A ship emission plume experiment was conducted about 100 km off the California coast during the NOAA Intercontinental Transport and Chemical Transformation (ITCT) 2K2 airborne field campaign. Measurements of chemical species were made from the NOAA WP-3D aircraft in eight consecutive transects of a ship plume around midday during 2.5 hours of flight. The measured species include NOx, HNO3, peroxyacetylnitrate (PAN), SO2, H2SO4, O3, CO, CO2, nonmethane hydrocarbons (NMHC), and particle number and size distributions. Observations demonstrate a NOx lifetime of ∼1.8 hours inside the ship plume compared to ∼6.5 hours (at noontime) in the moderately polluted background marine boundary layer of the experiment. This confirms the earlier hypothesis of highly enhanced in-plume NOx destruction. Consequently, one would expect the impact of ship emissions is much less severe than those predicted by global models that do not include rapid NOx destruction. Photochemical model calculations suggest that more than 80% of the NOx loss was due to the NO2 + OH reaction; the remainder was by PAN formation. The model underestimated in-plume NOx loss rate by about 30%. In addition, a comparison of measured to predicted H2SO4 in the plumes suggests that the photochemical model predicts OH variability reasonably well but may underestimate actual values. Predictions of in-plume O3 production agree well with the observations, suggesting that model-predicted peroxy radical (HO2 + RO2) levels are reasonable. The model estimated ozone production efficiency ranges from 6 to 30. The largest model bias was seen in the comparison with measured HNO3. The model overestimated in-plume HNO3 by about a factor of 6. This is most likely caused by underestimated HNO3 sinks possibly involving particle scavenging. However, limited data availability precluded a conclusive test of this possible loss process.
 Several theoretical studies have evaluated the chemical and climate impact of ship emissions with global models. Capaldo et al.  suggested that ship emissions are a dominant source of SO2 over much of the world's ocean and may produce a significant fraction of NSS (non-sea-salt) sulfate aerosol in these areas. These authors also estimated CCN (cloud condensation nuclei) emission rates and predicted a shift of −0.11 Wm−2 in global radiative forcing due to ship emitted particulate matter. Conversely, Endresen et al.  concluded that the overall effect of ship emissions on radiative forcing is small due to canceling effects. Lawrence and Crutzen  focused their effort on the effect of ship NOx emissions. Their chemical transport model (CTM) predictions showed significantly higher marine boundary layer NOx levels than those omitting ship emissions. In heavily trafficked shipping lanes the predicted enhancements were over 100 fold. Consequently, these NOx increases would lead to a quite noticeable effect on both surface ozone and OH levels. The estimated ozone enhancement in the central North Atlantic and Pacific was over a factor of 2 and model-predicted OH increases were up to a factor of 5 compared to expected levels if ship emissions were neglected.
Kasibhatla et al.  performed a similar modeling study with a more refined emission database and an improved geographical distribution of oceangoing ships but found that the predicted enhancement in NOx was not supported by observations. Comparison with data from the North Atlantic Regional Experiment (NARE) 1997 revealed that the model consistently over-estimated the observed NOx levels by up to factor of 10. Even with further refined emission data, the CTM study by Endresen et al.  still over-predicted the median boundary layer NARE NOx level by nearly a factor of 6. Interestingly, even when ship emissions were not considered, the model still over-predicted NOx by about a factor of 2 for both studies. Davis et al.  further compared the global modeling results of Kasibhatla et al.  with the data sets generated from 5 NASA Global Tropospheric Experiment (GTE) airborne campaigns from 1991 to 1999 in the North Pacific. This comparison revealed that the predicted NOx levels were 3.3 and 5 times higher than the observed values for spring and fall seasons, respectively. Again, the model was found to over-predict NOx even for the no-ship case.
 To reconcile these large differences between measurements and model predictions, several hypotheses have been put forward involving both overestimation of emission inventories and missing plume chemical and dynamical processes. One hypothesis is that CTM simulations do not properly treat the chemistry in ship plumes where NOx can be oxidized with a lifetime of a few hours. By contrast, the NOx lifetime is typically estimated to be around 1 day in the marine boundary layer by CTM models [Kasibhatla et al., 2000]. This means that a large fraction of the ship emitted NOx can be rapidly removed from the boundary layer before it is diluted to the grid size of a CTM. Power plant plume studies have demonstrated shorter in-plume NOx lifetimes in comparison to those estimated for background conditions [e.g., Ryerson et al., 1998]. By contrast, the lifetime of SO2 is primarily controlled by heterogeneous processes that are largely independent of plume chemistry. It is interesting to note that CTM-predicted SO2 levels are only about a factor of 2 lower than observed values [Davis et al., 2001]. Even this difference may be simply due to a bias in emission estimates. The latest global SO2 emission estimate shows a ∼11% increase, which would lead to a modest reduction in the magnitude of the model bias.
 Several studies have focused on the impact of detailed ship plume dynamics on the chemistry of the emissions using Lagrangian box models [von Glasow et al., 2003; Song et al., 2003a, Davis et al., 2001]. Song et al. [2003a] suggested that the in-plume NOx lifetime would be a factor of 2.5–10 shorter than in the ambient MBL primarily due to elevated plume OH levels. von Glasow et al.  show similar results and suggest that neglecting detailed plume chemistry in a global model overestimates the effect of ship emissions on ozone production by 50% and on OH levels by a factor of 2.
 The ship emissions of particles and SO2 were also examined in box model studies. In particular, von Glasow et al.  assessed the impact of aerosol particles on plume chemistry and concluded that the background (sulfate and sea salt) aerosols can have a significant influence on the gas phase chemistry of the ship plume while the ship emitted particles (including soot) had little additional effect. Song et al. [2003b] suggested that the major loss of SO2 is heterogeneous removal and the meteorological stability is the largest factor that controls the plume SO2 levels. The predicted SO2 and NSS are at the highest levels under stable conditions when the dilution rate is slower, in comparison with neutral or unstable conditions.
 Although modeling studies have shown the potential impact of ship emissions, observations are needed to test these predictions. In fact, several of the theoretical studies have stated the need for experimental data to better elucidate the dynamical and chemical processes of ship plumes [e.g., Corbett, 2003; von Glasow et al., 2003, Davis et al., 2001; Lawrence and Crutzen, 1999]. Observations of ship plumes can help us better determine the rate of the plume dispersion and also provide the observed NOx concentration as model constraints to examine various aspects of the plume chemistry.
 In this work we report the results of an experimental investigation of ship plumes made during NOAA ITCT (Intercontinental Transport and Chemical Transformation) 2K2 field campaign onboard the WP-3D aircraft on 8 May 2002 off the coast of California. During this flight, many gas phase and aerosol chemical species, aerosol size distributions, and meteorological parameters were measured in successive transects of a ship emission plume. Analysis of these data enables us to assess our understanding of ship plume chemistry, factors controlling NOx loss, and the chemical interaction between particulate matter (both background and ship emitted) and plume gases.
 A photochemical box model is used here primarily to provide estimates of short-lived compounds such as OH and H2SO4. The model contains explicit HOx-NOx-CH4 chemistry and parameterized NMHC chemistry; it has been used extensively in other airborne studies [Davis et al., 1996; Crawford et al., 1999; Chen et al., 2000; Olson et al., 2001]. The NMHC chemical mechanism is a modified version of the condensed mechanism of Lurmann et al.  to accommodate low NOx conditions (e.g., organic peroxide formation and feedbacks) and to represent explicit chemistry of acetone, propane, and benzene. A detailed listing of reactions can be found in the Appendix of Crawford et al. . The rate coefficients for gas phase reactions are those recommended by DeMore et al. . The photolysis rate coefficients, j values, were calculated for clear-sky conditions using a DISORT (discrete ordinates radiative transfer) four-stream implementation of the NCAR TUV (Tropospheric Ultraviolet Visible) radiative transfer model. The total ozone column data used in the TUV calculations was taken from satellite observations (http://toms.gsfc.nasa.gov/ozone/ozone.html). On the basis of visual observations and photographs taken during the flight, clear sky conditions were deemed appropriate for the radiative calculations. Rigorous ground based comparison has shown that TUV predicted j(NO2) is highly consistent with observations [Shetter et al., 2003].
 The model also includes surface and heterogeneous loss processes for soluble species (e.g., HNO3, H2SO4). These loss processes are parameterized as first-order reactions and the loss coefficients are derived from Logan et al. . However, when aerosol number/size distribution data are available, the rate of heterogeneous loss (or aerosol scavenging) is calculated using the formulation cited in the work of Fuchs and Sutugin . The mass accommodation coefficients and/or reaction probabilities were taken from recent literature (see later discussions).
 The model input consists of the 1 s observations of NO, CO, O3, H2O, SO2, temperature and pressure, and grab sample observations of NMHC. The high temporal resolution is required for plume analysis because of the small size of the plumes (<20 km) and the high speed (∼100 m s−1) of the aircraft. The model output includes OH, HO2, RO2 (organic peroxy radicals), H2SO4, HNO3 and rates of photochemical ozone formation and destruction. We opt to use the instantaneous photostationary state solution (dc/dt = 0) for the model output. This type of solution can provide adequate estimates for short-lived species in plume conditions. The same approach was also taken in earlier studies conducted in similar conditions [e.g., Frost et al., 1999, 2002]. Additional sensitivity model runs were carried out to examine the impact of longer lived species (i.e., CH2O) on model OH estimates. Limited Lagrangian calculations with parameterized dilution terms (see equation (4)) were also conducted to simulate the in-plume evolution of longer lived species (e.g., HNO3, PAN and O3). In these cases, the rates of photochemical production and loss were estimated using model calculations constrained by the observations recorded during the 8 plume transects. Interpolated values were used at the times between these transects.
3.1. Sampling Strategy and Background Conditions
 The ITCT 2K2 ship plume experiment was conducted around noon on 8 May 2002 about 100 km off the coast of California. Satellite images taken during the MAST study in 1994 demonstrate that this area has heavy ship traffic [Durkee et al., 2000a]. In situ observations of chemical tracers and wind patterns suggest that this region was moderately polluted, probably due to a combination of aged continental air with more recent inputs from the shipping lanes and possibly offshore platforms. This was supported by back trajectories that were essentially parallel to the Pacific coast for several days. The observed median background levels in the marine boundary layer outside of ship plumes for NOx, O3, and CO were ∼150 pptv, 40 ppbv, and 130 ppbv, respectively. Modestly elevated NMHC (nonmethane hydrocarbon) levels were also observed, e.g., C3H8 ranged from 300–400 pptv with a median of 370 pptv. The corresponding ratios of C2Cl4/CO and C2H2/CO were estimated to be 0.04 and 1.5 (pptv/ppbv), which are indicative of anthropogenic influences from urban/industrial sources [Smyth et al., 1996; Blake et al., 1999]. Particle number and size distribution measurements give an average total background volume concentration and surface area concentration of 7 μm3 cm−3 and 90 μm2 cm−3, respectively. In addition, average particulate sulfate and sulfur dioxide levels were 1.0 μg cm−3 and 0.5 ppbv, respectively. These levels are much higher than expected in a remote marine boundary layer and are probably due to sulfur rich emissions from ships.
 On the day of the ship experiment, visual observations indicated clear sky conditions with a few scattered clouds. The wind direction was consistently between 300°–325° and the average wind speed was 9–11 m s−1. The height of the marine boundary layer determined from aircraft altitude profiles varied from ∼350 m to ∼250 m through the sampling period. The observed atmospheric lapse rates suggest the meteorological stability was between neutral and unstable.
 Two ship plumes designated as east and west were encountered during the flight. The location of each plume is apparent in the measurements of NOy color-coded along the flight track of the WP-3D in Figure 1. The orientation of the ship plume is also consistent with the wind direction which is annotated in the figure. In the case of the east plume, there is no known information about the ship location or course. The west plume was a mixture of emissions from two ships close in location. The position, heading, and speed of the ships were visually determined. Ship 1 was heading WNW and Ship 2 was heading NNW; both at a speed of ∼5 m s−1 (∼9.7 knots). The aircraft sampling height for the plumes was ∼100 m above sea level. The initial sampling of the west plume was carried out by circling the ships. The closest sampling of the plumes occurred within 5 km downwind from the ships. The plumes from both ships were individually identified in circular sampling around the ships but merged into one further downwind. The data from the circular samplings were used to obtain emission factors for both ships. The coalesced plume was then systematically intercepted 8 times downwind, identified by the letters A-H as seen in Figure 1, which corresponds to plume ages of up to ∼2.5 hours (see discussion in section 3.3). It can be seen in Figure 1 that both plume tracks were essentially straight lines, indicating steady winds. The east plume was also periodically intercepted during the experiment but this data is analyzed at a lower level of detail as the background levels outside of the plume as well as the origin of the plume were less well defined. Discussions later in the text are focused on the west plume, unless specified otherwise. Two points should be noted about the plume sampling. First, two ships with different headings might be expected to form two separate plumes; however, visual observations showed the two ships were in close proximity, which facilitated the merge of their emissions into one plume. This merged plume was then consistently sampled in approximately a Lagrangian manner, i.e., the plane moved downwind at the speed close to that of the plume. Second, the heading of the ships almost directly into the wind, high wind speed, and the neutral to unstable stability conditions are factors all in favor of rapid dilution of the plume. This implies that this study may have been carried out at emission concentrations below typical values for ship plumes.
 One of the ships that formed the west plume was later identified as a 26,562 ton bulk carrier. Its diesel engine runs on marine fuel oil and generates ∼9120 to 10,730 horse power. The fuel consumption rate at sea is 27 tons day−1. The service speed is 14.5 knots (26.9 km hour−1).
3.2. Ship Emission Factors
 One major uncertainty in evaluating the impact of ship emissions is the estimate of the emission rates. Direct observations of the ship plumes can be used to estimate the emission factors (i.e., emission of certain species in gram per kilogram fuel consumed) [e.g., Hobbs et al., 2000; Sinha et al., 2003] which are critical parameters in emission inventory studies. For example, Hobbs et al.  derived the emission factors for NOx, SO2, and particles using the ratio of plume enhancement of effluents to CO2, e.g., ΔNOx/ΔCO2, ΔSO2/ΔCO2, etc. The emission factors were then calculated based on the mass ratio of CO2 to marine fuel oil of 3.2. The value of this ratio was derived from works by Tuttle  in which the hydrogen atom to carbon atom ratio was found to be 1.8 in marine fuel oil. Similarly, we have derived emission factors for NOx, SO2, and aerosol particles based on two plume observations closest to the ships (see Figure 2). The observed values of NOy, SO2, CO2, and total particle concentration, Ntotal, are shown in Figure 2. The ratios of ΔNOx/ΔCO2, ΔSO2/ΔCO2, and ΔNtotal/ΔCO2 were evaluated using the data shown in this figure. For the particle emissions the calculation was carried out over two size ranges 0.005–0.15 μm and 0.15–1.0 μm (in diameter). The estimated emission factors are summarized in Table 1 for both ships. The uncertainties quoted in the table reflect both instrument precision and the atmospheric variability in background levels. The latter is especially important in determining the magnitude of the CO2 enhancement because of the small difference between the plume and background. The estimated NOx emission factors are consistent with other in situ estimates [e.g., Hobbs et al., 2000; Sinha et al., 2003] and the values commonly used in emission inventory studies [e.g., Corbett et al., 1999]. The total particle emission factors are also consistent with those estimated by Sinha et al. ; assuming that Ntotal provides a meaningful comparison point as the measurement size bins are quite different.
Table 1. Summary of Estimated Emission Factors for Residual Fuel Ships
Typical ranges estimated from reported average values for various types of vessels [European Commission and ENTEC UK Limited, 2002, Tables 2.11–2.13]. Ranges for sulfur are based on heavy fuel oil sulfur contents averaging 2.7% by weight; individual vessels may periodically use fuel with lower sulfur contents than assumed in the ENTEC factors.
Nnucl = particles in size range of 0.005–0.15 μm in diameter.
Npm1 = particles in size range of 0.15–1 μm in diameter.
Ntot: = particles in size range of 0.005–7 μm in diameter.
 The SO2 emission factors are significantly lower than those estimated for ships burning marine fuel oil [e.g., Corbett et al., 1999; Corbett and Koehler, 2003; Hobbs et al., 2000; Sinha et al., 2003]. This may indicate that the sulfur content in the fuel used in the ships observed during ITCT was much lower. Average fuel sulfur contents are about 2.7% for international marine fuels [Corbett and Koehler, 2003; Endresen et al., 2003; European Commission and ENTEC UK Limited, 2002; International Maritime Organization and Marine Environment Protection Committee, 2001]. Of course, fuel-sulfur levels vary according to when and where fuel is purchased. A recent inventory [Starcrest Consulting Group LLC et al., 2004] of commercial marine vessels in Los Angeles confirmed that average sulfur levels used aboard ships on the west coast of North America is near the world average (averaging 2.64%); interestingly, this inventory also observed some vessels with main engine fuel-sulfur levels at or below 2%. The fuel delivery report obtained for this study indicates that the fuel sulfur content for the identified ship was most likely in the range of 1.9 to 2.2%. It is reasonable to expect that that the other vessel was consuming similar type of fuel. On the basis of the fuel sulfur content values, the SO2 emission factors are estimated to be 38–44 gSO2 kg fuel−1, which is still significantly higher than the values listed in Table 1. This discrepancy suggests that the SO2 emission factors derived from observations are inconsistent with the estimated fuel sulfur content.
 We have also investigated whether expected CO2 emissions are consistent with the observed plume enhancements, based on main engine fuel consumption of the identified bulk carrier at 27 ton/day. At the observed ship speed of ∼5 m/s (9.72 knots) and the rated vessel speed of 14.5 knots, engine load was 67% of full speed (∼30% of full power) which translates to a fuel consumption rate of ∼0.3 kg C (carbon)/s. Using the mass ratio of CO2 emitted to fuel consumed, the CO2 emission rate would be 1.09 kg/s. A Gaussian plume dispersion model was then used to estimate the CO2 mixing ratio at 5 km downwind where the ship plume was first spotted. The Gaussian plume dispersion scheme used in the calculation is that described by Song et al. [2003a], which is based on the offshore dispersion model developed by Hanna et al. . The model results show a high sensitivity to the meteorological stability. As mentioned earlier, the observed lapse rates suggest conditions were between neutral and unstable. The model-predicted centerline CO2 plume enhancements for these two conditions were 0.22 and 2.0 ppmv, respectively. This brackets the observed CO2 enhancement of ∼1 ppmv. Considering the large uncertainties in the dispersion model, it is clear that better models need to be incorporated in ship plume studies to represent the plume dispersion processes under actual ambient conditions.
3.3. Evolution of the Ship Plume
Figures 3a–3c shows three transects of the WP-3D of the west plume labeled west A, C, and H. These plots are time series of 1 second observations of NOx, SO2, and Ntotal (total particle number concentration). The width of the sampled plume ranges from five to ten km (∼50 s to ∼100 s of flight time) and requires the use of high resolution data to accurately characterize the plume even though the imprecision is large in some of the 1 second data, particularly for SO2. In Figures 3a and 3b, significant enhancements can be seen for all these species above background for west A and C. In the case of west H, the plume SO2 level is indistinguishable from the background; however, the enhancements of NOx and Ntotal are still evident, especially for Ntotal. The large variations observed during the west A transect suggest that the two plumes were partially mixed. The smaller variations observed in the later plume transects indicate that the plumes have become more thoroughly mixed.
 The decay of ship effluents is illustrated in Figure 4 which plots the average plume mixing ratio versus the age of the plume for all 8 transects for NOx, SO2, and Ntotal. Also shown are the corresponding background levels estimated from the data collected 30 s before and after the plume. The plume age is defined as the time elapsed between emission and sampling. Since ship information (i.e., speed, heading, and location) is limited to the initial visual observations, it is difficult to determine the absolute age of the plumes sampled in transect A through H. However, the relative plume age between transects can be readily calculated using the observed wind speed and locations of the plumes. The aircraft position is determined from GPS coordinates. Therefore the estimated relative plume ages are reasonably precise, although the absolute age may have a significant offset. The plume age for transect A shown on the plots was estimated as the time required for the plume to travel from the location where ships were initially spotted by NOAA WP-3D.
 A data filter is used to compute the averages displayed in Figure 4. This filter is based on total particle concentration, Ntotal. Ntotal is expected to experience changes due only to dilution. Particle loss through dry deposition and coagulation are estimated to be negligible. In addition, the observation of ultrafine particle (5–10 nm) suggests the maximum nucleation effect on Ntotal to be less than 3% for west A and this effect will rapidly decrease for the later transects. For a given transect (e.g., west A), the filtered averages contain only the data points for which the concurrent in-plume Ntotal enhancement (plume value minus background) is higher than 50% of the peak value. The peak value is defined as the 90th percentile of a given transect. The purpose of this filter is to better represent the center plume levels of the variables of interest.
 As shown in Figure 4, the average plume concentrations of the ship effluents decrease as the plume ages. The magnitude of the decrease in the plume enhancement is different for each of these three species. The observed decreases can be explained by plume dispersion and other losses including chemical conversion, particle scavenging, and surface deposition. Since the losses (i.e., deposition and coagulation) and production (i.e., nucleation) of Ntotal are negligible, the rate of dispersion can be derived by assessing the rate of decay of ΔNtotal (plume average minus background) with respect to time. Assuming a constant background, a dispersion lifetime of ∼2.5 ± 0.6 hours is determined by a least squares regression analysis weighted by the relative uncertainties estimated for ΔNtotal. The R2 value is estimated at 0.78.
 For the case of NOx, there are significant additional losses due to photochemical processes. The first-order loss rate coefficient kNOx for all photochemical losses can be estimated using the concentration ratio approach as described by Ryerson et al. , i.e.,
where ΔNOx is the NOx enhancement due to ship emission above background, ΔNtotal is the plume enhancement in Ntotal; and tp is plume age. By contrast, this method is not suitable to estimate SO2 lifetime. As shown in Figure 4b, 6 out of 8 points are not clearly distinguishable from the background. This is due mostly to the large variation in background measurements reflecting the instrument limitation in determining low concentration levels.
 The decrease of plume NOx is due to both plume dispersion and chemical losses. The concentration ratio approach analysis, i.e., equation (1), is applied on data shown on Figure 4 and an effective west plume NOx lifetime of 109 ± 21 min is derived from a weighted regression analysis. The R2 value is estimated at 0.86. An identical analysis was also performed on the east plume of unknown origin (ships were not visually observable.). This yielded an observed NOx lifetime of 113 ± 23 min which indicates the chemistry is similar between the east and west plumes. The NOx lifetime derived from ITCT observations is close to the minimum of the instantaneous NOx lifetimes reported by Song et al. [2003a]. For this lifetime, 80% of the ΔNOx (plume average minus background) observed in transect west A would be removed in about 2.5 hours, the time when the west plume reaches west H. The NOx loss was also evaluated using the mass balance approach of Ryerson et al. . A net NOx flux (ship NOx only) is estimated by integrating over the ship plume transected by WP-3D. These results indicated that the net flux estimated for west H is only about 17% of that of west A. This suggests that about 83% of ship emitted NOx was removed from the plume which is in excellent agreement with the concentration ratio approach.
 Previous studies have suggested that the shortened plume NOx lifetime is due to elevated OH levels which result from the high levels of plume NOx. The ITCT OH measurement did not have sufficient time response for measurements in the plume transects so we use H2SO4 as a surrogate species to examine the plume HOx chemistry. This reflects that the source of H2SO4 is the OH initiated oxidation of SO2 and it has a short lifetime (∼5 min for the observed surface area of 120 μm2 cm−3) due to aerosol scavenging. As shown in Figure 5a, plume average H2SO4 maximizes at west C, corresponding to a somewhat reduced NOx level of 1.2 ppbv, while SO2 levels were nearly constant as shown in Figure 4b. Assuming the observed H2SO4 trend was modulated by OH, the peak at west C is qualitatively consistent with previous studies that the highest OH does not correspond to the highest NOx [e.g., Song et al., 2003a] (see further discussion in section 4.1).
Figure 5b shows the plume average and standard deviation of HNO3 mixing ratio (filtered by Ntotal) as well as the average background level and its standard deviation. The HNO3 background levels were generally lower than the 25 pptv precision of the 1 s nitric acid measurement. The average background HNO3 level was estimated at ∼5 pptv ± 5 pptv by averaging 30 s of data before and after each plume. The plume average HNO3 is significantly higher than the background levels, but is much lower than plume NOx levels, i.e., 40–100 pptv for HNO3 versus 280–1600 pptv for NOx. Either HNO3 is not efficiently formed by NOx oxidation in the ship plume, or there is a rapid loss process for HNO3.
 Another way to assess the NOx to HNO3 conversion is to examine the relationship between measured NOx and NOy. Figure 6 shows the relation between plume averages of NOx and NOy, i.e., ΔNOx and ΔNOy. The plume averages were normalized by ΔNtotal to minimize the difference in the impact of plume dispersion on these nitrogen species. As shown in figure, ΔNOx/ΔNtotal and ΔNOy/ΔNtotal are strongly correlated with a R2 value of 0.99 and a slope of 0.86. A nearly identical relationship was also seen in the data collected in the east plume. This indicates that NOx is the major component of NOy and there was no significant accumulation of HNO3 in the plume. Previous studies of intense NOx sources (e.g., power plant) have shown significant buildup of gas phase HNO3 and a reduction of the NOx to NOy slope to values much less than one [Ryerson et al., 2001; Neuman et al., 2002]. Thus the low in-plume HNO3 levels, relative to NOx, suggest that either OH oxidation is not the major NOx sink or that there is a strong HNO3 sink. Model simulations that assess the sources and sinks of HNO3 as well as the in-plume sink of NOx are discussed in section 4.2.
4. Model Results
 In this section, we present the results from the modeling analysis. The gas phase plume chemistry is evaluated in terms of OH levels, NOx lifetime and oxidation products. Finally, the interactions between aerosol particles and plume chemistry are assessed.
4.1. OH Levels
 The hypothesis of rapid in-plume NOx loss is based on the prediction of highly elevated OH levels. Figure 7 shows the average plume OH levels from model calculations constrained by observations. These averages are also filtered by Ntotal as previously described. Also shown in the figure are the corresponding observed plume NOx values and the model calculated background OH level, i.e., OH levels calculated outside of the ship plume. It can be seen that plume OH levels are about a factor of 1.2 to 2.7 higher than background levels. The highest plume OH average of ∼1.6 × 107 molecules cm−3 is seen at west D, corresponding to a NOx level of ∼0.9 ppbv. It is also clear that plume OH itself varies by nearly a factor of 1.6. This variation is primarily driven by changes in NOx levels that impact the HO2 to OH ratio. While the total HOx (OH + HO2) level decreases with increasing NOx level due to the losses from the NO2 + OH reaction, increasing NOx also shifts the partitioning in favor of OH via the HO2 + NO reaction. The model estimated HO2/OH ratio ranges from 13 for west A, where NOx is highest, to 57 for west H, where NOx is lowest. Background ratios are around 90. The overall average plume HOx level is about a factor of 1.5 lower than the background value, with this ratio ranging from ∼3 at the high NOx level at west A to near unity for the lower NOx at west H. The OH + NO2 reaction is the predominant HOx loss within plume, contributing nearly 40% of the total in contrast to only ∼1% contribution to the total loss under background conditions. It should also be noted that since the ITCT ship experiment was conducted around midday local time the primary HOx production from O(1D) + H2O reaction was fairly constant with a variation of less than 15%. Finally, the particle scavenging of HO2 is estimated to have a 5% or less effect on model OH levels, which is based on estimates using in situ observed particle number/size distributions and an assumption of HO2 reaction probability of 0.2.
 Although observation-constrained model predictions suggest highly elevated OH levels in the ship plume, these predictions must be verified by observations to test the validity of current photochemical theory for ship plumes. For this reason the H2SO4 data were used to test model OH predictions. The production of gas phase H2SO4 is effectively determined by the SO2 + OH reaction. The major H2SO4 sink is particle scavenging which can be evaluated using the formulation developed by Fuchs and Sutugin  and in situ observed number and size distribution data. The value for the sulfuric acid mass accommodation coefficient used in this calculation is 0.7, based on a recent recommendation by Sander et al. . To reduce the uncertainty from large fluctuations due to instrumental noise and ambient variations, both SO2 and the particle number and size distribution data are averaged for each ship plume transect with the Ntotal filter. The resulting first-order H2SO4 scavenging coefficient corresponds to a lifetime of 4.5 to 6 min. Over 95% of the scavenging is attributed to submicron particles as they account for 90% of the total surface area, which averages around 120 μm2 cm−3. Compared to the background, the in-plume scavenging rate coefficient is about 45% higher. On the basis of the lifetime estimates, we assume steady state for H2SO4. The calculated and observed H2SO4 are shown in Figure 8. Comparison between observations and model values reveals a high degree of covariation with a R2 value of 0.9. There is a significant offset of 1 × 107 molecules cm−3 and a slope of 2.0. This suggests that model-predicted OH may be low by up to a factor of two. Uncertainty in modeled OH, measured SO2 and H2SO4, and the estimated scavenging rate could all play a role in this discrepancy.
 The marine boundary layer of the study has moderately elevated nonmethane hydrocarbon (NMHC) levels. Consequently, the intermediate products of NMHC oxidation (e.g., CH2O) may be significantly higher than the steady state values calculated from the model [Fried et al., 2003; Olson et al., 2004]. To explore this potential impact on OH, we have carried out sensitivity calculations based on the 95th percentile of CH2O observation (i.e., ∼900 pptv) made in the MBL during a recent NASA field program TRACE-P. This amount of CH2O, however, has a less than 5% effect on plume OH levels because HOx produced from CH2O photolysis is far smaller than that from O(1D) + H2O reaction. Another possible OH source involves heterogeneous reaction between NO2 and soot to produce nitrous acid (HONO). However, based on a reaction probability of 5 × 10−4 measured by Longfellow et al. , we estimate this source is negligible for the conditions encountered in this study.
4.2. NOx Lifetime and Oxidation Products
 In section 3.2, NOx lifetimes of 109 ± 21 and 113 ± 23 min were directly estimated from observations for the west and east plumes, respectively. Here we compare the model derived quantities with these values to gain an understanding of the major chemical process responsible for plume NOx destruction.
4.2.1. NOx Instantaneous Chemical Lifetime
 The instantaneous NOx lifetime for a given transect is defined as the ratio of the total burden over the total loss rate. The quantity can be estimated by calculating the ratio of average NOx loss rate across the transect to the average NOx concentration. Figure 9 shows the model estimated NOx instantaneous lifetime for the west plume transects and corresponding average OH levels as a function of plume age for all 8 plume transects. The NOx lifetime ranges from 2.5 hours to just over 3 hours, which are substantially shorter than the background NOx lifetime of ∼6.5 hour for the time period of the ITCT ship experiment. Similarly, the model NOx lifetime estimated for the east plume varies from 2.1 to 2.8 hours. The variation of the lifetime is anticorrelated with OH levels. This strong anticorrelation reflects that OH + NO2 is the main NOx loss process, contributing ∼85 to 90% for west A through D and ∼65 to 75% for the rest of later transects. The other significant ship plume NOx sink involves formation of PAN. Note that heterogeneous losses of NOx have been found to be negligibly small. For the ship plume experiment, the PAN lifetime is dominated by thermal decomposition and ranges from 7.5 to 8 hours which is substantially longer than the duration of the experiment itself. Thus production of PAN would be observed as a NOx loss during the experiment. The PAN reaction sequence is shown below:
The net rate of NOx loss from these processes, L(NOx)PAN, can be estimated by the difference between PAN formation rate and the NOx feedback from PAN, i.e.,
where k1, k2, k3 and j4 are rate coefficients for reactions 1 to 4, respectively.
 The rate of PAN formation can be predicted by a model based on the observed values of precursors; while the rate of feedback to NOx is estimated using observed PAN values together with model calculated OH and j4 as well as reaction rate coefficients from DeMore et al. . It should be noted that there are very limited PAN observations available in and around the west plume during the experiment. The available observations suggest that there is no statistically significant difference in PAN between earlier transects and later transects and the difference between inside and outside of the plume is also indistinguishable. Thus we have used the average of 135 pptv for [PAN]obs in equation (2). We predict a 60 pptv PAN increase from west A to west H assuming PAN concentration is at background level at west A. This increase is about 3 times higher than the reported measurement precision [Roberts et al., 2004] and should be detectable. The absence of any observed change in PAN may be attributed to several factors, including the limited number of in situ observations (i.e., 5 measurements near the fringe of the plume and 1 close to the center of the plume), kinetic coefficients, and the uncertainty in model prediction of CH3C(O)O2, which may be considerably larger than that of OH due to the uncertainties in NMHC oxidation mechanism.
4.2.2. Equivalent NOx Chemical Lifetime
 The instantaneous lifetime represents the rate of NOx decay at a particular point in time. The concept of equivalent lifetime is used here to address the average loss rate of NOx over the 2.5 hour of processing observed for the ITCT ship plume. The instantaneous NOx chemical lifetime is a function of plume age as shown in Figure 9. The working definition of equivalent NOx chemical lifetime is the time required to reproduce the NOx decrease (defined by the ratio of starting and ending concentrations) for a given time period under the assumption of exponential decay. On the basis of this definition, the equivalent lifetime can be derived from instantaneous lifetime using the equation below:
where τeqv and τinst are the equivalent NOx and instantaneous NOx chemical lifetimes, respectively; and Δt is the time period of interest. From the lifetime values shown in Figure 9, a model equivalent lifetime is calculated to be 143 ± 40 min. For the east plume, the estimated equivalent lifetime is 135 ± 37 min. The uncertainty cited can be attributed to those in model calculations, e.g., OH and CH3C(O)O2. If only OH oxidation were considered, the τeqv value would be 172 ± 45 min. Comparing to the observed τeqv of 109 ± 21 min, the model τeqv is significantly longer even though the uncertainties overlap each other. This may suggest that the model OH could be underestimated by about 30%, which is qualitatively consistent with the suggestion of low model-predicted OH determination from the H2SO4 analysis. More importantly, both observed and model τeqv values are much smaller than the calculated background NOx lifetime of ∼6.5 hours for the time period of the ITCT ship experiment (Note: the diurnal average NOx lifetime is estimated to be 21 hours for the background conditions). Finally, we would like to point out another source of this difference between model and observations. The model τeqv is defined to reproduce the difference between the NOx levels at the start point and end point of the time period of interest. The observed τeqv, however, represents the average rate of the NOx decay, as it is derived from regression analysis. If only the data from west A and west H (start and end points) were used, the resulting τeqv would be 127 ± 21 min, which is closer to the model value.
 In the past two sections, we have shown that ship plume NOx lifetime is considerably shorter than for the unperturbed marine boundary layer. This short lifetime is believed to be primarily due to a large enhancement in OH levels. Thus one would expect that the plume has much higher HNO3 concentrations than background. The photochemical model predicts that around 300–400 pptv of HNO3 should have been observed from west C to west H transects. This prediction is under the assumption that the major sinks of plume HNO3 are plume dispersion and dry deposition with a velocity of ∼0.6 cm sec−1 for the wind speed observed during the ITCT ship experiment [Pryor and Sorensen, 2000]. The boundary layer depth used here is 300 m.
 The observed HNO3 mixing ratios range from 40 to 100 pptv with an average of 69 pptv (Figure 5b). The low HNO3 values are consistent with the observed relation between NOx and NOy (see section 3.2). However, model to measurement comparison of H2SO4 suggests model-predicted OH levels may be as much as a factor of 2 too low. The model also under estimates the observed τeqv by at least 30%. Consequently, the model does not likely overestimate the rate of NOx oxidation by OH which leads to HNO3 formation. Thus additional or stronger sinks are needed to reconcile the difference between model and observed HNO3. An average HNO3 lifetime of ∼20 min would reconcile the model predictions with average observations calculated from the west A to west H transects. This lifetime is nearly a factor of six shorter than the current lifetime based on plume dispersion and dry deposition. If there were no additional sinks, the dry deposition velocity would have to be ∼25 cm sec−1, more than one order of magnitude higher than literature reported values. This rapid HNO3 loss is not unique to this experiment. Similar rates of HNO3 loss were also observed in power plant and refinery plumes during TexAQS 2000, which could not be explained by known sinks [Neuman et al., 2003]. Another potential sink is particle scavenging of HNO3. On the basis of the observed aerosol number size distribution, it would require a reaction probability, γ, of 0.12. This value is at, if not above, the upper limit of current literature reported value. The average observed total particle surface area for the plume transects was ∼120 μm2 cm−3 and the average for the background was 90 μm2 cm−3. This difference can mostly be attributed to the particle size range under 150 nm. The scavenging due to this size range is estimated to be ∼45% of the total for the plume transects and ∼29% for the background conditions. Considering the high wind speed of 9–11 m s−1, it is not unreasonable to assume that background particles larger than 150 nm are mostly sea salt. This assumption is, at least partially, based on reported observations of significant portion of submicron sea-salt particles down to as low as ∼200 nm [Murphy et al., 1998; Wang et al., 2002]. In this context, one would conclude that sea-salt scavenging was the major sink of HNO3 while the contribution from ship emitted particles remains significant.
 It should be noted that thermodynamic models cannot be used to predict the partitioning between gas phase HNO3 and particulate NO3− because there are no data on either NH3 or NH4+ available for the experiment. The few available PILS submicron plume NO3− measurements show no significant difference between the inside and outside of plume; while we predict more than a factor of ∼5 increase in particulate NO3− would be observed in transect H if HNO3 were lost to particles. However, it should also be recognized that the PILS would not be able to observe a significant nitrate increase if most submicron nitrate was in organic forms. The total particle volume increase is estimated at 1 ± 0.4 μm3 cm−3 from uptake of gas phase HNO3 and H2SO4. This value is estimated using a γ value of 0.1 for HNO3 and a sticking coefficient of 0.7 for H2SO4 together with observed gas phase concentrations. Comparing to average observed plume particle volume of ∼8 μm3 cm−3, this growth is relatively minor and not likely to be observed given the instrument precision and natural fluctuations. Another possibility for the rapid loss of HNO3 is efficient uptake on larger sea salt aerosol which is expected to have diameters typically greater than 1 μm [Seinfeld and Pandis, 1998]. However, there is not enough observed surface area in this size range to account for the HNO3 loss even at a diffusion limited rate. While we cannot draw more definitive conclusions about the contribution of particle scavenging to a HNO3 loss because of the limited availability of data, this analysis serves to further constrain the magnitude of additional HNO3 losses necessary to explain observations within pollution plumes.
4.3. Ozone Production
 We have estimated ship plume ozone evolution based on the following equation:
where [O3]plm and [O3]bkg are ozone concentrations of plume and background, respectively; P(O3) is photochemical ozone production; Lphoto is the photochemical ozone loss frequency; and Ldisp is the loss frequency due to plume dispersion. The values of P(O3) and Lphoto are calculated by the model and the value of Ldisp is estimated from the decay of ΔNtotal (see section 3.1). Figure 10 displays the comparison of the model simulated and observed ozone as well as the background ozone value. The model calculation is initialized by the average of ozone observed in west A. As shown in the figure, the model captures the observed trend and the model bias is less than 5%. This good agreement can be viewed as an indirect confirmation of model-predicted peroxy radicals, mostly HO2. Although the west A ozone concentration is lower than the background level, model calculations indicate photochemistry is a net source of ozone at a rate of 2.3 ppbv hour−1. The initial ozone concentration is lower than background due to titration by emitted NO as evidenced by the highly elevated NOx prior to west A. Even higher net ozone production is estimated for west B, C, and D ∼2.9 ppbv hour−1. For west H, this rate decreases to 0.9 ppbv hour−1.
 The ozone production efficiency, defined as the ratio of ozone produced to NOx consumed, ranges from 6 for west A to 30 for west H with an average value of 10 and is anticorrelated with the NOx level. This average value agrees well with that derived from a regression analysis between ozone and the amount of NOx removed. The total ozone production for the sampling period is 4.0 ppbv and a net increase of 2.5 ppbv over the background. The largest production channel is HO2 + NO which contributes ∼76%; while the major ozone sinks involves two sequences of reactions: (1) ozone photolysis followed by reaction of O(1D) + H2O and (2) O3 + NO followed by NO2 + OH. The former dominates in lower NOx transects, i.e., west E through west H, contributing over 50%; while the second sequence accounts for 40–50% of the total ozone loss for west A through west C.
 The observations demonstrate rapid plume NOx loss, i.e., ∼80% of ship emitted NOx was removed within the duration of the experiment ∼2.5 hours. The equivalent ship plume lifetime derived from the observed NOx decay is ∼1.8 hours. By contrast, the model estimated NOx lifetime in the background marine boundary layer is 6.5 hours for midday conditions. The model-predicted plume NOx lifetime is 2.4 hours, about 30% higher than that derived from observations. The model analysis primarily attributes this rapid NOx loss to oxidation by elevated plume OH levels. Formation of PAN is estimated to also have significant contribution as a temporary sink of plume NOx. However, PAN is a reservoir of NOx that may further extend the influence of ship plumes to the remote marine atmosphere. It should be noted that the estimate of net plume NOx loss through PAN is based on limited in-plume observations, which may not be representative of the whole plume and there is little experimental evidence for enhanced PAN production in the plumes. Fast PAN measurements would be desirable in future experiments to further elucidate and quantify the role of PAN.
 To test our understanding of ship plume chemistry, model/measurement comparisons of H2SO4 and O3 were conducted as indirect tests of OH and HO2, respectively. Comparison of H2SO4 showed that the model prediction is highly correlated with the observations with a R2 value of 0.9, but model values are significantly lower than the observations. An increase of model OH by a factor of 2 would remove the discrepancy. The model underestimated the observed rate of NOx loss by about 30%, which also suggests model OH estimates may be lower than the actual ambient values; although this difference is well within the estimated uncertainties for model and observations. However, if the model predictions of PAN formation are in error the discrepancy is even larger. The model simulation of plume ozone is within 10% of the observations. This suggests that the model-predicted HO2 is quite reasonable. Of course, a more conclusive test would involve in-plume measurements of both OH and HO2.
 The largest model disagreement was with the observed levels of HNO3 in the ship plumes. The observed plume level varies from 40 to 100 pptv, with an average of 69 pptv. Model calculations predict plume HNO3 levels ranging from ∼300 to 400 pptv. We believe that the model over-estimation is caused by a missing or under-estimated sink for HNO3. To reproduce the average observation, an average HNO3 lifetime of ∼20 min is required. One potential process that can account for this rapid loss is particle scavenging with assumptions of irreversible loss and reaction probability of >0.1. However, the time resolution of particulate nitrate measurements could not provide a clear indication of in-plume nitrate increase. The lack of NH4+ measurements prevented a test to check if the thermodynamic equilibrium would allow uptake of HNO3. On the basis of estimates of gas phase uptake of HNO3 and H2SO4, the total particle volume would have increased by 1 μm3 cm−3, or 12%. This increase is significant, but smaller than the combined uncertainty due to measurement precision and atmospheric variability. We believe this is a critical issue that should be addressed in the future. Fast and accurate observations of particle chemical composition (including organic nitrate) would be required to address this problem.
 Ship emission factors are derived from observations for NOx, SO2, and total particle concentration. While good agreement with previously reported values are obtained for NOx and total particle concentration, the values for SO2 emission factor appears to be a factor of 1.6 to 1.9 lower than literature values which are generally consistent (i.e., within 25%) with the values derived from the fuel sulfur content aboard the identified ship. This suggests that the sulfur chemistry in this experiment occurs at significantly lower levels than expected for more typical ship emissions.
 The authors would like to first thank the personnel of the NOAA Aircraft Operations Center for making this experiment a successful one. Special thanks go to Josh Graml of Mariners' Museum Research Library and Archives in Newport News, Virginia, and Vic Delnore of NASA Langley Research Center for their help in locating the information on the ship encountered during the experiment. Chris Patton of the Port of Los Angeles is acknowledged for obtaining fuel quality data for the voyage of ship one from the operator. This work was supported in part by NOAA OGP-NA06GP0410. The postmission analysis performed at NASA Langley Research Center was supported by the Tropospheric Chemistry Program of NASA's Earth Science Enterprise. The ITCT 2K2/PEACE campaigns were conducted under the framework of the International Global Atmospheric Chemistry project (http://www.igac.noaa.gov/).