Environ. Sci. Technol. 1987,21, 1224-1231
Environment: Toronto, Ontario, 1985; Report ARB-14185-AQM. Bardswick, W. S.; Chan, W. H.; Orr, D. B. Water, Air, Soil Pollut. 1986, 30, 981-990. Handbook of Analytical Methods for Environmental Samples; Ontario Ministry of the Environment, Laboratory Services and Applied Research Branch Toronto, Ontario, 1983; Vol. 1 and 2. Anlauf, K. G.; Bottenheim, J. W.; Brice, K. A.; Wiebe, H. A. Water, Air, Soil Pollut. 1986, 30, 153-160. Chan, W. H.; Chung, D. H. S. Atmos. Environ. 1986,20,
1397-1402. Summers, P. W. In Precipitation Scavenging;Semonin, R. G., Beadle, R. W., Eds.; Technical Information Centre, ERDA Springfield, VA, 1977; pp 88-94.
(9) Finlayson-Pitts,B. J.; Pitts, J. N., Jr. Atmospheric Chemistry; Wiley-Interscience: New York, 1986. (10) Munger, J. W.; Eisenreich, S. J. Environ. Sci. Technol. 1983, 17, 32A-42A. (11) Keene, W. C.; Galloway, J. N. Atmos. Environ. 1984, 18, 2491-2497. (12) Guiang, S. F.; Krupa, S. V.; Pratt, G. C. Atmos. Environ. 1984,18, 1677-1682. (13) Summers,P. W.; Barrie, L. A. Water,Air, Soil PoElut. 1986, 31, 523-535. (14) Topol, L. E. Atmos. Enuiron. 1986, 20, 347-356. (15) Dasen, J. M. Atmos. Enuiron. 1987, 21, 137-141.
Received for review March 21,1986. Revised manuscript received June 26, 1987. Accepted August 25, 1987.
Aerosol Formation and Growth in Atmospheric Aromatic Hydrocarbon Photooxidation Jennlfer E. Stern, Rlchard C. Flagan, Danlel Grosjean, and John H. Selnfeld" Department of Chemical Engineering, California Institute of Technology, Pasadena, California 9 1125
An experimental study of aerosol formation in aromatic hydrocarbon/NO, systems has been conducted in an outdoor smog chamber. Aerosol size distributions were measured as a function of time in toluene, m-xylene, ethylbenzene, or 1,3,5-trimethylbenzene photooxidations to determine the rates of new particle formation and the effects of initial particles on aerosol formation and growth. Aerosol yields from the aromatic gas-phase photooxidations were found to be approximately 2-5% by mass of the starting aromatic species. Simulations of the aerosol behavior in these experiments have been carried out using an integral model that includes a vapor source, homogeneous nucleation, condensational growth, and particle loss by deposition. Predictions from the model are in relatively good agreement with the experimental observations. Results indicate that the nucleation mechanism in these systems is still not completely understood.
Introduction Aromatic hydrocarbons are important components of anthropogenic atmospheric emissions (1). Earlier studies have shown that aromatics are important precursors of organic aerosols in the atmosphere (2-7). Organic aerosol formation occurs via gas-phase degradation routes that produce a condensable vapor that is converted to aerosol via homogeneous nucleation and heterogeneous condensation. These mechanisms have not been studied previously in the aromatic hydrocarbon/NO, system. In a system with a source of condensable vapor and no initial particles, the partial pressure in the vapor phase can build up to supersaturated levels, at which point homogeneous nucleation can occur and aerosol particles are formed. Subsequent condensation onto these particles will help relieve the vapor-phase supersaturation, causing nucleation eventually to cease. After this point, subsequent gas-to-particle conversion will occur by condensation. However, in the presence of initial particles, condensational growth can begin as soon as a low vapor pressure species is generated. With a sufficiently large number of initial particles it is possible that the vapor will never achieve the supersaturation needed for nucleation, as condensational growth will always be the dominant mechanism for gasto-particle conversion. We have studied experimentally in an outdoor smog chamber the effect of preexisting 1224
Environ. Sci. Technol., Vol. 21,No. 12, 1987
particles on the subsequent aerosol behavior in aromatic photooxidations in order to observe the nucleation suppression occurring with varying levels of initial particles. The analysis of the experimental data tests our ability to simulate simultaneous nucleation and condensation in such photooxidation systems. The principal unknown quantity in the description of aerosol formation and growth is the rate of homogeneous nucleation of the aerosol precursors. Indeed, predicting the rate of homogeneous nucleation of a substance is one of the long-standing challenges in condensed matter physics. The two most popular nucleation theories, the so-called classical theory and the Lothe-Pound theory, differ in their predicted rates by something approaching 20 orders of magnitude (8). Thus, the data obtained from the smog chamber experiments on the rates of particle formation and growth will aid in assessing our ability to describe nucleation of organic aerosol constituents. Moreover, studying how those rates are modified in the presence of foreign particles provides additional measures of the nucleation rate as the condensing species concentration is altered by condensation onto the preexisting particles. In the next section we present a description of our experimental facility. Next we discuss the inversion of the raw aerosol data, followed by a summary of the measured aerosol yields. Finally, we describe the aerosol model we have used and present the results from the simulations.
Experimental Description The essential nature of the experimental system is described in Leone et al. (9). The Teflon chambers used for this study were constructed of 10 Teflon panels of 1.2 m X 10 m each. Gas-phase sampling was done from Teflon tubing that extended approximately 30 cm into the chamber through Teflon ports. Aerosol-phase sampling was carried out through separate copper lines extending about 15 cm into the chamber. These copper lines minimized depositional losses of the aerosol. Most of these experiments were conducted in dualchamber mode. In dual-chamber mode, a poly(viny1 chloride) (PVC) pipe was placed across the chamber at its midpoint to divide the reactor into two equal sides. Air leaks across the divider were undetectable. The volume of each side of the divided chamber, as measured by the injection of known amounts of NO or NOz,was approxi-
0013-936X/87/0921-1224$01.50/0
0 1987 American Chemical Society
mately 25 m3, When operation was in the dual-chamber mode, an actuator valve was used to automatically switch sampling from one side to the other. In most experiments, the actuator switched sides every 10 min. In order to ensure that the lines had been flushed, sufficient time (at least 2 min) was allowed between the time the actuator switched and the time measurements were taken. Following the gas-phase injections, initial aerosol was optionally added to the entire chamber or to one side of the divided chamber. The seed particles were ammonium sulfate aerosol, generated with a stainless steel collision atomizer from a solution of 2 g L-l (NH4)2S04in water. The aerosol was passed through a 85Krdecharger before injection. To minimize depositional losses before beginning the experiment, the seed aerosol was usually the last species injected. The gas-phase measurements obtained were 03,NO, toluene, ethylbenzene, m-xylene, 1,3,5-trimethylbenzene, chamber temperature, sample line temperature, relative humidity, and total solar radiation. Since this work has focused on the aerosol measurements, the details of the gas-phase measurements will not be reiterated here; they are similar to those reported by Leone et al. (9). The gas-phase data and the analysis of these data are reported elsewhere (IO). . Five aerosol instruments were available for monitoring the particulate-phase behavior in the system. In singlechamber mode, two TSI Model 3030 electrical aerosol analyzers (EAA's) sampled continuously from the chamber contents. These instruments were controlled by a PDP 11-03 Series minicomputer housed in the adjacent laboratory. The computer also collected data from the EAA's in cycles of about 3 min each. The EAA's nominally measured particles from 0.01 to 1.0 ,um in diameter. A Royco Model 226 laser optical particle counter (OPC), housed inside the laboratory, was used to monitor aerosol growth in the O.l-pm diameter and larger size range. Since the OPC can only operate properly with number concentrations below 1500 ~ m -a~dilution , system was built to deliver approximately 100 times dilution of the sample before it entered the OPC. The OPC printed out readings every 2 min. Two Environment One Model Rich 100 condensation nuclei counters (CNC's) were available for determining overall number concentration in the chamber. In dual-chamber mode the aerosol instrument configuration was slightly different. The EAA's were still controlled and monitored by the laboratory computer, however each EAA sampled from a different side of the chamber. The sample lines were switched manually for several cycles at intervals during each experiment to determine any effects of instrument bias. The OPC and one CNC were connected to the center switching valve, thus sampling only one side continuously. A full schematic of the experimental setup is shown in Figure 1. After each experiment, the chamber was filled with purified air and "baked out" in sunlight for an entire day. Between some experiments the "bake out" was repeated. It was found that continual flushing of the chamber contents with clean air helped minimize ozone levels during "bake out". Using this procedure, we encountered no problems in reproducing experimental results from one day to the next. A total of 40 smog chamber experiments were carried out between June 1985 and May 1986. Thirty experiments were carried out in dual-chamber mode, for a total of 70 sets of initial data. Of the 40 experiments, there were 17 toluene photooxidations, 10 of which were in single-chamber mode, 7 m-xylene photooxidations, 6 1,3,5-trimethyl-
r-I-cJ 0 /"I*,
INSIDE LAB-
-OUTSIDE
_-__
G o i phose mmplei lhiDu¶h Teflon l i n e l A l l O I O l I(l",PllS
l n r o q h ropp*c liml
LA8
Figure 1. Schematic of the outdoor smog chamber facility used for the experiments.
benzene photooxidations, 5 ethylbenzene photooxidations, and 5 experiments in which different aromatics were injected into each side of the chamber. The experiments were between 2 and 6 h in length; most experiments were conducted for 4 h. The initial HC/NO, ratios varied between 6.4 and 44.2 ppmC/ppm. The average temperature ranged between 291 and 322 K. Initial aerosol loadings ranged from no initial particles to 1.7 X lo4~ m -although ~, usually the initial aerosol concentration was below 8 X lo3 ~ m - ~Table . I contains a complete listing of the experiments carried out for this study.
Aerosol Data Inversion We address the issue of data inversion here because it is one that often plays a pivotal role in the analysis of aerosol data. The EAA and the OPC are the main sources of information about the aerosol size distribution in our system. However, the EAA and the OPC do not respond over the same size range of particles, and both instruments exhibit nonideal behavior. The EAA is optimal for sensing particles from 0.02 to 0.3 pm in diameter; the OPC is optimal for particles above 0.17 ,urn. The size range of overlap for the reliable signals from the two instruments is therefore quite narrow. Hence, there are few data sets in any experiment where both instruments can be compared, and some assumptions have to be made as to how to derive the most information from the available data. Several algorithms exist for the inversion of data from indirect measurement techniques such as the aerosol measurements (11-19). In our case, we seek the aerosol size distribution that gives rise to the observed EAA or OPC readings. For inverting large streams of data (approximately 100 data sets for each instrument for each experiment) a quickly converging, robust, iterative scheme was needed. For the inversion of our data, we have employed a routine based on Twomey's algorithm (15). The Twomey algorithm, however, can give rise to oscillatory results that still satisfy the mathematics of the problem; therefore, a smoothing step was added to the algorithm (20). The resulting so-called smoothed-Twomey routine allows the input of experimental tolerances for each channel of the instrument and attempts to choose the smoothest result that fits the observed signals within these input tolerances. Kernel function data were obtained for the EAA from Richards (21) and for the OPC from a calibration done by Crump with the present instrument (22). Although these inversions gave satisfactory results for the EAA and the OPC individually, it was necessary to combine our observations to produce one set of number distributions for each experiment. We determined that this could only be done by combining the data from both Environ. Sci. Technol., Vol. 21, No. 12, 1987
1225
Table I expt 1
2 3 4 5 6 7 8 9 10 llAd 11B 12A 12B 13A 13B 14A 14B 15A 15B 16A 16B 17A 17B 18A 18B 19A 19B 20A 20B 21A 21B 22A 22B 23A 23B 24A 24B 25A 25B 26A 26B 27A 27B 28A 28B 29A 29B 30A 30B 3 1A 31B 32A 32B 33A 33B 34A 34B 35A 35B 36A 36B 37A 37B 38A 38B 39A 39B 40A 40B
aromatica TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL TOL XYL XYL XYL XYL XYL XYL XYL XYL XYL XYL XYL XYL XYL TMB XYL EB TMB EB XYL XYL TMB TMB TMB TMB TMB TMB TMB
TMB TMB TMB TMB TMB EB EB EB EB EB EB EB EB EB EB TMB XYL TOL EB
[HC/NO,Io, PPmC/PPm 12.9 14.2 17.2 13.9 10.1 21.3 14.3 16.3 17.7 12.2 21.4 17.1 15.6 15.6 14.1 14.1 33.3 33.3 28.5 29.5 29.9 28.7 30.4 14.4 9.2 8.5 6.4 6.5 7.9 8.1 8.2 18.3 17.9 17.8 21.4 21.8 33.2 33.2 31.4 33.1 28.0 44.2 17.8 17.8 10.7 10.7 8.7 8.6 11.6 20.3 16.8 17.1 20.3 21.1 14.9 15.1 10.4 10.5 14.9 15.7 16.7 38.8 34.4 35.4 32.7 32.4 28.1 15.9 20.7 20.5
T, K
Nbb io3
307 315 311 318 313 309 313 309 309 312 319
7.0 1.0 7.4 4.7 0.0 8.0 0.0
322 315 312 315 312 315 310 305 308 291 294 296 297
0.0
6.8 3.0 0.0 0.0 3.6 3.4 6.7 0.0 2.0 0.0 16.9 0.0 5.9 0.0 2.0 2.1
6.3 0.0 1.6 0.0 2.1 0.25 0.5 0.5 3.0 0.0 7.4 0.25 e
Nul,: IO3 12.0 6.1 6.3 13.2 11.7 4.0 16.5 10.8 6.4 5.2 9.5 20.0 29.0
-
22.3 23.7 15.1 -
-
11.5 3.4
-
4.6 2.0 4.7 5.7 14.7
e
306 299 305 304 302 307 309 310 311 309 309 304 311 310 307 308
e e e e 5.2 0.4 2.0 0.4 2.2 0.1 0.7 0.2 3.4 0.8 5.0 0.4 6.6 0.2 e e 5.8 1.1 0.7 0.7 3.0 0.2 5.4 0.0 1.3 0.8 0.4 1.0
11.5 2.7 4.0
-
3.2 2.3 5.9
-
5.5 2.1
-
6.8 -
7.6 10.2 17.0
-
5.4 2.8
18.0 5.6 -
‘TOL = toluene, XYL = rn-xylene, TMB = 1,3,5-trimethylbenzene, and E B = ethylbenzene. Initial number. Maximum number observed. (-) indicates no observed nucleation. dA and B indicate the two sides of the divided chamber. “No aerosol measurements taken. 1226
Environ. Sci. Technol., Vol. 21, No. 12, 1987
instruments and inverting all of the experimental observations simultaneously. Although the multiinstrument aerosol inversion has not been addressed previously, success with the smoothed-Twomey algorithm for the single-instrument case suggested that perhaps the multiinstrument problem could be solved with this method. Simultaneous inversion of the OPC and EAA data produced size distributions at approximately 3-min intervals throughout each run. From these, one can obtain total number, total volume, and average particle size as a function of time for each experiment.
Aerosol Yields The first aerosol-related question we address is that of the quantity of the initial hydrocarbon that is converted to aerosol-phase precursor. This is an important question, for it provides information not only to help determine gas-phase reaction pathways but also to assess in an overall way the aerosol-forming potential of the organic species. The yields in these experiments were calculated on a mass basis, since molecular composition of the condensing species was unknown, making a molar yield impossible to obtain. The inverted aerosol volume was converted to mass by assuming that the aerosol had a density of 1 g ~m-~ Yields . were calculated at the point where the aerosol volume profile reaches its maximum to minimize the effect of wall loss. Finally, we note that no correction was made for the mass of initial aerosol in the system, as it represents a very minor fraction of the total aerosol. The average aerosol yields by mass for each aromatic were as follows: toluene, 4.8%; m-xylene, 3.5%; ethylbenzene, 1.9%; 1,3,5-trimethylbenzene, 2.4%. We note that these yields are somewhat dependent on the HC/NO, ratio for each experiment since there is a correlation between the system reactivity, which is dependent on the HC/NO, ratio, and the percent conversion to the aerosol phase. Previous studies of the gas-phase mechanisms of aromatic hydrocarbons have included aerosol yields based on percent of reacted carbon in order to complete a mass balance on the carbon loading in the system (5,9,23-27). These researchers have found aerosol yields in the range of 1-6% for the toluene and m-xylene systems. Although our yield determination was calculated on a mass basis instead of an elemental carbon basis, our values are generally in good agreement with those found in previous studies. Theoretical Analysis of Aerosol Formation and Growth The experiments we have carried out have generated a large data base with information on particle formation and growth in aromatic photooxidations. By applying aerosol models to these data, we can learn about the microphysics of aerosol dynamics in such systems. The model we have used for this analysis of the system is an integral aerosol model called the “SNM model”, because it describes the saturation ratio in the vapor phase, number concentration, and mass concentration of aerosol (28, 29). It treats the aerosol distribution as two monodisperse modes: a primary mode that contains the seed aerosol and a secondary mode created by homogeneous nucleation from the vapor phase. The processes included in this model are a source of condensable vapor, homogeneous nucleation, condensational growth, and particle loss by deposition, The number concentrations in our experiments were low enough so that the amount of coagulation is insignificant; for example, for a system with 10OOO cm-3 particles of 0.1-pm diameter, the characteristic time for coagulation is approximately 30 h. The model
consists of balances on mass and number in the two aerosol modes and a mass balance on the vapor phase: vapor concentration cW,/dt = RG - g,Rj - R,1 - Rc2 primary aerosol
(1)
secondary aerosol
(4) where RG represents the source rate of condensable vapor, RJ represents the nucleation rate, and Rcl and Rc2are the rates of condensation onto the two modes of aerosol, all in number cm-3 s-l. p1 and p2 are the deposition coefficients of the two modes in s-l, and ml is the monomer mass in grams. I t is assumed that the secondary aerosol forms at the thermodynamically determined critical cluster size g, and grows rapidly to an assumed cluster size g,. The characteristic time for condensational growth of freshly nucleating particles in our system is approximately 1ms. Since this free-molecular regime growth is fast, the assumption that particles form a t size g, instead of g, is reasonable. The following parameters are needed for use in this model: physical property data (vapor pressure, surface tension, density, and molecular weight), the source rate of condensable vapor, deposition parameters (electric field, particle charge, and coefficient of eddy diffusivity), and nucleation parameters (nucleation rate expression, gJ. For nucleation and condensation modeling we need to know the vapor pressure, surface tension, density, and molecular weight of the condensable species. Moreover, we need to account for the temperature dependence of these properties since the average temperature of the outdoor smog chamber varied from experiment to experiment. However, the molecular composition of the condensing species in our system is unknown, and therefore we cannot easily predict the physical property data needed. In addition to the aerosol physical property data, the gas-phase source rate of condensable vapor is the other major unknown in our system. Ideally one would predict this rate from knowledge of the detailed gas-phase kinetics. Such knowledge is not yet available for these systems. In the absence of this information, the source rate can be estimated from the overall aerosol growth. To determine the rate of generation of condensable species, the measured aerosol volume profile was interpolated over 0.5-h intervals. The growth rate was determined in discrete intervals from each data point to the next. Thus, the source rate used in the model was variable, increasing or decreasing in steps. For intervals where the aerosol volume decreased due to depositional loss, the source rate was taken to be zero. The rate of deposition on the wall of the chamber is the result of diffusional and gravitational deposition as described in the theory of Crump and Seinfeld (30). In addition, if the particles are charged, electrostatic deposition is important, since Teflon film chambers tend to develop an electric field. McMurry and Rader (31)have shown that electrostatic deposition in a Teflon chamber is especially significant for particles between 0.1 and 1.0 bm, the size range of interest in our experiments. We assumed the turbulent mixing coefficient in the CrumpSeinfeld theory to be k, = 0.1 s-l and the electric field on the Teflon to be 40 V cm-l, approximately the values found
by McMurry and Rader in their Teflon chambers. If the aerosol particles are singly charged, the theory predicts a deposition coefficient, p, on the order of s-l, which is dependent on particle size. The predicted number concentrations are not particularly sensitive to the choice of deposition parameters, since nucleation occurs over an extremely short time period, during which depositional losses are insignificant. We chose the deposition parameters to match the removal observed during the period of steady condensational growth that follows nucleation. The parameters reported here predict depositional losses close to those observed experimentally. We have assumed initially that each aromatic starting species will produce one condensable species. It is believed that the aerosol precursors come from the ring-preserving reaction pathway of the aromatic photooxidation mechanism. Therefore, we have assumed that the condensable species for all of the aromatics have a molecular weight of 150, representative of a nitrogenated or oxygenated ring compound. We assumed a surface tension of 30 dyn cm-l, independent of temperature, a value typical of organic liquids. The temperature effects were included by variation of the vapor pressure of the condensing species. The Clausius-Clapeyron equation states that for an ideal gas, vapor pressure and temperature are related by In psat= A / T + B (6) where A and B are species-dependent constants. Since our experiments were carried out over a range of temperatures, we assume that this functional form describes the variation in vapor pressure of the condensable species between experiments. Because the molecular composition of the aerosol species was not known, it was necessary to estimate the vapor pressure for the condensed species from the experimental data. From our observations of nucleation in systems with few or no initial particles, it is possible to determine the vapor pressure required to fit the observed number of particles resulting from the nucleation event. Doing so for each relevant experiment gave a collection of (psat,T) points for each aromatic; from these we could determine a best fit of the vapor pressure for each system using eq 6. The vapor pressure estimates are given in Figure 2 for toluene, m-xylene,and 173,5-trimethylbenzene. The ethylbenzene system did not generate sufficient aerosol, and the temperature range of those experiments was narrow enough that there were not sufficient data to determine a psat/Trelationship. Since the aromatic decay in the ethylbenzene system most closely resembled that of toluene, the toluene vapor pressures have been used provisionally for ethylbenzene. 1,3,5-Trimethylbenzene data were taken over a limited temperature range, so a constant average vapor pressure was used for data analysis of this species. Although for experimental purposes it was necessary to employ initial aromatic concentrations of about 1 ppm, ambient aromatic levels are on the order of 50 ppb (32). If the aerosol yield for toluene, for example, is about 3% as we have found, there should be approximately 1.5 ppb of gas-phase toluene aerosol precursors produced in the atmosphere. The vapor pressure of the toluene aerosol constituents that is consistent with the observed rate of nucleation is approximately dyn cm-2,which is about 0.01 ppb. This level yields a saturation ratio of about 150, which is sufficiently high for gas-to-particle conversion to occur under atmospheric conditions. Finally, we need to consider the nucleation rate expression and the size of the nucleating particles, The nucleating cluster size in the SNM model g, was assumed Environ. Sci. Technot., Vol. 2 1 , No. 12, 1987
1227
0
0
/
,--
N
1
/
1 (K-') N,, ooserved ( x 1003
j
