Environ. Sci. Technol. 2006, 40, 3536-3543
Investigation of r-Pinene + Ozone Secondary Organic Aerosol Formation at Low Total Aerosol Mass ALBERT A. PRESTO† AND NEIL M. DONAHUE* Center for Atmospheric Particle Studies, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213
We present a method for measuring secondary organic aerosol (SOA) production at low total organic mass concentration (COA) using proton-transfer reaction mass spectrometry (PTR-MS). PTR-MS provides high time resolution measurements of gas-phase organic species and, coupled with particle measurements, allows for the determination of aerosol yield in real time. This approach facilitates the measurement of aerosol production at low COA; in fact aerosol mass fractions can be measured during R-pinene consumption as opposed to only at the completion of gasphase chemistry. The high time resolution data are consistent with both the partitioning theory of Pankow (Atmos. Environ. 1994, 28, 185 and 189) and the previous experimental measurements. Experiments including the effect of UV illumination and NOx reveal additional features of R-pinene + ozone product photochemistry and volatility. The high time resolution data also elucidate aerosol production from R-pinene ozonolysis at COA < 10 µg m-3 and show that extrapolations of current partitioning models to conditions of low COA significantly underestimate SOA production under dark, low-NOx conditions. However, extrapolations of current models overestimate SOA production under illuminated, higher-NOx conditions typical of polluted regional air masses.
1. Introduction The production of secondary organic aerosol (SOA) from the ozonolysis of monoterpenes (C10H16) is well-documented. (1-12) Oxidative cleavage of endocyclic carbon-carbon double bonds found in many monterpenes (i.e., R-pinene and limonene) leads to the production of multifunctional species. These products often contain alchohol, carbonyl, or acid moeities and therefore exhibit lower saturation vapor pressures than the parent molecule (7, 9, 11, 13). For many species, ozonolysis is a more potent source of SOA than oxidation by either OH or NO3 radicals (4), in part because of the multifunctional products generated by ozonolysis. It is very important for SOA production experiments to be placed in the context of ambient organic carbon levels. A difficulty in all SOA studies is to capture semivolatile partitioning (and thus SOA production) at relatively low total organic aerosol mass (COA), specifically between 1 and 10 µg * Corresponding author phone: (412)268-4603; fax: (412)268-7813; e-mail:
[email protected]. † Current address: National Energy Technology Laboratory, Pittsburgh, PA 15236. 3536
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 11, 2006
m-3; experimental signal-to-noise ratios are challenged by both low COA and long kinetic time scales in low-concentration experiments. Consequently, the vast majority of SOA experiments have been conducted at elevated COA, and application to ambient conditions requires model-dependent extrapolation. This is a significant shortcoming, as COA e 5 µg m-3 is typical of the continental atmosphere, comprising a significant fraction of regional aerosol exceeding the 15 µg m-3 annual average standard of the U. S. EPA (14, 15). The important quantity is the aerosol mass fraction (AMF ) ξ), the particle mass produced following the oxidation of a given precursor mass; ξ ) COA/∆CROG, where COA (µg m-3) is the mass of aerosol formed by the oxidation of ∆CROG (µg m-3). This is conventionally called a “yield”, but we reserve yield to describe the mass yield (R) of chemical products; ROG f R1p1 + R2p2 +... Chemical yields are assumed to remain constant in groups of experiments over a range of COA, while ξ is variable because of semivolatile partitioning effects. Because partitioning is the key observable, it is most important that experiments be conducted over ambient ranges of COA; ambient terpene levels are not necessary or even desirable, unless extremely difficult experiments using organic seeds are being conducted. Variable chemical yields due to gas-phase chemistry are also a challenge to SOA experiments. Most past investigations of SOA production via terpene ozonolysis have focused on dark, NOx-free (or low-NOx) conditions (3, 4, 7, 8, 10). There are practical reasons for this choice. Experiments are typically conducted in environmental (or “smog”) chambers, and maintaining a dark chamber during ozonolysis limits interference from photochemically generated OH radicals (16). Ozone reacts rapidly with NOx (16), particularly NO; thus ozonolysis experiments that contain NOx face possible interference from the products of O3-NOx reactions, including NO3 radicals. Relying solely on these dark, NOx-free conditions, however, presents a serious limitation on the application of experimental results to the atmosphere. The moeities responsible for the lowered vapor pressures of the ozonolysis products also lead to elevated absorption cross sections in the ultraviolet (17); the products are thus vulnerable to photodecomposition. Organic oxidation mechanisms are sensitive to nitrogen oxides, primarily because under highNOx conditions NO and NO2 can react with peroxy radicals (RO2) that might otherwise react with other peroxy radicals (RO2 and HO2). These products can have different vapor pressures, and thus branching ultimately influences SOA formation. Thus, to fully understand SOA production from terpene ozonolysis, experiments must be conducted not only in dark chambers but also in the presence of UV light and variable NOx. Previous work from this laboratory focused on SOA production under changing conditions of UV illumination (11) and NOx concentration (12) (as measured by [VOC]0/ [NOx]0). SOA mass fractions from the ozonolysis of R-pinene decreased in the presence of UV light by a more-or-less uniform 0.03 (a very substantial amount when actual AMFs at low aerosol mass are on the order of 0.05) (11). Increasing NOx concentration (either NO2 or a mixture of NO and NO2) also reduced SOA mass fractions, again by a substantial amount (12). In each case, we concluded that the AMF reductions were a direct result of changes in gas-phase chemistry leading to a more volatile product distribution. The challenge we address here is to extend SOA production experiments to low-COA levels under the necessarily wide range of conditions. With numerous variables to vary, 10.1021/es052203z CCC: $33.50
2006 American Chemical Society Published on Web 04/28/2006
TABLE 1: Experiments Conducted for This Study date
[VOC]0 (ppb)
O3 (ppb)
[NOx]0 (ppb)
[VOC]0/[NOx]0 (ppb C/ppb)
COA (µg/m-3)
final AMF
06/08/05 06/14/05 06/20/05 06/28/05 07/08/05 07/13/05 07/22/05
11.3 6.30 24.5 135 11.5 13.4 43.8
220 280 340 390 290 260 350
26 6.0 5.3 6.3 41 5.5 5.0
4.3 10.5 46 214 2.8 24 88
0.215 2.68 10.7 192 0.707 6.43 46.6
0.0034 0.076 0.078 0.25 0.011 0.085 0.19
including the NOx and UV levels described above, exploring the parameter space of aerosol mass to span the appropriate range of volatilities threatens an overwhelming number of experiments. Consequently, we seek experimental approaches designed to scan a range of aerosol masses during a single experiment. The obvious starting point is to observe SOA production during the early portion of an experiment while the precursor (R-pinene in this case) is being oxidized. For these experiments to succeed, the ozone-terpene reaction must be ratelimiting for SOA formation; if either secondary reactions or nucleation/condensation kinetics delay SOA formation, then measurements will provide only a lower limit of SOA yields (18). However, in many cases this initial reaction is indeed rate-limiting, leaving only the time resolution of the measurements themselves at issue. In recent years, proton-transfer reaction mass spectrometry (PTR-MS) has proven an effective tool for measuring gas-phase concentrations of organic species (19-26). PTRMS has been used to monitor several classes of species similar to those found in terpene ozonolysis experiments, including organic acids (26), aldehydes and ketones (19, 23, 26), and terpenes (24, 25) at sub-parts-per-billion (ppb) levels. PTRMS also provides relatively high time resolution measurements; typical sampling times range from a few tens of seconds to several minutes. In contrast, gas chromatographic methods often require as much as 30 min between samples and therefore do not provide the time resolution afforded by PTR-MS. The high time resolution and sensitivity to organics make PTR-MS a powerful tool in the investigation of terpene ozonolysis, specifically enabling the observation of gas-phase precursor concentrations in real time. This permits direct observations of SOA mass fractions during the initial phase of oxidation experiments. In this study we present PTR-MS measurements of terpene removal for the R-pinene-ozone system under a range of conditions. The experiments presented here include the conditions of experiments presented previouslyswe perform both dark and UV-illuminated experiments over a range of [VOC]0/[NOx]0. Most importantly, the PTR-MS measurements allow us to access the lower end of the aerosol mass distribution and thus provide a much stronger constraint on the low-COA SOA mass fractions in this critical system. As we shall show, our observations are consistent with our earlier data (and other published data), but the new data at low aerosol mass reveal much larger AMFs than extrapolations based on two-product fits to earlier data.
2. Experimental Section Many of the experimental details are described elsewhere (11, 12); here we will present details specific to this study. SOA production experiments are conducted in a 10 m3 Teflon chamber (Welch Fluorocarbon), suspended inside a temperature-controlled room (15-40 °C, held at 22 °C during this study). Ozone and NOx concentrations are measured by gas-phase analyzers for each species (Dasibi 1008-PC and API 200A). Particle concentrations (for ∼10-800 nm diameter
notes NOx, UV UV NO2
particles) are monitored by a scanning mobility particle sizer (SMPS, TSI 3936). For most experiments, a gas chromatograph with a flame ionization detector (GC-FID, Perkin-Elmer AutoSystem XL; J&W Scientific DB-624 capillary column, 30 m × 0.530 mm) coupled to a preconcentrator (Entech 7100A) is available for measuring gas-phase concentrations of organic species. The chamber is also equipped with three banks of UV lights (General Electric model 10526 black lights). The experiments conducted in this study constitute several variations on the reaction between R-pinene and O3. The various experimental conditions are consistent with experiments presented in two previous studies (11, 12) and include the addition of UV radiation, NO2, and NOx. We present experiments from four primary groupings: R-pinene + O3, dark; R-pinene + O3, UV-illuminated; R-pinene + O3 + NO2, dark; R-pinene + O3 + NOx, UV-illuminated. All experiments also include 2-butanol in sufficient quantity to scavenge >95% of OH radicals formed by the ozonolysis reaction (27). The product of the reaction of OH with 2-butanol, 2-butanone, does not form SOA even at high (parts per million) concentrations. Experiments that include NOx contain a mixture of NO and NO2; these experiments are performed in the presence of UV light to regenerate NO via NO2 photolysis. We classify experiments by the initial [VOC]0/[NO2]0 or [VOC]0/[NOx]0 ratio (ppb C/ppb) (1, 12), even in cases where no additional NOx is added to the chamber. Table 1 details the experiments conducted for this study. A commercial PTR-MS (Ionicon Analytik GmbH) was employed to monitor gas-phase concentrations of organic species. PTR-MS is a chemical ionization technique based on the transfer of protons from H3O+ ions to organic analytes (19, 21). The ionization takes place in a drift tube held at ∼2.15 mbar where the analyte molecules are exposed to the H3O+ ions. A full description of PTR-MS construction, operation, and theory is given elsewhere (19-22); the following discussion will focus on aspects of the PTR-MS pertinent to our work. For species with proton affinities greater than that of water (∼697 kJ mol-1) the proton-transfer reaction is often nondissociative (19, 23) (R1), and the analyte is detected at a mass-to-charge ratio of the molar mass plus one.
H3O+ + R f RH+ + H2O
(R1)
However, the proton-transfer reaction causes fragmentation of several types of organic species important to this work, including terpenes. Tani et al. (24, 25) reported that, in addition to the protonated molecular ion (m/z 137), terpenes also produced fragment ions at m/z 67, 81, and 95. Fragment ions at m/z 137 and 81 were used to determine R-pinene concentrations in this study, and we calibrated PTR-MS measurements of R-pinene concentration using GC-FID. Figure 1 shows a time series of a typical experiment (07/ 13/05). Both the SMPS and the PTR-MS R-pinene raw signals have been corrected for perturbations: wall loss for the SMPS and a minor interference from a reaction product for R-pinene. The raw data and the corrections are presented in VOL. 40, NO. 11, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
3537
the ξ in the context of the partitioning theory of Pankow (28, 33, 34), which allows aerosol production to be expressed as a function of the total organic aerosol concentration (28). n
∑
ξ)
i)1
( ) Ri
1+
FIGURE 1. Time series of a typical experiment (experiment 07/13/ 05). The corrected aerosol mass concentration is determined by assuming that a size-independent first-order loss rate describes wall losses and by assuming spherical particles with a density of 1.0 g cm-3. Injection of ozone (t ≈ -0.25 h) results in r-pinene consumption and the generation of SOA. the Supporting Information. Injection of O3 into the chamber (t ≈ -0.25 h) leads to a decrease in R-pinene concentration and the nucleation of aerosol. Because our goal with the PTR-MS sampling is to investigate gas-phase concentrations of organic species, a Teflon filter is placed between the smog chamber and the PTR-MS for all experiments. The filter prevents particles from entering the PTR-MS and possibly altering the instrument response. Several tests were conducted to verify that the filter does not affect the measured concentration of volatile species. To find partitioning based on eq 2 we need in theory to measure only the mass of precursor consumed (∆CROG with the PTR-MS and GC-FID) and the mass of aerosol produced (COA with the SMPS), noting that we normalize to an assumed aerosol density of 1 g cm-3 pending future density constraints. PTR-MS CROG data are smoothed with a spline to the 3 min sampling interval of the SMPS, resulting in a precision of 0.2 µg m-3 + 2%. The accuracy is estimated at 5%. The SMPS volume mode is typically 250 nm at the end of an experiment (see, for example, Figure 3 in Presto et al. (11)), with a precision of 0.01 µg m-3 + 1% and again an accuracy of 5%. Systematic factors include chamber cleaning (overnight flushing with clean air, UV, and high ozone), chamber mixing (on the order of a 3 min time scale), wall effects (small during the rise period used here), nucleation and condensation kinetics, and chemistry. Overall, we estimate the AMF measurements to have a combined precision and accuracy of approximately 10%, with an additional 0.01 added in quadrature, dominated by the systematics and not instrumental precision. The precise concordance between the PTR-MS and the GC-FID data in Figure 1 is encouraging. It also suggests that interpolations of GC data may be valid for some time-resolved AMF data. However, this can only be taken so far; the sampling interval for the GC preconcentrator is many minutes, and the R-pinene decay curve is not perfectly exponential. Consequently, interpolation of GC data to the earliest times (just after t ) 0 in Figure 1) certainly involves substantial errors. This is just where the widest range and lowest values of COA are observed.
3. Secondary Organic Aerosol Partitioning Previous studies have characterized SOA production by defining the aerosol mass fraction, ξ, accounting for aggregate condensable mass rather than focusing on individual product yields (3, 4). Chamber studies (3, 4, 6, 28-32) have considered 3538
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 40, NO. 11, 2006
(1)
C/i
∆M0
Ri is the (mass-based) yield of product i, and C/i (µg m-3) is the effective gas-phase saturation concentration of species (or group) i. Previous studies found that a two-product model, with parameters R1, R2, C/1, and C/2, adequately describes chamber-based SOA measurements at a single temperature for both biogenic (4) and anthropogenic (28) precursors. The two products used in the model are intended as lumped species; one set of parameters (R1 and C/1) describes the ensemble of heavy (relatively nonvolatile) species, and the other set of parameters (R2 and C/2) describes the ensemble of light (relatively volatile) species. Completely volatile species are not treated. Our analysis follows partitioning theory, but we explicitly separate the production of products from the partitioning itself, deriving two terms for each semivolatile “product” i: the mass yield of that product, Ri, and the partitioning coefficient for that product, ξi, which is the condensed-phase mass fraction of that species or group. The critical parameters are the saturation concentration of species i, C/i , in µg m-3, and the total concentration of aerosol organic carbon, COA ) ∆M0. (We use COA to emphasize that it is the total organic aerosol mass concentration that appears in the partitioning equation, regardless of its source, and that the partitioning coefficient ξi is unitless).
ξi )
( ) 1
1+
C/i
n
ξ)
∑Rξ
i i
(2)
i)1
COA
While C* can in theory be derived from the pure saturation vapor pressure of an identified compound, this is essentially irrelevant; not only are the species used in AMF fits composites of numerous products, but neither their vapor pressures nor their activity coefficients in real atmospheric organic aerosols are known. The quantity C* itself is the empirically determined value in these studies. An important caveat is that all studies based on SMPS data measure aerosol volume as opposed to aerosol mass (particle sphericity is a safe assumption). The particle density is thus an important unknown. Various values have been used in the literature, which introduces a potentially vexing source of confusion. Rather than assuming a density, we shall present a “normalized AMF” (ξ′ ) ξ/F or, alternatively, the AMF assuming F ) 1); this is consistent with most previous work (4, 12, 35). Attempts to constrain the aerosol density under different conditions will lead to progressive refinements in F, and the ultimate AMF can be calculated with ξ ) F ξ′. The formulation for ξi facilitates estimation of product partitioning by inspection, regardless of the total concentration of product i or the concentrations of other species. When the total concentration of organic aerosol, COA, is equal to C/i , product i partitions evenly between the vapor and the condensed phases. Likewise, i resides ∼10% in the condensed phase when COA ) C/i /10 and ∼90% in the condensed phase when COA ) C/i × 10. Stanier et al. (36) present a broad framework for fitting AMF data to multiproduct models. As noted by the authors, the number of products included in the model is a function of the model applicationsfewer products can be used to
attain greater computational efficiency, or a larger number of products can be applied for improved accuracy. However, to treat the production of semivolatile products from R-pinene + ozone consistently in the context of other semivolatile compounds present in the atmosphere, we condider SOA production with a logarithmically spaced basis set, {C/i }, that extends well beyond the atmospheric range of COA at both the low and the high volatility limits (37). Fitting only the Ri with a fixed C* basis set is relatively straightforward (36), but it is only well-conditioned when data exist with COA within an order of magnitude of a given C*. In a separate work, Pathak et al. (38) present AMF data and optimal parameters for atmospheric modeling from multiple terpenes. Those results should be used for modeling, but the parameters themselves are unimportant; what is supremely important is the agreement among various experimental data sets, the degree of agreement among parametrizations that are interpolating experimental data, and the divergence of those parametrizations when they are used to extrapolate beyond the range of experimental data. The wide range of C/i used in the basis set provides for high accuracy and additionally allows for the aerosol to be parsed into a collection of theoretical products with saturation concentrations similar to those generated by the ozonolysis reaction (37). Aerosol organic carbon is typically 5 µg m-3 in many environments, though it can occasionally rise to 50 µg m-3 or more in highly polluted areas. Thus, to capture the full range of products contributing to atmospheric organic aerosol, we must consider C/i between ∼0.05 and ∼500 µg m-3. Additional products, with larger C/i (103-104 µg m-3), are required if we wish to ultimately consider the effects of further oxidation of product species, including aerosol aging.
4. Fitting the Basis Set We can fit data with eq 2 using a basis set, in which a collection {C/i } is specified and the {Ri} are free parameters (37), or using a multiproduct model where {C/i ,Ri} are free parameters (28). Specifically, a two-product model has four free parameters, as does a basis set with C* ) {0.1, 1, 10, 100} µg m-3. However, these two cases are not identical; the parameters in the two-product model are far from orthogonal and show very high covariance after fitting. This means that small changes to the data setsfor instance, adding new datas can result in very large changes in the parameters and yet little change in the actual AMF curve. Comparing parameters from fits is always dangerous, but in two-product models it is impossible. The parameters in the basis-set model are far less covariant, as they have no freedom to move in volatility space; we thus expect much less variation with added data, unless the new data contain substantial new information content. Despite this, neither approach should be trusted in extrapolation, because neither is based on a priori knowledge of volatility behavior outside of the range of the data; indeed the major point of this paper is that data must span the range of ambient COA to constrain ambient SOA production. Another advantage of the basis-set formalism is that parameter estimations are intuitive and well-conditioned. For the reasons mentioned above no one can look at data to be fit with a two-product model and estimate the four parameters; however, one can easily estimate the basis-set parameters by visual inspection of data, as we shall show. There is relatively little cross-talk between neighboring bins when they are separated by an order of magnitude, as in this case. For a given bin with a saturation concentration equal to a given COA (for example, 10 µg in Figure 2), 90% of the next-lower-volatility bin is in the condensed phase, and only 10% of the next-higher-volatility bin is in the condensed phase. Thus, for the purposes of illustration only one bin at a time has substantially semivolatile behavior. This separation
FIGURE 2. Fitting SOA data with a basis set (see text). Each bin in turn has 50/50 partitioning for COA ) Ci/. That bin is shown in green and white on top of lower-volatility material in gray for each successive Ci/. The green portion represents the condensed phase, with the y-value at the top of the green bar showing the approximate total aerosol mass fraction in each case (normalized to G ) 1). Almost all of the lower-volatility material is in the condensed phase for each case, as shown by the accumulating solid gray bars with total height A ) ∑ ri; Ci/ < COA. Parameter estimation proceeds from left to right, as indicated by the arrows; final parameters are determined through standard nonlinear least-squares fitting. and the intuitive nature of the log spacing motivates our choice for {C/i }. Fitting data to a basis-set description should be thought of as a progressive exercise, depicted in Figure 2 for data that will be discussed in detail in the next section. The data presented here span the range 0.02-300 µg m-3, and we shall fit them to mass yields {Ri} for a basis set of {0.01, 0.1, 1, 10, 100, 1000} µg m-3. Yield parameters are estimated starting with low-COA data at 0.01 µg m-3, where that basis material will be 50/50 partitioned (depicted with a green bar within a clear box; the total clear box height is R1 e 0.005). This material is almost completely condensed at 0.1 µg m-3 (in reality the partitioning is 90%), and there is little if any sign in the data for additional material with C* ) 0.01 µg m-3. Next, the data at 1 µg m-3 are treated by first assuming that the less volatile material is completely in the condensed phase (R1 + R2 is now shown as a solid gray bar). The residual between the data and R1 + R2 must be 50% of R3, which is thus approximately 0.05. This process continues up to the full range of the data, as shown. In reality there is a 10% coupling between each neighboring bin that needs to be formally treated with regressive optimization, but the simple procedure described here gives a good and intuitive first guess. That coupling can be a serious source of error for small-yield bins, such as the first two in this example, where residual partitioning from higher-volatility bins with much higher mass yields can overwhelm the native signal. There are two classes of uncertainty for this approach. First, for a given basis set, the parameter uncertainties and confidence intervals are easily determined given standard nonlinear least-squares formalism. In this case the optimal parameters are {0.004 ( 0.003; 0 ( 0.001; 0.051 ( 0.007; 0.09 ( 0.01; 0.12 ( 0.02; 0.18 ( 0.06}, and the 95% confidence interval is shown in subsequent figures with dashed lines. The parameter uncertainties themselves are not sufficient to estimate the uncertainty in the aerosol mass fraction because of parameter covariance. The formal confidence interval can meet this need, but a more subjective assessment of the uncertainty is that ξ is known to within approximately 10% over the range 0.5-500 µg m-3. A comprehensive set of VOL. 40, NO. 11, 2006 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
3539
parameters for R-pinene and other terpenes + ozone, including temperature and composition effects, is presented in a separate paper (38). The second uncertainty concerns the basis set itself and extrapolation. In this example the data could be fit with equal fidelity with the yield in either of the first two bins set to zero. While a slight shift in the other parameters would propagate through the basis set, the major difference in prediction would be found for extrapolations with COA < 0.1 µg m-3. Fortunately in this case that is below the range of substantial interest in the atmosphere. The same issue applies to the highestvolatility bin. In general, the extrapolation error can be bounded on the low-volatility end by selecting low-volatility end members to the basis set that spans the extremes as in this example. The higher-volatility end is more difficult to constrain with anything other than a maximum mass yield that must not be too much greater than the mass of the precursor. A final major advantage to the basis-set representation concerns mixtures with other semivolatile material (37). Within the assumption that the organic compounds in these experiments form ideal solutions with ambient organic aerosol, when data are fit to a common basis set the entire collection of semivolatile material can be represented with a single basis set {C/i } and an ensemble of yields {Ri} for each SOA reaction and fluxes {φi} for each semivolatile primary source (39). Even chemical transformations among the various bins driven by homogeneous and heterogeneous reactions can be readily treated (37). As a result a very rich array of semivolatile organic aerosol sources can be modeled without undue computational burdens to large chemical transport models.
5. Results For each experiment, the normalized AMF is calculated by two separate methods. The final AMF is determined in the same manner as previous work from this laboratory (11); this value is traditionally reported for chamber-based experiments (3, 4, 35). We also use the PTR-MS data to calculate a time-dependent AMF. The time-dependent AMF allows us both to investigate how aerosol yield evolves with time and to explore aerosol production at conditions of low total aerosol concentration. The final AMF results are listed in table 1. The AMF values are consistent with previous work from this group (12) as well as other sources (4, 35). The AMF decreases (relative to dark, NOx-free experiments) in the presence of high concentrations of NO2, NOx, or UV light. This is also consistent with previous results (11, 12). As shown in Figure 3a for experiment 07/13/05, the AMF rises as R-pinene is consumed. The aerosol mass peaks approximately 1 h after the initiation of aerosol formation; this coincides with the R-pinene concentration falling to approximately 15% (∼e-2) of its initial concentration. After peaking, the AMF remains constant. The PTR-MS data also allow for the construction of the classic AMF versus organic aerosol concentration plot (28) for each experiment. Figures 3b-d show normalized AMF vs COA for darkened low-NOx, high-NOx, and UV-illuminated low-NOx conditions, respectively. The figures are plotted on a logarithmic scale to emphasize the atmospherically important region at low COA (
