Trace Metals in Soot and PM2.5 from Heavy-Fuel-Oil Combustion in a Marine Engine.

Heavy fuel oil (HFO) particulate matter (PM) emitted by marine engines is known to contain toxic heavy metals, including vanadium (V) and nickel (Ni). The toxicity of such metals will depend on the their chemical state, size distribution, and mixing state. Using online soot-particle aerosol mass spectrometry (SP-AMS), we quantified the mass of five metals (V, Ni, Fe, Na, and Ba) in HFO-PM soot particles produced by a marine diesel research engine. The in-soot metal concentrations were compared to in-PM2.5 measurements by inductively coupled plasma-optical emission spectroscopy (ICP-OES). We found that <3% of total PM2.5 metals was associated with soot particles, which may still be sufficient to influence in-cylinder soot burnout rates. Since these metals were most likely present as oxides, whereas studies on lower-temperature boilers report a predominance of sulfates, this result implies that the toxicity of HFO PM depends on its combustion conditions. Finally, we observed a 4-to-25-fold enhancement in the ratio V:Ni in soot particles versus PM2.5, indicating an enrichment of V in soot due to its lower nucleation/condensation temperature. As this enrichment mechanism is not dependent on soot formation, V is expected to be generally enriched within smaller HFO-PM particles from marine engines, enhancing its toxicity.

Introduction

Marine engines represent a major source of atmospheric particulate matter (PM), with global emissions comparable to road traffic and aviation.[1] As close to 70% of shipping emissions occur within 400 km of coastlines,[2] and marine-engine PM may contribute significantly to local mortality.[3] The remaining 30% of emissions are the major source of anthropogenic air pollution in the marine environment.

Ocean-going marine engines typically operate on the cheap residual fuel known as Heavy Fuel Oil (HFO). Due to its high sulfur content, HFO has already been prohibited within the so-called Sulfur Emission Control Areas (SECAs, coastal zones near Europe and North America). However, as ∼80% of global shipping in 2015 originated from ports outside of SECAs,[4] large urban populations remain exposed to significant levels of HFO exhaust.[5−7]

Epidemiological studies have suggested a connection between <span class="Chemical">HFO-combustion PM and mortality[8] as well as morbidity.[9] Multiple toxic species are found in HFO; HFO PM is rich in transition metals and polycyclic aromatic hydrocarbons (PAHs), both of which play a role in its toxicity.[10−14] A significant mass fraction of these transition metals may be present in the smaller PM fractions,[15−19] such as PM2.5 (PM with diameters <2.5 μm), which are small enough to penetrate to and affect the human bronchi and alveoli.[20] When insoluble, such particles may remain in the lungs for days,[21−24] catalyzing oxidative stress.[25] The even smaller PM0.1 size fraction is more likely to pass through exhaust aftertreatment systems[26] and are small enough to enter the bloodstream after inhalation.[24] The passage of PM0.1 through aftertreatment systems is particularly noteworthy, as such aftertreatment systems are required in SECAs when HFO is combusted.

The most abundant trace metal in <span class="Chemical">HFO is vanadium(V), which is also particularly toxic.[27] Nickel (Ni) and other metals are present in smaller amounts. Such metals are typically present as oxides, though ∼10% exist as organometallic compounds.[28−30] Some fraction of these metallic compounds may vaporize during fuel combustion before nucleating into new particles.[18,31,32] These nucleated particles may then act as in-flame condensation nuclei for soot formation,[16,33] leading to metal or metal-oxide inclusions in soot with diameters of ∼5–10 nm.[34] These inclusions have the potential to influence soot emissions by catalyzing soot oxidation.[31,35,36] The inclusions may be composed of metal alloys or oxides, such as Ni, Ni2Fe, Fe2S, V2O3, and Ni3Fe2O3.[33] Popovicheva et al.[33] identified these compounds in a ship-exhaust PM sample using scanning electron microscopy (SEM), which provided single-particle elemental as well as morphological information. However, only a limited number of particles can be analyzed quantitatively by SEM due to its labor intensity.

In contrast to SEM, online aerosol mass spectrometry provides the opportunity to quantify metals in aerosol particles with high throughput. The recently developed Soot-Particle Aerosol Mass <span class="Chemical">Spectrometer (SP-AMS),[37] which uses a 1064 nm-laser vaporizer to specifically vaporize particles which absorb light efficiently at that wavelength, has the ability to selectively quantify metals in soot particles since few materials other than soot do so.[37−40] Here, we operated the SP-AMS laser vaporizer in tandem with the standard AMS thermal vaporizer[41] to obtain additional information about particle composition.

In this manuscript, we refer to the particles vaporized by the <span class="Chemical">SP-AMS laser (and not the AMS thermal vaporizer) as soot particles. Since these particles have aerodynamic diameters in the range of 60 to 600 nm (discussed further below), soot particles are a subcategory of PM2.5. A soot particle is a flame-synthesized particle primarily comprised of an aggregate of refractory black carbon (rBC) spherules, where rBC is the light-absorbing, insoluble, graphene-like PM that vaporizes at ∼4000 K.[42,43] It follows that soot may contain additional condensed material, such as sulfates and organics, as well as metal inclusions. We note that in the primary engine emissions studied here, the mass fraction of organics and sulfate internally mixed with soot was negligible,[44] so that the potentially ambiguous scenario of a particle containing only 10% rBC is irrelevant here. In HFO PM, a second type of rBC particle may be observed: char. Char particles are the supermicron, graphitized, often porous carbonaceous residues of fuel droplets.[16,32,33,45] Due to their different formation mechanisms, the trace-element composition of soot particles will be different than that of char particles. Crucially, the supermicron diameter range of char particles means that they are too large to be measured by the SP-AMS, so that SP-AMS measurements are intrinsically selective for soot particles, whereas filter samples may include both soot and char.

In this work, we present SP-AMS measurements of the intern<span class="Chemical">ally mixed metal content of soot generated by a marine four-stroke diesel research engine operated on HFO. We use the term “metals” to describe the metal mass fraction of alloys, oxides, or other compounds. (Our observations indicate that metal alloys and/or oxides were the dominant species present in our sample.) The specific metals measured were vanadium, nickel, iron, sodium, and barium (V, Ni, Fe, Na, and Ba). In addition to the SP-AMS data, additional measurements of rBC mass concentration by laser-induced incandescence (single-particle soot photometer, SP2) and bulk PM2.5 metal content by inductively coupled plasma-optical emission spectroscopy (ICP-OES) are presented. SEM measurements were performed to exclude the presence of char particles in our samples. Finally, measurements on two metal-free distillate fuels (marine gas oil, MGO, and diesel fuel) are used as controls.

Materials and Methods

Experimental Section

The engine used in this study was a four-stroke, <span class="Chemical">single-cylinder research engine installed at the Institute of Piston Machines and Internal Combustion Engines at the University of Rostock in Germany. The engine has a 150 mm bore and a 180 mm stroke and operates at a nominal 1500 rpm with a maximum power of 80 kW. The engine can be operated with both HFO and distillate fuels (here, MGO and diesel fuel, with diesel defined according to EN 590), and its layout represents a typical medium-speed large-bore engine. Further details are provided in Streibel et al.[34] Engines of these dimensions may be used on smaller ships as a main power supply, on large ships for ancillary power (at sea or while harbored), or as a backup power supply, e.g., in hospitals.

Aerosol samples from the engine exhaust were taken through an insulated sampling probe at 300 °C to a cyclone with a nominal cutoff diameter of 2.5 μm and then 10-fold diluted by a two-stage sampling sys<span class="Gene">tem (Venacontra, DAS, Finland). The first dilution stage employed a porous-tube dilutor[46,47] in which compressed air flowed through pores in a cylindrical sampling volume to provide a sheath flow and minimize wall losses of vapors and particles. The second stage employed an ejector dilutor with a total flow rate of 150 L min–1, which provides a good mixing of the sample with dilution air. Dilution ratios were calculated in real time from CO2(g) measurements. Four parallel PM2.5 filter samples were collected simultaneously with a modified speciation sampler (Rupprecht & Patashnik 2300, Thermo Scientific, Waltham, USA). The PM2.5 samples were collected either on quartz-fiber filters (T293, Munktell, Sweden), PTFE membrane filters (PFF, Zefluor 1 μm, Pall, USA), or polycarbonate (Millipore, UK) with a flow rate of 10 normal liters per minute each. The quartz-fiber filters were conditioned before sampling by baking at 500 °C for at least 12 h. Filters were stored in sealed glass containers until sampling. Immediately after sampling, the filter samples were stored at −25 °C. Thermal/optical EC was determined with a DIR Model 2001a carbon analyzer according to the Improve A protocol.[48]

SP-AMS and <span class="Gene">SP2 (single particle soot photometer, described below) samples were taken following an additional 100-fold dilution relative to the filter samples, using two 10-fold ejector dilutors (PALAS GmbH, Germany). To allow a comparison between the real-time and filter-based measurements, real-time measurements were averaged over the filter sampling periods (typically 30 min), unless specified otherwise. Other real-time measurements from these experiments are reported elsewhere.[44]

SEM was performed on HFO particles collected on poly<span class="Chemical">carbonate filters (Millipore, UK) at a representative engine load of 50% load. A few nanometers of gold were deposited onto the filters to enhance sample conductivity prior to measurement with a Hitachi SU3500. The elemental composition of 56 particles was measured by Energy Dispersive X-ray spectroscopy (EDX) with the SEM. Simultaneous high-resolution images of the EDX-measured particles were not obtained. As these SEM-EDX measurements were performed as a supplement to the other measurements, they were much more limited in scope.

Inductively Coupled Plasma-Optical Emission Spectroscopy (ICP-OES)

Inorganic elements were determined from PTFE filter samples u<span class="Chemical">sing inductively coupled plasma-optical emission spectroscopy (ICP-OES, Spectro Ciros Vision, SPECTRO Analytical Instruments GmbH & Co. KG, Kleve, Germany). Each precisely weighed sample was added to a quartz vessel containing concentrated aqueous HNO3 (analytical grade, Suprapur, Merck, Germany). The HNO3 was sub-boiling distilled before use. The samples were then digested in a microwave digestion system (Paar GmbH, Germany) and diluted with deionized water (Milli-Q, Millipore, Germany) before measurement. The ICP-OES analytical procedure is described in full detail in the SI.

Single Particle Soot Photometer (SP2)

A single particle soot photometer (<span class="Gene">SP2; Droplet Measurement Technologies, CO, USA) was employed to measure rBC mass concentrations by laser-induced incandescence[38,49] using a 1064 nm continuous-wave laser. Data were analyzed using the PSI SP2 Toolkit version 4.112. The SP2 quantitatively measures the rBC mass of single particles between ∼0.7 fg (∼80 nm rBC mass-equivalent diameter, drBC) and ∼200 fg (∼600 nm) with unit collection efficiency. Outside of this range, smaller particles are not detected; larger particles are still counted, but rBC mass is not quantified due to signal saturation. The rBC mass-equivalent diameter drBC was calculated according to a BC density of 1800 kg m–3.[50] The SP2 was operated as described in Laborde et al.[51] Mass calibration was performed with nebulized reference rBC particles (fullerene-enriched soot, Alfa Aesar; stock 40971, lot FS12S011), which have been validated as a reference by Laborde et al.[52] The reference rBC particles were mass-classified (Aerosol Particle Mass analyzer model 3061, Kanomax) in order to determine a linear mass-response curve. The SP2 also measures particle scattering cross sections, which provides optical sizing of the particles (including non-rBC material); the scattering detectors were calibrated using 269 nm spherical polystyrene latex size standards.

Lognormal <span class="Disease">fits to the measured drBC size distributions were used to infer correction factors to account for unmeasured rBC mass in particles smaller or larger than the size-detection limits given above. An uncommonly large correction factor of 1.46 was obtained for HFO, due to the presence of a significant fraction of high-mass particles from residual fuel that saturated the detector of the SP2. This is discussed in detail in a separate manuscript[44] and is not relevant to this manuscript, considering the overall low in-soot metal concentrations reported here.

Soot-Particle Aerosol Mass Spectrometer (SP-AMS)

The Soot-Particle Aerosol Mass Spectrometer (<span class="Chemical">SP-AMS; Aerodyne Inc., USA) employs a 1064 nm, continuous-wave laser to vaporize rBC as well as internally mixed impurities.[37] The resulting vapor molecules are neutral,[53] except for those metals with exceptionally low ionization energies.[40,54] Vaporized species are analyzed by a high-resolution time-of-flight mass spectrometer after electron ionization.

The above-mentioned vaporization laser was alternately switched on and off, <span class="Chemical">similarly to Corbin et al.[39] With the laser off, nonrefractory PM (nrPM) is flash vaporized on a 600 °C porous-tungsten vaporizer and subsequently analyzed similarly. These measurements are subsequently referred to as “AMS” measurements. The porous-tungsten vaporizer is also present in laser-on mode, in which case it vaporizes any nrPM which does not absorb sufficiently at 1064 nm, such as externally mixed OM or sulfate particles. The AMS measurements were used to obtain the reported OM concentrations, sulfate concentrations, and elemental ratios H:C and O:C following standard methods;[41,55−57] further details are provided in the SI. The ionization efficiency of the AMS (and SP-AMS) was calibrated using 300 nm mobility-diameter-selected ammonium nitrate (NH4NO3) particles.[41] The ionization efficiencies relative to nitrate (RIEs) used for OM and sulfate were 1.4 and 1.2, respectively.[58,59]

SP-AMS rBC mass concentrations <span class="Chemical">CrBC,SP-AMS were quantified from the sum of carbon-cluster ions (C1–9+) using an RIE of 0.2.[37,39,60] We note that the ratio C1+:C3+ was unity for our samples, in contrast to the previously reported value[39] of 0.6 (Figure S1). CrBC,SP-AMS is proportional to the mass of rBC reaching the SP-AMS laser vaporizer, which must depend on the transmission efficiency E of particles from the SP-AMS aerodynamic lens to its laser vaporizer. E is a function of the particle free-molecular-regime aerodynamic diameter,[61] which is a function of the soot-aggregate monomer diameter, morphology, and internal mixing.[61−64] The reliable constraint of E for SP-AMS mass quantification is a topic of current research[63,64] and outside the scope of this work, as evaluated and discussed in the SI. Below, we therefore used the SP2-measured rBC mass concentrations CrBC,SP2 to normalize CrBC,SP-AMS (eq ) since the reliability of the SP2 in measuring rBC mass concentrations independently of mixing state has been thoroughly confirmed by previous studies.[52,65,66]

We quantified in-soot metal concentrations with the <span class="Chemical">SP-AMS using the method recently demonstrated by Carbone et al.[40] That is, we used the RIEs reported by Carbone et al. to quantify the concentrations of V, Ni, Fe, Ba, and Na vaporized by the SP-AMS laser as CM,SP-AMS. We note that the RIEs reported by Carbone et al. for Na and Ba (only) result in conservative lower-limit concentration estimates, as detailed in the SI. This does not apply to the concentrations reported for V, Ni, or Fe, for which quantification is accurate.

Each metal <span class="Chemical">signal was quantified from the signal of its most abundant isotope. These isotopic signals were integrated to obtain peak areas, which were then divided by the relative abundance of that isotope to yield the total mass concentration CM,SP-AMS of that metal. A peak-specific SP-AMS limit of quantification of 20 ng m–3 was estimated for all metal ions via a novel procedure, described in the SI. Each relative abundance was directly measured and validated as consistent with literature. A small mass fraction of Fe (12%) and V (3.5%) was present as metal-oxide ions (MO+); the mass of Fe or V in oxide ions was quantified and added to the corresponding mass of metal ions when reporting the mass concentrations of these species. The SI provides complete details on this novel procedure, as well as on the above-mentioned isotopic abundances and metal-oxide ions.

Calculated Quantities

The mass ratio of a given metal M to BC in <span class="Chemical">PM2.5 was calculated aswhere CM,ICP-OES is the mass concentration of that metal M in the PM2.5, and CEC,PM is the EC mass concentration measured by the thermal–optical IMPROVE protocol after the PM2.5 cyclone for the same filter, representing CrBC,PM, which previous studies have found to agree well with rBC mass concentrations.[52,65,67]

The mass ratio of a given metal M to rBC in soot particles was determined aswhere CM,<span class="Chemical">SP-AMS and CrBC,SP-AMS are the mass concentrations of metal M and of rBC measured by SP-AMS. We observe that the ratio (M:BC)in-soot is approximately equal to the mass fraction of the metal in soot, since CM,SP-AMS was on the order of 1% and since the presence of nonrefractory material internally mixed with the measured soot particles was negligible, as shown elsewhere.[44] However, the discussion below is not dependent upon this observation.

Following from eq , the emission factor of a given <span class="Chemical">metal M per mass of CO2 is

This equation uses CrBC,<span class="Gene">SP2 rather than CrBC,SP-AMS or CEC,PM, respectively, because of the E issues noted above and because of the high time resolution of the SP2 relative to the thermal–optical analysis. Emission factors for other PM components were calculated similarly to eq , replacing CM with the appropriate measurement.

Uncertainties

ICP-OES PM2.5 uncertainties were c<span class="Chemical">alculated as the quadratic sum of the standard deviation of the background, and 5% of the signal of the latter is an estimated proportional error[68]. When averaging data to ICP-OES-filter-sampling periods, the standard error of the mean was used.

SP-AMS and <span class="Chemical">AMS uncertainties were calculated by combining peak-integration uncertainties with ion-counting uncertainties.[69] Peak-integration uncertainties were obtained with an improved-efficiency implementation of the Monte Carlo calculations described by Corbin et al.[69] The peak-width uncertainty used was 3%. We estimate that the worst-case uncertainty of the metal RIEs reported by Carbone et al.[40] are on the order of ∼20%.

SP2 c<span class="Chemical">alibration and particle-counting uncertainties were generally smaller than the engine variability, such that the standard error of a typical mean rBC mass concentration (representing temporal variability) was larger than the propagated statistical uncertainty. The SP2 correction factor for the mass fraction of rBC particles that were too large or too small to be quantified, described above, resulted in a correction factor for HFO of 1.46 ±  0.15, where 0.15 is the estimated day-to-day variability of the 1.46 correction factor.

Results and Discussion

Overview of PM2.5 Composition

The composition of the <span class="Gene">ship-engine HFO PM was dominated by OM, BC, and sulfates. Since ammonia and nitrate emission factors were negligible, we infer that the dominant sulfate species was sulfuric acid. The pie chart in Figure shows the mean relative emission factors of OM, BC, sulfate, and ash for a typical engine load of 50%. Note that these emission factors reflect the composition of the primary particulate emissions of this engine and do not account for the formation of secondary sulfate and secondary OM, which may occur in the atmosphere on a time scale of hours for OM and days for sulfate. More details on these emission factors are given in Corbin et al.,[44] and the relationship between such emission factors and engine load has been discussed by Müller et al.[70] Ash is defined as all elements other than CHOS and was measured by ICP-OES. Note that this definition excludes the oxygen contained in metal oxides.

Figure 1

Trace-element composition of the fuel (mass fraction) and of the PM2.5 and soot (emission factors). All elements except C, H, O, and S are shown. The inset pie chart shows the overall PM composition, including OM elemental composition. Uncertainties are standard errors of the mean (s.e.m.) for the PM samples and 10% for the fuel composition. Si in PM2.5 was not determined (n.d.). Other missing bars correspond to below-LOD signals; LODs are given in the text and in Tables S2 and S3.

Trace-element composition of the <span class="Chemical">fuel (mass fraction) and of the PM2.5 and soot (emission factors). All elements except C, H, O, and S are shown. The inset pie chart shows the overall PM composition, including OM elemental composition. Uncertainties are standard errors of the mean (s.e.m.) for the PM samples and 10% for the fuel composition. Si in PM2.5 was not determined (n.d.). Other missing bars correspond to below-LOD signals; LODs are given in the text and in Tables S2 and S3.

The pie chart of Figure also includes the OM to OC mass ratio (OM/OC) and the element<span class="Chemical">al ratios H/C and O/C, all from the high-resolution AMS data[55,56] (see the SI for details). The OM/OC, required to convert thermally evolved carbon measurements to total OM mass, was 1.2 on average. This is lower than the value of 1.6 recommended for urban aerosols by Turpin and Lim,[71] due to the near absence of oxygen in our OM (O/C 0.02  ± 0.02, H/C 1.87  ± 0.07).

Using SEM, we identified soot aggregates, spheric<span class="Chemical">al, tar-ball-like particles, and irregular ash particles in the HFO PM. No porous char cenospheres[45] were observed. This observation rules out the hypothesis that supermicron char particles, not detectable by SP-AMS due to their size, influenced our results. Applying SEM-EDX to 56 randomly chosen micron-sized particles, we detected Ca, S, Al, Zn, V, Si, and Ni in irregular combinations. This irregularity highlights the strength of the SP-AMS, which sampled ∼105 particles per second, such that counting statistics are a negligible source of uncertainty. Obtaining a comparable SEM-EDX data set was infeasible for this study. We reiterate here that the SP-AMS methodology we have employed has been directly validated in the laboratory[40] and does not rely on the SEM for quantification.

Emission Factors of Trace Metals in Soot Particles

The trace-element compositions of the <span class="Chemical">fuel (relative to the bulk mass) and PM2.5 (as emission factors) are shown by the leftmost and central bars in Figure , respectively. These compositions were both measured by ICP-OES. Overall, HFO contained ∼110 mg/kg of trace metals (with signals above LOD), whereas MGO and diesel contained negligible amounts (only Fe in MGO was above the general detection limit of 0.1 mg/kg, at 0.2 mg/kg). Trace elements in both the HFO fuel and PM were dominated by V, with Ni, Na, Ca, Fe, and other less-abundant species present. The relative presence of V and Ni is discussed further below. The ratio of fuel composition to emission factor for each element was ∼2–10 in PM2.5 and ∼100 in soot (Figure , Figure S4).

The rightmost bars for each <span class="Chemical">metal in Figure show that the emission factors of metal inclusions in soot particles, as measured by the SP-AMS (“in-soot metals”, eq ), were much lower than for the PM2.5. This is primarily because the PM2.5 measurements additionally include externally mixed metal oxides and salts. V concentrations were a factor of 25 lower in soot than in the PM2.5. For Ni, this factor was 84. Given that the rBC mass was a factor of four lower than the overall PM2.5 mass (Figure ), this translates to a relative depletion of metals in soot by about an order of magnitude, compared to the metals present in the overall PM2.5. The error bars in Figure indicate standard errors of the respective means. These were much larger for in-soot metals than for in-PM2.5 metals, indicating a substantial variability in the internal mixing of soot with each metal. Note that this variability does not indicate noise in the SP-AMS data; it is much greater than the expected instrument precision (discussed in the SI). In addition, the metal-in-soot concentrations were not correlated with the total PM2.5 metal concentrations (Figure S9), demonstrating the independence of these two quantities.

We note that the reported metal ions were gener<span class="Chemical">ally atomic ions (M+) of the respective metals, with two exceptions. First, a very small mass fraction of V (3.5%) was observed as VO+ (see the SI and Table S5). Second, a more substantial mass fraction of Fe (12%) was observed as FeO+ and Fe2O+. Oxide ions of all other reported metals were not present. This indicates that the metallic species present in our samples were either metallic compounds or oxides which thermally degraded upon heating.[72]

We did not observe metal-<span class="Chemical">sulfate ions in the SP-AMS mass spectra, in spite of the presence of substantial amounts of sulfate mass in the PM (Figure ). This is not a limitation of the SP-AMS. Carbone et al.[40] used an SP-AMS to measure VSO, CaSO, and FeSO ions in PM generated by a heating station operated on an aqueous emulsion of HFO and distillate fuel. This difference between our results and those of Carbone et al. provides physical insight into the composition of HFO PM. In particular, since emulsions lead to lower combustion temperatures,[73] resulting in the preferential formation of metal sulfates rather than oxides,[18,32] the absence of metal sulfates in our mass spectra represents an indicator of combustion temperature. Our results therefore indicate that studies which use laboratory combustors[74,75] may produce PM which is not representative of the HFO PM produced by four-stroke marine diesel engines. Toxicological studies should therefore use ship engines[13] rather than boilers[12,76] to simulate marine pollution.

Metal to BC Mass Ratios in Soot and PM2.5

The mass ratios of each quantified metal to BC are shown in Figure for each <span class="Chemical">metal where in-soot concentrations were quantified by SP-AMS. These ratios are shown for the specific cases of in-soot mass, (M:BC)in-soot, and in-PM2.5 mass (M:BC)in-PM. The ratio (M:BC)in-soot was less than 0.03 in all cases, corresponding to a few percent or less mass contribution to the refractory mass of soot particles. For toxicology studies, this depletion means that only a small fraction of the metal mass is rendered biologically inaccessible by incorporation into soot particles.

Figure 2

Metal:BC mass ratios in soot and in PM2.5. The ratios were obtained from the in-soot and in-PM2.5 metal concentrations (defined respectively by eq and eq ) in combination with the in-soot BC mass or all-PM2.5 BC mass, as defined in those equations. For soot, the data points are colored darker for higher CSP-AMS,M concentrations. Fewer data points are plotted for soot due to limited availability of SP2 data. Boxes show 25–50–75th percentiles; error bars are s.e.m.

Metal:BC mass ratios in soot and in <span class="Chemical">PM2.5. The ratios were obtained from the in-soot and in-PM2.5 metal concentrations (defined respectively by eq and eq ) in combination with the in-soot BC mass or all-PM2.5 BC mass, as defined in those equations. For soot, the data points are colored darker for higher CSP-AMS,M concentrations. Fewer data points are plotted for soot due to limited availability of SP2 data. Boxes show 25–50–75th percentiles; error bars are s.e.m.

However, this small mass fraction may be sufficient to cat<span class="Chemical">alyze BC oxidation within the engine. Such BC oxidation is a major factor in determining the BC emissions of an engine, since the vast majority of BC produced by a flame is generally destroyed (oxidized) before leaving it.[42]

To evaluate this hypothe<span class="Chemical">sis, we compare our measured in-soot metal content to the data reported by Bladt et al.,[35] who observed a large change in temperature-programmed oxidation results for BC with 10  mass% of Fe and higher. Although those authors did not report measurements below 5 mass%, their reported exponential dependence of oxidation temperature on Fe mass% suggests that our observed few percent metal content may have a significant influence on BC emissions. We also note that in addition to BC oxidation, the presence of metals may influence BC formation.[36] These effects would mean that the soot emission factors of HFO are a function of the metal content of the fuel, which is expected to vary significantly between fuels.[77] Further study is required to establish whether this fuel-to-fuel variability is substantial in the context of engine-to-engine variation.

The high outlier for Fe in Figure is noteworthy, suggesting that (<span class="Chemical">Fe:BC)in-soot may have been externally mixed for this point. Uniquely to all other SP-AMS-quantified metals, this outlier corresponds to a period of measurement where CSP-AMS,Fe was poorly correlated with CSP-AMS,rBC. We conclude that this point was likely biased high due to the anomalous presence of light-absorbing Fe or Fe compounds.

Enhanced Vanadium-to-Nickel Ratio in Soot

The ratio V:Ni has been recommended for the source apportionment of residu<span class="Chemical">al-fuel PM in atmospheric studies. In this section we examine whether this ratio changes between in-PM2.5 and in-soot metals. Also, since metals are incorporated into soot particles by vaporization/condensation, the metal-in-soot ratios provide quantitative insights into the degree to which some metals may be enhanced in ship-exhaust PM2.5 or PM0.1 particles relative to coarse-mode PM.

Our PM2.5 data (Figure ) indicate a mean and standard deviation V:Ni of 5.2 ± 0.5. This ratio is comparable to those reported by two previous studies. <span class="Chemical">Based on a source-apportionment analysis of atmospheric data, Viana et al.[78] suggested that a V:Ni of 4–5 is representative of HFO combustion. Consistent with this range of values, Agrawal et al.[79] observed a ratio of 4.5 ± 0.1 for an in-use ship. However, Lyyränen et al.[16] reported a lower ratio, close to unity, possibly due to a difference in fuel composition. Our PM2.5 data do not show a dependence of V:Ni in PM2.5 on the engine load, also consistent with Agrawal et al.[79]

Figure 3

Mass ratio V:Ni in soot particles and in the PM2.5 of HFO exhaust. Also shown are the ranges of values measured in the fuel itself (green thin line) and in field measurements of atmospheric particles (gray shading, measured by Viana et al.[78]). The horizontal scatter of the PM2.5 data is added for clarity only. Error bars are s.e.m.

Mass ratio V:Ni in soot particles and in the PM2.5 of <span class="Chemical">HFO exhaust. Also shown are the ranges of values measured in the fuel itself (green thin line) and in field measurements of atmospheric particles (gray shading, measured by Viana et al.[78]). The horizontal scatter of the PM2.5 data is added for clarity only. Error bars are s.e.m.

Whereas V:Ni was similar between the <span class="Chemical">fuel and the PM2.5, V:Ni in soot particles was 20–100, which is 4- to 25-times higher than in the PM2.5. This enhancement is well outside of the SP-AMS uncertainty range. Source apportionment studies based on SP-AMS or other laser-vaporization instruments with high sensitivities to BC, such as laser desorption/ionization,[5,7,80,81] will therefore measure different V:Ni ratios than ICP-OES or similar techniques. We note that the in-soot V:Ni appeared to increase at lower engine loads, in contrast to the in-PM2.5 V:Ni. Additional measurements at high engine loads are needed in future studies to corroborate this potential trend.

We also at<span class="Gene">tempted to measure the composition of micron-sized particles using SEM-EDX, for comparison with the bulk PM2.5 and soot particles results. However, this method was of limited statistical power, as we found that sufficient amounts of V and Ni were present in only 3 of 56 measured particles. The V:Ni ratios of these particles was 2.44 ± 0.34, 1.07 ± 0.68, and 1.23 ± 0.28. This limited evidence suggests that the V:Ni ratio of supermicron particles more closely resembles that of PM2.5 than that of soot.

When switching from HFO <span class="Chemical">fuel to distillate fuels, we did not observe V or Ni in the emissions. Thus, the possibilities of V and Ni originating directly from the engine itself (e.g., via abrasion) as well as the interference of resuspended particles can be ruled out. Also, this shows that V:Ni is a specific tracer for HFO emissions and allows ship-exhaust to be identified only when HFO is the fuel. This presents a challenge to future ship-exhaust source apportionment studies, as HFO emissions must be minimized within SECAs.[82] We note that the use of exhaust scrubbers[83] can be used to accomplish this minimization and that exhaust scrubbers are least effective for the smallest size fraction.[26] Therefore, our observed enrichment of toxic vanadium in soot particles indicates a per-mass enhanced toxicity in scrubbed HFO exhaust.

Overall, we conclude that <span class="Chemical">vanadium inclusions in soot were more likely to form than nickel inclusions. We compare this conclusion with the thermodynamical calculations of Sippula et al.,[18] which indicate that VO2 should be the first species to nucleate in the engine, followed by iron and nickel (at 1200–1300 °C). Our observations are in agreement with these predictions.

Implications for HFO PM Toxicity

Since our observed enhancement of <span class="Chemical">vanadium in soot particles was attributed to the higher-temperature condensation of vanadium in the engine compared to other metals, this enhancement is independent of the chemistry that leads to soot formation. Therefore, V is expected to be enhanced not only in soot particles but also in all small (diameters < 100 nm) particles emitted by HFO combustion in marine diesel engines. This would mean that the V:Ni relevant for source-apportionment of atmospheric PM to HFO particles may depend on the measured size fraction. Moreover, since vanadium is especially toxic,[27] and since the penetration of particles into the human lungs is size-dependent, with smaller particles penetrating deeper into the lungs and alveoli, this would correspond to an enhanced toxicity of smaller HFO particles due to their enhanced vanadium content. Although incorporation into soot particles would render this vanadium biologically inaccessible, we observed (above) that this effect sequesters <3% of the total PM2.5 vanadium. Relevant to this conclusion, we note that the translocation of insoluble PM into the blood and subsequent accumulation in organs is a strong function of particle size, being significantly enhanced for diameters below ∼100 nm.[24] The temperature dependence of this small-particle enhancement in V also indicates that the toxicity of HFO PM is dependent on the combustion conditions, in particular on temperature, for a given fuel. When performing size-resolved PM exposures, size-resolved chemical composition determination is therefore essential.