Rethinking Organic Aerosols: Semivolatile Emissions and Photochemical Aging

See allHide authors and affiliations

Science  02 Mar 2007:
Vol. 315, Issue 5816, pp. 1259-1262
DOI: 10.1126/science.1133061


Most primary organic-particulate emissions are semivolatile; thus, they partially evaporate with atmospheric dilution, creating substantial amounts of low-volatility gas-phase material. Laboratory experiments show that photo-oxidation of diesel emissions rapidly generates organic aerosol, greatly exceeding the contribution from known secondary organic-aerosol precursors. We attribute this unexplained secondary organic-aerosol production to the oxidation of low-volatility gas-phase species. Accounting for partitioning and photochemical processing of primary emissions creates a more regionally distributed aerosol and brings model predictions into better agreement with observations. Controlling organic particulate-matter concentrations will require substantial changes in the approaches that are currently used to measure and regulate emissions.

Airborne particles pose serious health risks and have a controlling influence on Earth's climate. Organic aerosol (OA) is a major component of fine-particle mass throughout the atmosphere (1). OA comprises primary organic aerosol (POA, particle mass directly emitted from sources such as motor vehicles and forest fires) and secondary organic aerosol (SOA, particle mass formed in the atmosphere from the oxidation of gas-phase precursors) (1). The relative contribution of POA and SOA to the overall OA budget remains controversial. Research using individual organic compounds to estimate POA (2, 3), organic-to-elemental C ratios to estimate SOA (4, 5), and atmospheric chemistry models (6, 7) indicates that POA dominates the urban OA budget. However, recent field measurements indicate SOA dominance, even in heavily urbanized areas, that cannot be explained by the oxidation of known SOA precursors (810). The persistent discrepancies between measured OA concentrations and predictions of atmospheric chemistry models underscore the substantial uncertainty regarding the sources of OA (6, 11, 12).

Combining laboratory, field, and modeling results, we propose two major amendments to the current OA conceptual framework: (i) accounting for gas-particle partitioning of POA and (ii) explicitly representing gas-phase oxidation of all low-volatility vapors in current SOA-production mechanisms. This replaces the current static representation of POA emissions with a far more dynamic picture in which low-volatility material evaporates, oxidizes, and recondenses over time. This paper presents experimental results supporting this revised framework, first showing that primary emissions indeed evaporate substantially upon dilution to ambient conditions and then showing that photo-oxidation of those vapors in a smog chamber produces SOA far exceeding that from traditional precursors. Finally, we implement these amendments in a chemical transport model to illustrate their implications for our understanding of OA in the atmosphere.

Sources emit thousands of organic compounds, ranging from CH4 to species with 30-or-more C atoms (1315). The amount of POA depends on the gas-particle partitioning of this complex mixture (16, 17). However, emission inventories and models treat POA as nonvolatile, implicitly assuming that standard emissions tests represent the full range of atmospheric conditions. This is wrong (17). To illustrate the semivolatile character of POA, Fig. 1 presents a compilation of diesel exhaust data measured at different levels of dilution, extending from conventional emissions sampling to typical atmospheric conditions. The POA emission factor (EF) decreases with increasing dilution because of the evaporation of semivolatile organic compounds (SVOCs) (17). This evaporation causes POA concentrations to decrease considerably more than does dispersion alone. Furthermore, the data follow well-established partitioning theory (17, 18), indicating that POA levels also vary with temperature. This dynamic picture complicates the definition of POA and underscores the need for our first amendment to the current OA conceptual framework: explicit treatment of gas-particle partitioning of primary emissions.

Fig. 1.

Partitioning data and volatility distribution of diesel POA measured at 300 K. The circles in (A) indicate POA EF data from previously published dilution-sampler measurements (21) and new results obtained at higher levels of dilution. The data are plotted as a function of OA concentration (COA), following the principles of partitioning theory (17, 18). The values are normalized by the traditional POA EF, measured by means of a quartz filter at low levels of dilution. The results show substantial evaporation of POA with decreasing COA and that only a quarter of the traditionally defined POA exists in the particle phase at atmospherically relevant COA. The curve shows a fit based on absorptive partitioning (19), along with a 95% confidence interval (CI). The red bars in (B) are the volatility distribution determined by this fit (19), plotted in terms of C* (related to effective saturation vapor pressure). These bars reveal the volatility distribution of traditionally defined nonvolatile POA emissions and therefore sum to one (Σ bars = 1) (19). The hatched bars show an assumed volatility distribution of additional IVOC emissions estimated from other data (1315, 19) that indicate that IVOCs contribute one to threetimes the POA emissions (here, we assume 1.5 times) (19). Secondary axes show un-normalized fuel-based EFs and levels of dilution for our experiments.

Partitioning depends on the volatility distribution of the emissions. Figure 1B shows the volatility distribution for diesel exhaust at 300 K. It is presented in terms of lumped species that span a volatility basis set of effective saturation concentrations (C*), allowing a physically meaningful treatment of partitioning for all organics (18). Conceptually, a volatility distribution can be constructed from speciated emissions data by lumping species with similar saturation vapor pressures. However, this is not possible because less than 10% of the condensed and semivolatile mass has been speciated (1315). We determined the distribution in Fig. 1B by fitting isothermal dilution data with absorptive partitioning theory (19), as done in well-established analyses of SOA data (20).

The distribution shown in Fig. 1B comprises emissions of all organics less volatile than approximately a C12 n-alkane, covering the range of material that can change phase between the high-temperature conditions at the end of the exhaust pipe and the cool, highly dilute ambient atmosphere. Based on partitioning at typical atmospheric conditions, this distribution includes “nonvolatile” (C* < 0.1 μg m–3), “semivolatile” (SVOC; 0.1 μg m–3 < C* < 1000 μg m–3), and “intermediate-volatility” (IVOC; 1000 μg m–3 < C* < 100,000 μg m–3) organic compounds. Even more volatile species dominate the overall emissions of reduced organic C (1315); however, these species are largely accounted for in models, either as part of ozone chemistry mechanisms or as traditional SOA precursors.

The red bars in Fig. 1B show the volatility distribution of the traditionally defined POA emissions that are currently treated as nonvolatile in models and inventories. The majority of these emissions have a C* > 100 μg m–3 and therefore exist largely in the gas phase at typical atmospheric conditions. Some of the emissions are misclassified as POA because dilution-sampler measurements are typically conducted at unrealistically high concentrations, which biases gas-particle partitioning relative to much more dilute atmospheric conditions (17, 21).

Most of the emissions shown in Fig. 1B are vapors in the atmosphere, but given their low volatility, they may be important SOA precursors. The first several generations of oxidation typically produce compounds with lowered vapor pressures (19, 22). Low-volatility vapors should also produce SOA much more efficiently than the traditional high-volatility SOA precursors (such as monoterpenes and light aromatics) that are thought to dominate ambient SOA production (6, 7, 23). Figure 1B indicates that the aggregate emissions of low-volatility vapors are substantial compared with the POA emissions. However, these vapors are largely uncharacterized, instead appearing as an unresolved complex mixture of presumably branched aliphatic and cyclic hydrocarbons (13, 14).

To investigate this hypothesis, diluted diesel exhaust was exposed to ultraviolet (UV) light inside our environmental chamber at initial aerosol loadings near typical ambient conditions (19). The evolution of both the gas and the particle phase was monitored. Typical results are shown in Fig. 2. UV illumination initiated photochemistry, which caused an initial burst of SOA formation, followed by steady production through the remainder of the experiment. After 3 hours of aging, SOA has almost doubled the initial aerosol mass. Approximately half of the trimethylbenzene was oxidized over the course of the experiment, indicating roughly one generation of processing and ∼2 × 106 molecules cm–3 of OH, which is typical of a summer day (24).

Fig. 2.

Results from the photochemical oxidation of diesel exhaust in an environmental chamber. (A) The wall-loss–corrected aerosol mass, measured with a scanning mobility particle sizer, assuming a density of 1 g cm–3. The wall-losscorrection is based on measured loss of the particle number, accounting for the effects of coagulation, and it assumes that semivolatile vapors are in equilibrium with both the suspended aerosol and the material deposited on the wall during the experiment (19). The gray area indicates the primary aerosol (POA + other species). The red area shows the upper-bound estimate of the contribution of known SOA precursors to the suspended aerosol mass (19). We attribute the blue area to SOA formed from SVOC and/or IVOC oxidation. AMS results are shown in (B) to (D). The AMS OA spectrum can be described by two components: the initial diesel spectrum (diesel OA) and an oxidized residual spectrum (residual OA). The relative contribution of these two components to the overall OA spectrum is shown in (B). The spectra of these two components are compared to reference spectra obtained from factor analysis of ambient AMS data in (C) and (D). The spectrum of the diesel OA component is quite similar to the hydrocarbonlike OA (HOA) factor (8, 9). By the end of the experiment, the spectrum of the oxidized residual OA component is quite similar to that of the OOA factor (8, 9). The solid line in (B) indicates the fractional contribution of POA to the suspended OA, based on the initial particle mass and the wall-loss rate and analogous to the results shown in (A). The excellent agreement between this line and the relative contribution of the two AMS components indicates that these two independent approaches yield the same estimate of SOA. m/z, mass/charge ratio.

We examined the potential contribution of traditional-SOA precursors empirically by spiking the chamber with additional aromatics. Adding enough toluene to roughly double the potential SOA production from the measured aromatics caused only a slight inflection in the aerosol-mass time series. Oxidation of traditional precursors can thus explain only a small fraction of the SOA formed in the chamber.

We also calculated the traditional SOA production using measured precursor consumption and published SOA yield curves (19). Our model accounts for 58 SOA precursors, but ∼90% of the calculated SOA production is from the measured decay of light aromatics, such as toluene. Initial concentrations of these species were at most a few parts per billion by volume. The calculation is complicated by losses of particles and vapor to the wall of the environmental chamber (19). Our maximum estimate of the contribution of SOA from the oxidation of known precursors is shown in Fig. 2A. It indicates that traditional SOA contributes at most 15% of the new aerosol, we hypothesize that the majority of the substantial unexplained SOA production is due to the oxidation of low-volatility gas-phase species—our second amendment to the current OA conceptual framework.

SVOC and IVOC vapors constitute a potentially large source of SOA. Their mass emission rate is several times that of condensed-phase compounds (Fig. 1B) (19); therefore, SOA produced from the oxidation of these vapors will probably exceed the POA emissions. These vapors also oxidize on atmospherically relevant time scales. Large saturated organics have OH rate constants on the order of 3 × 10–11 cm3 per molecule s–1 (25) and therefore will undergo one generation of oxidation every 4 hours at typical summertime OH levels. Winter OH levels are three to five times lower (26), resulting in slower but still significant processing over multiday time scales associated with regional transport. Finally, the total non-CH4 hydrocarbon burden in urban environments can exceed 1000 μg m–3 (27), as compared to typical OA concentrations of less than 10 μg m–3. Therefore, even at only a few percent of the total hydrocarbon budget, SVOC and IVOC vapors could be an important source of ambient OA.

The particle composition during our chamber experiments was characterized with an Aerodyne aerosol mass spectrometer (AMS). The measured OA mass spectra can be described as a combination of two components (19): (i) diesel OA, whose spectrum is the same as that of the fresh primary emissions (Fig. 2D), and (ii) an oxidized residual that becomes progressively more oxidized during the experiment. By the end of the experiment, the oxidized residual is quite similar to the oxygenated OA (OOA) factor that often dominates ambient OA levels in Pittsburgh, Pennsylvania, and elsewhere (8, 9). This oxidized residual does not look like the mass spectra of SOA formed from aromatics (28), confirming that traditional precursors are a minor source of SOA in these experiments. Figure 2B shows that the relative contribution of these two mass-spectrum components evolves over time. This evolution matches the independent estimate of the primary-secondary split, based on the measured wall loss. Therefore, the relative contribution of the oxidized residual spectrum provides a good estimate of SOA, supporting the conclusion drawn from ambient AMS data that SOA dominates the ambient OA burden (8, 9).

To investigate the implications of our two major amendments, we used a three-dimensional chemical transport model, PMCAMx, to simulate pollutant concentrations across the eastern half of the United States (19). The model was modified to account for the volatility distribution and gas-particle partitioning of POA and to more explicitly represent SOA production from the oxidation of IVOC and/or SVOC vapors by means of a framework that is consistent with our experimental data (19).

Figure 3 presents maps of ground-level OA concentrations from four simulations. The traditional model with nonvolatile POA (Fig. 3A) predicts high POA concentrations (>3 μg m–3) in heavily urbanized areas and substantial urban-to-regional concentration gradients. Allowing the primary emissions to partition but not react (Fig. 3B) dramatically reduces the POA levels throughout the modeling domain, indicating that the majority of the traditional POA emissions are actually evaporated at ambient OA levels (Fig. 1A). Photochemical aging of the SVOC and IVOC vapors creates a considerable amount of regional SOA (Fig. 3, C and D). If one accounts only for partitioning and aging of the existing POA emissions (Fig. 3C), the predicted OA levels are lower than the traditional model. Therefore, even with aging, evaporation of traditional POA reduces OA concentrations. The most comprehensive simulation (Fig. 3D) adds additional IVOC emissions to the model. This creates large amounts of regional SOA. The net result is that regional (but not urban) OA levels exceed the traditional model.

Fig. 3.

Maps of predicted ground-level OA concentrations for four PMCAMx simulations: (A) a traditional model with nonvolatile POA emissions and (B to D) three simulations that account for the partitioning of primary emissions—one assuming nonreactive emissions and two considering photochemical aging. In (B), the red bars in Fig. 1B are used to represent the volatility distribution of all POA emissions, and the emissions are allowed to partition but not react; this results in substantial evaporation of the POA (19). In (C) and (D), gas-phase primary emissions are also aged by OH (19). (C) shows the same emissions scenario as (B), whereas (D) includes additional IVOC emissions by applying the entire volatility distribution shown in Fig. 1B to the existing POA emissions (19). The explicit representation of photochemical aging of IVOC and SVOC vapors used in (C) and (D) substantially increases the amount of anthropogenic SOA. In the traditional model, the oxidation of low-volatility vapors contributes only 25% of anthropogenic SOA, as compared to 85% in the revised model shown in (D). The balance of the anthropogenic SOA is from the oxidation of aromatics, which contributes essentially the same amount of SOA on an absolute basis in all simulations. The maps present averages over an 8-day period in July 2001 and show only a subset of the modeling domain (19).

The effects of our amendments on both the POA/SOA split and the total OA burden are shown in Fig. 4. The traditional model predicts substantial contributions from POA, whereas the revised model predicts that ambient OA is dominated by SOA during the summer. Such a shift is consistent with recent field measurements indicating dominant contributions from SOA (810) while remaining consistent with POA estimates based on low-volatility tracers. In terms of the overall OA budget, the revised model decreases predicted OA in urban areas by as much as 50% and increases it in many rural areas by 15 to 30% (Fig. 4C), reducing the large urban-to-regional gradients predicted by the traditional model and resulting in considerably better agreement with measured urban-to-regional OA ratios (Fig. 4D).

Fig. 4.

Predicted changes in the POA/SOA split and total OA between the current framework and the revised model (results shown in Fig. 3, A and D). (A) and (B) show the ratio of SOA to total OA for the two cases, respectively. (C) shows the ratio of the OA predictions for the two cases. (D) compares average measured urban-to-regional OA ratios to model predictions for four large cities. The measured ratios (yellow bars) are based on data from the U.S. EPA Speciation Trends Network. The regional concentrations are estimated with the use of a site located upwind of the city. The predicted ratios are based on the average OA concentrations for the grid cells in which the monitoring stations are located. For all cities, the base-case model significantly overpredicts the urban-to-regional ratio, whereas the revised model shows much better agreement. Balt, Baltimore, Maryland; NYC, New York City, New York; Pgh, Pittsburgh, Pennsylvania; Phil, Philadelphia, Pennsylvania.

This work has several implications for our understanding of OA. The semivolatile character of primary emissions requires that instead of measuring fixed POA EFs, we must measure the volatility distribution of the emissions. Models and inventories must account for these distributions and their evolution with photochemical age. Regulations and control technologies may also need to be revised to control SVOC and IVOC emissions because of their importance as SOA precursors. The results also imply that, except for people living close to sources, the majority of the population (even in urban areas) is exposed mostly to SOA. Ultimately, a relatively local urban emissions problem is transformed into a regional source of oxidized and presumably hydrophilic OA. The health consequences and climate effects of this oxidized material are almost certainly dramatically different from those of primary emissions.

Supporting Online Material

Materials and Methods

Table S1


References and Notes

View Abstract

Navigate This Article