Ultrafine Particles Near a Roadway Intersection - American Chemical

Sep 15, 2010 - Intersection: Origin and. Apportionment of Fast Changes in. Concentration. JOSEPH P. KLEMS,. M. ROSS PENNINGTON,. CHRISTOPHER A...
0 downloads 0 Views 3MB Size
Environ. Sci. Technol. 2010, 44, 7903–7907

Ultrafine Particles Near a Roadway Intersection: Origin and Apportionment of Fast Changes in Concentration JOSEPH P. KLEMS, M. ROSS PENNINGTON, CHRISTOPHER A. ZORDAN, AND MURRAY V. JOHNSTON* Department of Chemistry and Biochemistry, University of Delaware, Newark, Delaware 19716

Received June 14, 2010. Revised manuscript received August 26, 2010. Accepted September 7, 2010.

A wavelet-based algorithm was implemented to separate the high frequency portion of ambient nanoparticle measurements taken during the summer and winter of 2009 in Wilmington, Delaware. These measurements included both number concentration and size distributions recorded once every second by a condensation particle counter (CPC) and a fast mobility particle sizer (FMPS). The high frequency portion of the signal, consisting of a series of abrupt spikes in number concentration that varied in length from a few seconds to tens of seconds, accounted for 6-35% of the daily ambient number concentration with hourly contributions sometimes greater than 50%. When the data were weighted by particle volume, this portion of the signal contributed an average of 20% to the daily PM0.1 concentration. Particle concentration spikes were preferentially observed from locations surrounding the measurement site where motor vehicles accelerate after a red traffic light turns green. As the distance or transit time from emission to sampling increased, the size distribution shifted to larger particle diameters.

Introduction Ambient nanoparticles have many sources ranging from atmospheric processing of gases such as SO2, NOx, and NH3 to direct emission from industrial stacks and the tailpipes of motor vehicles (1-3). While the exact mechanism of action is still unclear, these particles have been linked to increased incidences of hospital visits and negative cardio-pulmonary health outcomes (4, 5). Particles associated with motor vehicle operation including tailpipe emissions, road dust, and tire wear are thought to constitute anywhere from 20 to 76% of ambient PM2.5 in urban areas (6), with tailpipe emissions alone accounting for 14 to 34%. As these ranges suggest, there is a highly variable level of exposure to vehicle emissions based on time and location. This heterogeneity is due, at least in part, to the plume dynamics associated with the freshly emitted aerosol from tailpipes. For example, particle number concentrations are substantially elevated within the first few hundred meters of a roadway (7). Freshly emitted aerosol from a tailpipe exists as a hot, concentrated plume containing volatile, semivolatile, and nonvolatile material. Dilution and rapid cooling of the exhaust * Corresponding author e-mail: [email protected]. 10.1021/es102009e

 2010 American Chemical Society

Published on Web 09/15/2010

in ambient air causes both condensation of lower volatility material and evaporation of higher volatility material to/ from the particle phase. Typically, the number concentration decreases and the size distribution shifts to larger diameters as the plume moves away from the point of emission (7-9). These are important points to consider since both number concentration and size distribution affect the potential health hazards of ambient PM, and both quantities vary depending on where a receptor is located in relation to the point of emission. Work has been done to quantify how nanoparticle sources and exposure change on a community wide level (10, 11), and while these studies are useful, they have not focused on the contributions of short-lived processes. Highly time-resolved measurements of ambient nanoparticle concentration (12) and composition (3) have shown that near a roadway there are short bursts of nanoparticles associated with passing vehicle traffic. Nanoparticle emission from a motor vehicle is a complex process that depends on many factors such as fuel type, emission control system, and engine load (13-15). As a vehicle accelerates from a stop, the engine load is greater and a greater number concentration of particles is emitted when compared to an idling or nonaccelerating vehicle. Minoura et al. (16) showed that changes in number concentration can be used to detect patterns within a single traffic light cycle. Particle emissions quickly rise as the traffic light turns green and gradually fall as the traffic at the intersection dissipates, until another red/green light cycle occurs. The authors also were able to track shifts in particle diameter as well as estimate the volatile organic fraction of these particles. However, the study stopped short of quantifying the contribution of these patterns to the ambient nanoparticle concentration. The present work attempts to both quantify the contribution of fast changes in nanoparticle concentration near a roadway to the total particle concentration, and also to further characterize the microenvironment of a traffic intersection. The measurements described herein are part of the Ultrafine Aerosol Characterization Experiment (ULTRACE), a collaborative study between the University of Delaware and the National Center for Atmospheric Research. Measurements were performed during the summer and winter of 2009 in Wilmington, Delaware with the goal of characterizing the source and composition of urban nanoparticles.

Experimental Section Site Geography. Measurements were taken in Wilmington, Delaware at a State of Delaware Air Quality Monitoring Site managed by the Delaware Department of Natural Resources and Environmental Control (DNREC). Figure 1 shows a satellite image of the measurement site with the location of the instrumentation trailer marked with a star and other important locations labeled A-D. Three major roads intersect at this location. Martin Luther King (MLK) Boulevard and Lancaster Avenue are the larger roads located immediately to the north of the trailer, running diagonally from left to right across the figure. Each is three lanes wide and traffic runs one way in the direction indicated. The smaller road immediately to the West is Justison St. which changes name to North Washington St. on the North side of Lancaster Ave. Points A-C correspond to the shortest distance from the trailer to Justison St., the intersection of Justison St. and MLK Blvd., and the intersection of Justison St. and Lancaster Ave., respectively. These points are located approximately 15, 45, and 61 m to the particle inlets for the instruments, and in directions approximately 300°, 340° and 25° from North. Point VOL. 44, NO. 20, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7903

FIGURE 1. Satellite image of the measurement site and specific locations discussed in the text. (Source: http://earth.google. com). D shows a local fire station about 126 m from the site at 70° from North. On average over 28,000 vehicles pass through this intersection daily, although traffic counts were not available during the study period. Wind Data, Particle Number Concentration, and Size Distribution. A summary of the relevant instruments and time periods of operation are given in Table 1. Particle size distributions were obtained once every second during the winter campaign with a fast mobility particle sizer (FMPS; model 3091, TSI Inc., St. Paul, MN), while the number concentration data were obtained once every second during both summer and winter campaigns with a condensation particle counter (CPC; model 3025A, TSI Inc., St. Paul, MN). Sampling inlets were fashioned in the following manner. The FMPS was attached to a 13-ft. length of 3/8-in. (OD) copper tubing which extended to a height of approximately 15 ft. above the ground. The CPC was configured in the same manner during the summer. During the winter, the CPC flow was diluted by a factor of 4 with clean air provided by a HEPA filter cartridge (model 12144, Gelman Sciences, Ann Arbor, MI) to prevent saturation of the CPC, which was observed during portions of the summer campaign. The inlets for the CPC and FMPS were located within 1 ft. of each other and the aerosol flows into the inlets were matched to ensure that the same air mass with the same sampling residence time was analyzed. These inlets transmitted particles with greater than 90% efficiency over the size range of this study. Wind speed and direction were obtained with a Vaisala 425H Ultrasonic Sensor in 1-min averages. This device was located in a separate trailer from the CPC and FMPS instruments. Because the wind data were provided by the minute, but the resolution of the CPC/FMPS was 1 s, a linear interpolation was used to estimate the wind speed and direction at each CPC/FMPS data point. Data Analysis. The CPC and FMPS data consisted of a series of abrupt spikes in particle number concentration on top of a less intense and slowly changing background. These spikes were typically on the order of a few seconds to tens of seconds long. To separate these particle spikes from the baseline number concentration, an algorithm was developed that utilized wavelet decomposition as the separation mechanism. Wavelet decomposition has attracted a great deal of attention recently due to the advantages it possesses over the Fourier transform, especially in terms of detecting and separating signals that do not occur at stable frequencies such as HPLC chromatograms (17), NMR spectra (18), and Raman spectra (19). The choice of a wavelet function is complicated and affected by many parameters, including signal length, level of decomposition, and wavelet selection. Detailed reviews are available elsewhere (20, 21). 7904

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 20, 2010

At its most basic level, wavelet decomposition is the application of a pair of high and low pass filters to a signal for the purpose of separating it into high and low frequency representations. The first level of decomposition yields a high pass representation of the signal referred to as the first level detail, or d1, and a low pass representation referred to as the first level approximation, or a1. The first level approximation, or low pass portion of the signal, is then input into a second level of wavelet decomposition, which is essentially another high and low pass filter pair that distinguishes a different band of frequencies. The second decomposition creates a second level detail, d2 and second level approximation, a2. The second level approximation can then be broken down further into another detail and approximation, and the process can be repeated a number of times, n, until the desired level of decomposition is achieved or the signal can no longer be broken down. The end result is that the signal has been partitioned into different bands of frequencies, represented by d1 to dn and a single approximation representing the lowest frequencies present in the signal an. If one were to recreate the original signal, every detail level would n be summed (∑i)1 di) and then combined with the final approximation an. Particular representations can also be summed to recreate only the portion of the signal of interest for a particular application. For example, d1 and d2 can be summed to generate only the portion of the signal which falls into those frequency bands. For the current application, this is advantageous because it provides the ability to recreate only the portion of the signal corresponding to the particle spikes, while leaving behind other time-dependent features of the signal. In developing the separation method, both the level of decomposition and the type of wavelet were assessed in relation to the ability of the selected wavelet to reconstruct the original signal, and to visually separate the abrupt particle spikes. Artifacts associated with instrumental noise were reduced by applying an 11 point rolling average to the CPC and FMPS data prior to wavelet decomposition. The results are summarized in Table S1 (Supporting Information). Among the 38 wavelets that were tested, no meaningful difference was observed between the reconstructed signal and the original signal, and all but one yielded particle spike contributions within two percentage points of one another. The lone wavelet that did not fall within this range provided an unreasonable background, consisting of a variable intensity square wave. The similarity among the wavelets tested indicates that for the present application the choice of wavelet is a noncritical factor, assuming a logical background is predicted. Multiple levels of signal decomposition were explored, and the d8 level was deemed the most appropriate for several reasons. First, the d8 level corresponds to events with periods between 256 and 512 s while the d9 level corresponds to events with periods between 512 and 1024 s. The d9 level is clearly an inappropriate choice for the separation of the particle spikes as it encompasses periods which are 5-17 min long. Furthermore, when the d9 level was tested with a segment of CPC data, it badly overpredicted the contribution of the particle spikes to the total signal (Figure S1). This places the upper limit of the signal decomposition at the d8 level. Second, the d7 level of decomposition corresponds to events which are 128 to 512 s long and when compared to the d8 level it was found to under-predict the contribution of particle spikes to the total signal (Figure S1). The particle number concentration associated with “spikes” in the data was determined in the following way. First, the total signal was decomposed to the d8 level and the resulting approximation, a8, was subtracted from the total signal. This gave the first estimate of the particle spike signal, S1. Next, because the value of the baseline cannot exceed the

TABLE 1. Summary of Measurements and Particle Spike Contributions instrument CPC summer CPC winter FMPS CPC (winter during FMPS) Vaisala 425H

measurement

dates

contribution

number concentration number concentration number concentration and size distribution number concentration wind direction and speed

July 1-3 and July 8-15, 2009 December 10-22, 2009 December 17-22, 2009

15 ( 3% 19 ( 4% number: 24 ( 6%; PM0.1 mass: 20 ( 4% 24 ( 5% NA

actual value of the signal at any point, if a point in S1 was less than zero, the corresponding value in a8 was set to its true value from the original signal. This created the corrected first baseline estimate, b1, which was then decomposed again to the d8 level. The process was repeated until bn and b(n+1) differed by less than 1%. The particle spike signal was obtained by subtracting b(n+1) from the original signal. The particle spike signal was analyzed in two ways. First, the contribution of particle spikes to the total number concentration was determined by dividing the spike signal by the total number concentration at each time point. Second, a matrix of particle spikes was constructed that matched the time an individual spike was observed with the number concentration (and size distribution from FMPS data) associated with that spike. Separate matrices were constructed from CPC and FMPS data. The size distributions in the FMPS matrix were used to estimate the contribution of spikes to the PM0.1, and the top 1000 spikes in this matrix (based on the second derivative of the number concentration) were then treated in two ways. First, number concentrations were combined with wind direction to generate rose plots to determine where the spikes originated. Second, the corresponding size distributions were vector normalized, and averaged over three angular regions of wind direction, which represented three well-defined distances from the probable location where the spike originated to the aerosol sampling inlets at the trailer. All wavelet decompositions, Fourier transforms, and data analyses were performed using the MATLAB (R2009b) computing environment and its associated Wavelet Toolbox.

December 17-22, 2009 continuously

shows the total number concentration and baseline for a portion of the FMPS data set. The difference between the two plots was used to obtain the number concentration associated with particle spikes, as discussed in the Experimental Section. Figure 3b shows the contribution of particle spikes, expressed as percentage of the total number concentration, for both the CPC and FMPS data sets during a 5-day period in December 2009. The contributions determined by the two instruments tracked each other very well (r2 ) 0.77). While two notable exceptions on early Saturday morning and late Sunday evening were registered, they did not substantially alter the time-averaged contribution of spikes to the total signal (Table 1). As Figure 3b shows, the contribution of the particle spikes to the total number concentration exhibited large variability from day to day and hour to hour, with hourly contributions often exceeding 40%. These variations are a reflection of the temporal changes in traffic density, traffic type, and prevailing weather patterns. The average daily contributions of particle spikes to number and mass concentrations are summarized in Table 1. The summer contribution to number concentration was slightly lower than winter, and there was no statistically significant difference between FMPS and CPC averaged data. The contribution of particle spikes to mass concentration was obtained from FMPS size distributions and represented an

Results and Discussion Highly time-resolved measurements with the CPC and FMPS showed a series of abrupt spikes in number concentration that varied in length from a few seconds to tens of seconds. When viewed over a relatively short period of time, these spikes appeared to occur at regular intervals. As an example, Figure 2a depicts the CPC trace during the morning of July 9, 2009. To determine the spacing of these spikes, a Fourier transform of the data was performed and the frequency distribution was converted into a period distribution. The resulting periodogram in Figure 2b shows that the shortest period (highest frequency) component occurs between 106 and 118 s. This period matches the 108 s cycle time of the traffic signal at the intersection adjacent to the measurement site, indicating that the particle concentration spikes are associated with traffic flow through the intersection. The narrow widths of the spikes suggest they arise from acceleration of vehicles that were initially stopped at a red light. The longer time period features in Figure 2b are not informative with respect to this study. If one performs a Fourier analysis of CPC traces during other time periods of the measurement, the feature around 108 s is consistently observed while the other features vary. Wavelet decomposition, as described in the Experimental Section, was used to decompose the CPC and FMPS data into separate contributions from the spikes and underlying baseline. The results are illustrated in Figure 3. Figure 3a

FIGURE 2. (a) CPC trace for the morning of July 9, 2009 showing recurring particle spikes. (b) Fourier transform of the CPC trace. The shortest period characteristic of the spikes matches the 108 s cycle time of a traffic light at a nearby intersection. VOL. 44, NO. 20, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7905

FIGURE 3. (a) Wavelet-based separation of the total particle number concentration (black) into spikes and an underlying baseline (red) for a portion of the FMPS data set on December 21, 2009. (b) Contribution of the separated particle spike signal to the total number concentration for the CPC (red) and FMPS (green) data sets during the FMPS measurement period, December 17-22, 2009. average of 0.021 µg/m3 or 20% of the daily PM0.1 level. The diurnal variation averaged over all data is given in Figure S2. The contribution of particle spikes to the total number concentration was about a factor of 2 higher in the morning than late afternoon or evening. Based on the Fourier transform in Figure 2b, it was suspected that the particle number concentration spikes arose from vehicles accelerating from a stop at the adjacent intersection. This hypothesis was explored through wind rose plots constructed from the highest intensity spikes. Figure 4a shows a plot of number concentration vs wind direction for the top 1000 spikes in the FMPS data set. The most intense spikes are detected preferentially at a few wind directions. Wind rose plots of spikes in the CPC data show a similar preference (Figure S3). The feature at 300° is attributed to the acceleration of vehicles on Justison St., location A in Figure 1. This wind direction corresponds to the shortest tailpipe to sampling inlet distance. A second feature at 345° is attributed to acceleration of stopped vehicles at the intersection, either at the head of Justison St. or more likely (because of higher motor vehicle traffic) on MLK Blvd., location B in Figure 1. Particle spikes are also observed in the 0-60° range, which encompasses the Washington St.-Lancaster Ave. intersection, location C in Figure 1, and the fire station, location D in Figure 1. The enhancement at 45° in Figure 4a is less apparent in the longer time-period CPC data sets (Figure S3), which most likely reflects increased activity at the fire station during the specific time period of FMPS measurement. If the assignments of specific locations to spikes in number concentration are correct, another observation should follow based on the distance, or range of distances, to the sampling inlet. Aerosol plumes originating far from the sampling inlet should have size distributions biased toward larger particle diameters, whereas plumes originating close to the sampling 7906

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 20, 2010

FIGURE 4. (a) Wind rose plot of the top 1000 particle spikes in the FMPS data set. (b) Angular average of the particle size distribution as a function of distance from tailpipe to sampling inlet: (9) 15 m, (b) 48-60 m, and (2) 70-126 m. inlet should have distributions biased toward smaller particle diameters (7-9). To test this hypothesis, size distributions from the FMPS data set were averaged over three specific angular ranges: 290-310° representing emissions from Justison St. approximately 15 m from the sampling inlet, 20-40° representing emissions from the Washington St.Lancaster Ave. intersection approximately 48-60 m from the sampling inlet, and 60-80° representing emissions the fire house and/or vehicles further down MLK Blvd. or Lancaster Ave. approximately 73-126 m from the sampling inlet. The size distributions are plotted in Figure 4b. The particles emitted from the direction of Justison St. (15 m) show the narrowest distribution and the greatest fraction of particles under 20 nm, while those from locations further away show broader distributions extending to larger particle sizes. A size distribution was obtained for spikes from 320 to 340°, representing emissions from the intersection of Justison St. and MLK Blvd. approximately 30-55 m from the sampling line. This plot, omitted from Figure 4b for clarity, is shown in Figure S4. Not surprisingly, it is similar in shape to the 20-40° plot as the distances overlap. The correlation of particle size with distance from the sampling inlet is consistent with particle spikes arising from accelerating vehicles at the intersection. A final test of the origin of particle spikes is the wind velocity dependence of the observed size distribution. As the wind speed increases, the size distribution should shift toward smaller particle diameters due to a shorter time from tailpipe to sampling. A wind direction of 290-310° was chosen for this analysis for two reasons. First, this is the only direction around the measurement site where there is definitively only

one source of particle emission. Other directions are more difficult to assess owing to multiple locations of vehicle acceleration. Second, the aerosol plume undergoes rapid changes as it is diluted by ambient air. To monitor these changes, the source of these emissions should be sufficiently close to the site to be able to observe them; Justison St. (location A) is the closest possible source of freshly emitted aerosol at 15 m away from the particle inlets for the CPC and FMPS. When the size distributions originating from Justison St. were area normalized and averaged over 4 wind velocity bins, a clear pattern emerged as shown in Figure S5. As the wind velocity increased, the size distribution during particle spikes shifted to smaller particle diameters. At the highest wind velocities the distribution was approximately bimodal, while at lower velocities it was trimodal. Interestingly when the same method was applied to particle distributions originating from 20-40° (location C) all of the observed distributions were trimodal. To explain why the size distributions from this direction do not change, wind velocity and distance data were transformed into the transit time from emission to the sampling inlet. These times did not include time spent in the sampling inlet (∼2 s). At the highest wind velocity observed from Justison St. (location A), the aerosol would have spent less than 2 s in transit, whereas at the lowest velocity it would have spent greater than 15 s in transit. The Justison St. aerosol plumes developed a trimodal distribution after a transit time of approximately 3-4 s. At the highest wind velocity observed from location C, the aerosol transit time was nearly 7 s, almost a factor of 2 longer than the transit time needed for the Justison St. aerosol to achieve a trimodal distribution. Therefore, even at the highest wind velocities observed, the aerosol plumes originating from location C would have spent more than enough time in the atmosphere to achieve a trimodal distribution before they were sampled and would not show the evolution from a bimodal to a trimodal distribution. The wind direction, velocity, and particle size distribution dependencies in this work are all consistent with spikes in number concentration originating from specific locations around an intersection. These locations match those where vehicles accelerate after a red light turns green. Number concentration spikes represent a significant fraction of the total particle concentration near an intersection, exceeding 50% during some time periods. Motor vehicle emissions represent an even greater fraction of the total particle concentration because they contribute to the baseline as well as the spikes. These results indicate that human exposure to ambient nanoparticles strongly depends not only on distance from a roadway but also distance from a traffic intersection.

Acknowledgments This research was funded by the Health Effects Institute through agreement 4775-RFPA06-4/07-9. Research described in this article was conducted under contract to the Health Effects Institute (HEI), an organization jointly funded by the United States Environmental Protection Agency (EPA) (Assistance Agreement CR-83234701) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views of HEI, or its sponsors, nor do they necessarily reflect the views and policies of the EPA or motor vehicle and engine manufacturers. We thank Richard Cavalere and TSI Inc. for providing the fast mobility particle sizer used in this study, and Steven Brown for suggesting the wavelet decomposition approach.

Supporting Information Available Five figures and one table referenced in the text. This information is available free of charge via the Internet at http://pubs.acs.org/.

Literature Cited (1) Kittelson, D. B.; Watts, W. F.; Johnson, J. P. Nanoparticle emissions on Minnesota highways. Atmos. Environ. 2004, 38 (1), 9–19. (2) Kulmala, M.; Vehkama¨ki, H.; Peta¨ja¨, T.; Dal Maso, M.; Lauri, A.; Kerminen, V. M.; Birmili, W.; McMurry, P. H. Formation and growth rates of ultrafine atmospheric particles: A review of observations. J. Aerosol Sci. 2004, 35 (2), 143–176. (3) Zordan, C. A.; Wang, S.; Johnston, M. V. Time-resolved chemical composition of individual nanoparticles in urban air. Environ. Sci. Technol. 2008, 42 (17), 6631–6636. (4) Oberdorster, G.; Oberdorster, E.; Oberdorster, J. Nanotoxicology: An emerging discipline evolving from studies of ultrafine particles. Environ. Health Perspect. 2005, 113 (7), 823–839. (5) Peters, A.; von Klot, S.; Heier, M.; Trentinaglia, I.; Hormann, A.; Wichmann, H. E.; Lowel, H. Exposure to traffic and the onset of Myocardial Infarction. N. Engl. J. Med. 2004, 351 (17), 1721– 1730. (6) Abu-Allaban, M.; Gillies, J.; Gertler, A.; Clayton, R.; Proffitt, D. Motor vehicle contributions to ambient PM10 and PM2.5 at selected urban areas in the USA. Environ. Monit. Assess. 2007, 132 (1), 155–163. (7) Zhang, K. M.; Wexler, A. S. Evolution of particle number distribution near roadways. Part I: Analysis of aerosol dynamics and its implications for engine emission measurement. Atmos. Environ. 2004, 38 (38), 6643–6653. (8) Zhang, K. M.; Wexler, A. S.; Zhu, Y. F.; Hinds, W. C.; Sioutas, C. Evolution of particle number distribution near roadways. Part II: The ‘road-to-ambient’ process. Atmos. Environ. 2004, 38 (38), 6655–6665. (9) Zhang, K. M.; Wexler, A. S.; Niemeier, D. A.; Zhu, Y. F.; Hinds, W. C.; Sioutas, C. Evolution of particle number distribution near roadways. Part III: Traffic analysis and on-road size resolved particulate emission factors. Atmos. Environ. 2005, 39 (22), 4155– 4166. (10) Krudysz, M.; Dutton, S.; Brinkman, G.; Hannigan, M.; Fine, P.; Sioutas, C.; Froines, J. Intra-community spatial variation of sizefractionated organic compounds in Long Beach, California. Air Qual. Atmos. Health 2009, 2 (2), 69–88. (11) Kasumba, J.; Hopke, P. K.; Chalupa, D. C.; Utell, M. J. Comparison of sources of submicron particle number concentrations measured at two sites in Rochester, NY. Sci. Total Environ. 2009, 407 (18), 5071–5084. (12) Ogulei, D.; Hopke, P. K.; Ferro, A. R.; Jaques, P. A. Factor analysis of submicron particle size distributions near a major United States-Canada trade bridge. J. Air Waste Manage. Assoc. 2007, 57 (2), 190–203. (13) Kittelson, D. B.; Watts, W. F.; Johnson, J. P. On-road and laboratory evaluation of combustion aerosols. Part 1: Summary of diesel engine results. J. Aerosol Sci. 2006, 37 (8), 913– 930. (14) Kittelson, D. B.; Watts, W. F.; Johnson, J. P.; Schauer, J. J.; Lawson, D. R. On-road and laboratory evaluation of combustion aerosols. Part 2: Summary of spark ignition engine results. J. Aerosol Sci. 2006, 37 (8), 931–949. (15) Jayaratne, E. R.; Meyer, N. K.; Ristovski, Z. D.; Morawska, L.; Miljevic, B. Critical analysis of high particle number emissions from accelerating compressed natural gas buses. Environ. Sci. Technol. 2010, 44 (10), 3724–3731. (16) Minoura, H.; Takekawa, H.; Terada, S. Roadside nanoparticles corresponding to vehicle emissions during one signal cycle. Atmos. Environ. 2009, 43 (3), 546–556. (17) Shao, X.; Cai, W.; Pan, Z. Wavelet transform and its applications in high performance liquid chromatography (HPLC) analysis. Chemom. Intell. Lab. Syst. 1999, 45 (1-2), 249–256. (18) Neue, G. Simplification of dynamic NMR spectroscopy by wavelet transforms. Solid State Nucl. Magn. Reson. 1996, 5 (4), 305–314. (19) Ehrentreich, F.; Summchen, L. Spike removal and denoising of raman spectra by wavelet transform methods. Anal. Chem. 2001, 73 (17), 4364–4373. (20) Daubechies, I. The wavelet transform, time-frequency localization and signal analysis. IEEE Trans. Inf. Theory 1990, 36 (5), 961–1005. (21) Parameswariah, C.; Cox, M. Frequency characteristics of wavelets. IEEE Trans. Power Delivery 2002, 17 (3), 800–804.

ES102009E

VOL. 44, NO. 20, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7907