Identifying Likely PM2.5 Sources on Days of ... - ACS Publications

A simple statistical method is described for identifying the likely importance of local sources of PM2.5 in a region on days when the National Ambient...
0 downloads 0 Views 1016KB Size
Environ. Sci. Technol. 2009, 43, 2407–2411

Identifying Likely PM2.5 Sources on Days of Elevated Concentration: A Simple Statistical Approach NANJUN CHU,† JOSEPH B. KADANE,† A N D C L I F F I . D A V I D S O N * ,‡ Department of Statistics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, and Departments of Civil & Environmental Engineering and Engineering & Public Policy, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213

Received June 4, 2008. Revised manuscript received December 24, 2008. Accepted January 22, 2009.

A simple statistical method is described for identifying the likely importance of local sources of PM2.5 in a region on days when the National Ambient Air Quality Standard is exceeded. The method requires only PM2.5 mass concentration and wind direction data, and makes use of the EPA database on PM2.5 emissions in the local region of interest. The method has been illustrated using data from the Pittsburgh Air Quality Study, and suggests that local sources can be very important in affecting PM2.5 exceedances. The results have implications for many of the urban areas in the eastern United States downwind of large sources in the Midwest, and shows that simple statistical tests can be of value in identifying regions where further testing with sophisticated air quality dispersion models and sourcereceptor models is warranted.

Introduction It is well known that large stationary sources in the Midwest portion of the United States contribute to airborne particle loadings in regions further east. Especially important are coal-fired power plants which emit sulfur dioxide that is converted to particulate sulfate during atmospheric transport. This secondary sulfate can be responsible for one-third to one-half of the total mass in PM2.5, airborne particulate matter with diameters smaller than 2.5 µm. Additionally, atmospheric transport can contribute substantial fractions of secondary organic aerosol created from emissions from motor vehicles and stationary sources considerable distances upwind (1). It is reasonable to expect that sulfate and organics from long-range transport are the main culprits leading to exceedances of the 24-h average National Ambient Air Quality Standard for PM2.5. While it is recognized that local sources may also contribute to poor air quality, the relative importance of such sources on exceedance days is not well established. Assessing the contribution of local sources to overall PM2.5 concentrations is usually accomplished through air quality dispersion modeling as well as source-receptor modeling of ambient data. Sophisticated air quality dispersion models have been developed for precisely this purpose, such as the * Corresponding author phone: 412-268-2951; fax: 412-268-7813; e-mail: [email protected]. † Department of Statistics. ‡ Departments of Civil & Environmental Engineering and Engineering & Public Policy. 10.1021/es801548z CCC: $40.75

Published on Web 02/25/2009

 2009 American Chemical Society

Industrial Source Complex model and more recently AERMOD of the U.S. Environmental Protection Agency (2). Using these models requires input data from emission inventories as well as meteorological and topological databases. Alternatively, source-receptor modeling generally requires a set of PM2.5 chemical composition data and possibly concentrations of other air contaminants. It is necessary to collect enough PM2.5 for bulk chemical analysis which requires timeintegrated sampling and costly analysis methods, or use of complex single-particle analysis instruments such as mass spectrometry. The processes involved can also be understood using stochastic models. Henry et al. (3) and Yu et al. (4) use nonparametric regression to identify likely pollution sources. Henry (5) back-calculates pollution trajectories, assuming a reasonably flat terrain and wind monitoring close to pollution sources. With the emphasis on increasingly sophisticated techniques, there has been a lack of interest in simple statistical methods that can nevertheless provide valuable and occasionally surprising findings. In this study, we illustrate one such method for easily determining whether local sources in a particular direction are likely to influence a receptor site on exceedance days. This knowledge can then be used to guide application of some of the more sophisticated techniques described above. The method takes advantage of one year of 10-min average PM2.5 mass concentrations and wind direction data obtained during the Pittsburgh Air Quality Study.

Experimental Section The Pittsburgh Supersite included over 30 instruments sampling various gases and particles during July 2001 to September 2002 (6). One of these instruments was a Rupprecht and Patashnick model 1400A tapered element oscillating microbalance (TEOM). The TEOM works on the principle that the natural frequency of oscillation of a hollow tube, which is the “tapered element,” varies according to its mass. One end of the tube is clamped while the other end is free to vibrate. Ambient air containing PM2.5 is first dried using a nafion dryer to reduce artifacts caused by high relative humidity. The dried air passes through a Teflon-coated borosilicate glass filter positioned at the free end of the tube, allowing the particles to collect on the filter. The mass on the filter is monitored every 2 s by measuring the frequency of vibration of the tube. The instrument reports PM2.5 mass concentrations using 300-s rolling averages, which have been subsequently combined to give an overall average concentration every 10 min. Wind speed and direction data have also been obtained at the same location with values reported each 10 min. The monitoring station was located on a hill in Schenley Park, adjacent to the Carnegie Mellon University campus about 6 km east of downtown Pittsburgh. This location is one of the highest points in the city. Both the TEOM and the wind instruments were positioned on the roof of a trailer about 3 m above the ground. The winds measured at this location are representative of regional winds without being greatly influenced by local topography, as demonstrated by comparing the wind rose for the monitoring site with that at the National Weather Service station at the Pittsburgh airport 25 km to the west. Similarly, the PM2.5 data have been shown to be representative of the region rather than local conditions, at least on most of the days in the sampling period (7). There are no heavily traveled streets within a few hundred VOL. 43, NO. 7, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2407

FIGURE 1. Sources of PM2.5 in the vicinity of Pittsburgh, PA, as taken from the EPA Emission Inventory (8). Names of counties in the region are shown. For each source, the area of the circle is proportional to the logarithm of the emission rate. meters of the station. There are also no significant stationary sources within 2 km except for the Bellefield Boiler Plant, a coal-fired steam heating facility 0.78 km WNW (291°) from the monitoring station that has been shown through statistical analysis to have negligible influence on the measured PM2.5 levels. This plant was observed to have substantial plume rise from a stack of approximately the same height as the monitoring station; it is likely that the plume was rarely if ever close to ground level upon reaching the PM2.5 sampler, consistent with results of the statistical analysis of the data. The sampling period of interest for this work is the one-year time interval from September 1, 2001 to August 31, 2002. All stationary sources within 70 km of the station and emitting more than 15 tons per year of PM2.5 according to the EPA emission inventory (8) are shown in Figure 1. The largest sources in Figure 1, those with emissions exceeding 100 tons/year, are listed in Table 1. Note that the Bellefield Boiler, with an emission rate of 17.5 tons/year, is indicated in the figure. There are five sources to the S and SE of the station within 25 km, of which four are large enough to be listed in Table 1. These sources are of special interest as they are believed to influence PM2.5 at the station to varying degrees, as explained below. Although most of the emissions listed in Table 1 are from elevated stacks, the two USS sources have a majority of their emissions at low levels.

Results and Discussion Emissions from myriad sources influence the receptor site. Of most interest, however, are sources contributing on days when the concentration exceeds the 24-h PM2.5 National Ambient Air Quality Standard (NAAQS). To attain this standard, according to the U.S. EPA (9), “the 3-year average 2408

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 7, 2009

TABLE 1. All Sources Listed in the 2002 EPA Emission Inventory within 70 km of the Monitoring Site and with PM2.5 Emissions Exceeding 100 tons/year name of source Adrian Allegheny Ludlum Corp Cheswick Power Station Shelocta Homer City USS Edgar Thompson Connellsville USS Clairton El Rama Masontown New Eagle West Finley Shippingport Monaca Wampum West Pittsburgh

PM2.5 emissions distance to angle (tons/year) station (km) (degrees) 1830 271 591 8550 6850 472 105 394 556 9530 181 136 1600 418 298 1320

68 26 17 56 63 8.8 58 16 21 65 24 65 46 42 59 66

36 44 49 64 83 127 149 160 175 179 186 218 298 304 326 327

of the 98th percentile of 24-hour concentrations at each population-oriented monitor within an area must not exceed 35 µg/m3 (effective December 17, 2006).” We assume that source emissions during the 2001-2002 Supersite operation were reasonably representative of emissions in a typical year. Therefore the conclusions of this paper are likely to apply into the future, when the 35 µg/m3 standard will be applicable. Several of the 365 days in the study period had to be eliminated due to missing data for PM2.5, windspeed, or wind direction, or due to windspeed below 1 m/s for which wind direction data are not considered reliable. Of the total 52,560

FIGURE 2. Wind direction estimated for each 24-h day as described in the text, plotted around the compass circle for each exceedance day during the sampling year. The two indicated sectors show 112-142° and 145-201°, respectively, as discussed in the text. 10-min periods during the year, 6.9% had missing PM2.5 data, 5.4% had missing windspeed data, 18.5% had missing wind direction data, and 31.9% had windspeeds less than 1 m/s. Those 10-min periods with valid data, a total of 28,427 periods, were then used to calculate 3-h and 24-h average values. Of the 322 days remaining after the elimination process, 24 had an average PM2.5 concentration exceeding the 35 µg/m3 standard, termed “exceedance days” in this paper. Thus 1/13.4 of the days of valid data were exceedance days, higher than the 1/50 allowable by the 2006 NAAQS. Figure 2 shows the estimated wind direction for each full 24-h day for the exceedance days. The wind direction for each day has been calculated using the von Mises distribution (10). This is an analog of a normal (i.e., Gaussian) distribution whose domain is circular. Its parameters are a mean µ and a measure of precision κ. As κ goes to infinity, the limiting distribution is normal; as κ goes to zero, the limiting distribution is uniform on the circle. The von Mises distribution weights each 10-min average wind direction equally. An alternative would have been to weight each 10-min average according to its windspeed. The 24-h values have been computed using the 10-min data. Also shown in Figure 2 are the locations of the 16 sources listed in Table 1. The four sources of special interest S and SE of the monitoring station are included among these 16. The two indicated sectors correspond to 127° ( 15° and 145°-201°. The first sector corresponds to a reasonable variability in wind direction around the USS Edgar Thompson steel works. The second sector corresponds to a reasonable variability in wind direction around the three sources at 160°, 175°, and 186° in Table 1. Of the 322 days of valid data, there are 96 days when the average wind direction was within one of these sectors or the 3° interval between them, i.e., in the range 112°-201°. The PM2.5 concentration exceeded 35 µg/ m3 on 17 of those days. In contrast, the PM2.5 concentration exceeded 35 µg/m3 on 7 days when the wind was from other directions. Thus on 18% of the days when the wind was from the general direction of these sources, the PM2.5 standard was exceeded. On only 3.1% of the days when the wind was from any other direction was the standard exceeded.

FIGURE 3. Wind direction estimated for each 3-h period as described in the text, plotted around the compass circle for each 3-h period when the PM2.5 concentration exceeded 50 µg/ m3 during the sampling year. The two indicated sectors show 112-142° and 145-201°, respectively, as discussed in the text.

To further investigate the occurrence of high PM2.5 concentrations when the wind direction was in the range 112°-201°, the 3-h average values of PM2.5 and windspeed were also considered. The von Mises distribution was used to estimate the wind direction for each 3-h period. Figure 3 shows 3-h wind direction data plotted for those time periods when the average PM2.5 concentration exceeded 50 µg/m3. There were 60 time periods out of 2134 when this threshold concentration was exceeded. (The number of possible time periods is less than the ideal 2920 due to 786 periods without valid PM2.5, windspeed, or wind direction data, or when windspeed was below 1 m/s for the entire 3-h period.) The PM2.5 concentration exceeded 50 µg/m3 in 6.0% (38/631) of the 3-h periods when the wind direction was in the range 112°-201°. The concentration exceeded 50 µg/m3 in only 1.5% (22/1503) of the 3-h periods when the wind was from other directions. Thus data from both averaging periods suggest that sources SE-to-S of the monitoring station contributedisproportionatelytoelevatedPM2.5 concentrations. Figure 2 shows that 4 out of 24 exceedance days occurred when the 24-h wind direction was in the range 127°-142°. There were no exceedance days when the wind direction was in the range 112°-127°. If the Edgar Thompson steel works were affecting the sampling site, which is reasonable as it is the closest source in Table 1, we would expect at least some exceedance days to occur in this latter range. A probable reason is explained by the wind direction frequency over the sampling year: the wind rose shown in Figure 4 indicates that the frequency is low in the quadrant N to E and increases only gradually from S to SE. Thus the lack of exceedance days when winds were in the range 112°-127° most probably reflects the relatively low frequency of winds from that interval. The original data show that there were only 9 days during the full sampling period when the 24-h wind direction was in that interval, compared with 19 days when the wind was in the range 127°-142°. Note that 3-h wind directions have been chosen for Figure 4 since the transport of emissions VOL. 43, NO. 7, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2409

FIGURE 4. Wind rose for the monitoring station during the sampling year. Percentages of 3-h wind direction readings within each 5° range of the compass circle are shown.

FIGURE 5. PM2.5 concentration rose based on the wind direction data used to construct Figure 4. All 3-h concentrations when winds were from a 5° range of the compass circle have been averaged to arrive at the concentration shown for that range.

from the source to the monitoring site would require sustained winds from the SE for nearly an hour, assuming typical 3 m/s winds. This also allows time for vertical mixing within the boundary layer. The original wind data show a similar wind rose based on 24-h values, although the resolution is poorer due to the smaller number of datapoints. There are no other directions with a preponderance of 24-h winds on exceedance days. This is surprising considering the number of large sources to the west of the sampling site and the frequency of winds from this direction. Examination of the 3-h data shows that there were twelve 3-h periods with concentration above 50 µg/m3 when winds were from SWto-W, 225°-270°; there were only 4 days out of 322 when the concentration exceeded the 24-h average NAAQS with the wind direction in that interval. Thus it is likely that the sources SW-to-W of the station did not have a large influence in elevating the 24 h average PM2.5 concentration above background to reach 35 µg/m3 at the monitoring site. The largest of these sources are coal-fired power plants with tall stacks. The height of the stacks may be responsible for elevated plumes that did not reach the earth’s surface when passing over the monitoring site. Of course, such emissions may have contributed to higher background concentrations further downwind. Furthermore, the 3-h average windspeed for the interval 112°-201° is only 1.85 ( 1.15 m/s, compared to the average windspeed for the interval 201°-315° of 2.60 ( 1.36 m/s. This suggests that atmospheric mixing and hence dilution may be smaller when winds are in the former interval compared with the latter. Heights of the mixing layer are also different on days when the average wind is in the former range; the measured morning mixing height at the Pittsburgh airport averages 165 ( 156 m for the former interval compared with 398 ( 355 m for the latter. For the 17 exceedance days in the former range, the average mixing height is 200 ( 66 m, suggesting a low mixing height may be preventing dilution on at least some of the exceedance days. Finally, the importance of local sources is to be expected when averaging periods are short, such as 3 or 24 h, since local plumes are less likely to be averaged by background levels. The 24-h NAAQS for PM2.5 is meant to protect the public from acute effects where concentrations can be most affected by nearby plumes. Additional analyses were conducted to examine wind directions for 24-h average PM2.5 concentrations above 25

and 30 µg/m3 and for 3-h average PM2.5 concentrations above 40 and 45 µg/m3. Wind directions were also examined for 1-h average PM2.5 concentrations above 45, 50, 55, 60, and 65 µg/m3. In all cases, the patterns resembled those observed in the analyses described earlier. Further analyses conducted separately for each season of the year indicated that most periods of elevated concentration occurred in the summer, when photochemistry is greatest and winds from 112°-201° are most common. It thus appears that exceedance days occur mainly when local primary particle emissions add to already high background levels of secondary PM2.5 to push concentrations above the 35 µg/m3 NAAQS. To further check this hypothesis, analyses were conducted on PM2.5 data collected at two sites by the Pennsylvania Department of Environmental Protection (DEP). The sites include Greensburg, roughly 55 km ESE, and State College, roughly 180 km ENE of the Pittsburgh Supersite location. These sites were chosen because PM2.5 concentrations should represent regional background when winds are from 112°-201° due to a lack of significant PM2.5 sources in those directions. Concentrations at these two sites were always less than those in Pittsburgh on the 24 exceedance days. On many of these days, however, the concentrations were within 10 µg/m3 of the respective Pittsburgh value. This suggests that regional background constitutes an important part of total PM2.5 on exceedance days in Pittsburgh. The overall effect of high background concentrations and local emissions can be seen in the PM2.5 concentration rose in Figure 5. Much greater concentrations are seen on days of winds in the range 112°-201°. Based on the analyses above, it appears that sources to the SE and S of the Pittsburgh Supersite had a significant influence on measured PM2.5 on days when the concentrations were elevated; sources in other directions from the monitoring site apparently had much less influence despite greater emissions and a high frequency of winds from those other directions. Note that we did not consider recirculation of air masses that may have passed over the city a day or two earlier, then changed direction and passed over the city once again (11). Investigation of recirculation is beyond the scope of this study but warrants further work. The simple statistical method illustrated here can be used for other pollutants and other urban areas where requisite data are available. The method has shown the advantages of

2410

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 7, 2009

focusing on days when the NAAQS is exceeded, trying more than one averaging time for pollutants and wind direction to search for patterns, and mapping pollution sources with emissions greater than a certain value. This procedure can help identify influential sources as well as sources that are likely not playing a major role. The method may be most valuable when used in conjunction with source-receptor modeling.

Acknowledgments The assistance of Xiting Yang, David Hwang, and Anna Anselmi is greatly appreciated. This research was conducted as part of the Pittsburgh Air Quality Study that was supported by U.S. Environmental Protection Agency under contract R82806101 and the U.S. Department of Energy National Energy Technology Laboratory under contract DE-FC2601NT41017. This paper has not been subject to EPA’s peer and policy review, and therefore does not necessarily reflect the views of the Agency. No official endorsement should be inferred.

Literature Cited (1) Blanchard, C. L. Spatial and Temporal Characterization of Particulate Matter. In Particulate Matter Science for Policy Makers: A NARSTO Assessment; McMurry, P., Shepherd, M., Vickery, J., Eds; Cambridge University Press: Cambridge, England; http://www.cgenv.com/NARSTO link to PM Science and Assessment, 2004. (2) Revision to the Guideline on Air Quality Models: Adoption of a Preferred General Purpose (Flat and Complex Terrain) Dispersion Model and Other Revisions; U.S. Environmental Protection

(3)

(4) (5) (6) (7)

(8)

(9) (10) (11)

Agency, Part III, 40 CFR Part 51, Federal Register 70 (216), page 68218, November 9, 2005. Henry, R. C.; Chang, Y.-S.; Spiegelman, C. H. Locating nearby sources of air pollution by nonparametric regression of atmospheric concentrations on wind directions. Atmos. Environ. 2002, 36, 2237–2244. Yu, K. N.; Cheung, Y. P.; Cheung, T.; Henry, R. C. Identifying the impact of large urban airports on local air quality by nonparametric regression. Atmos. Environ. 2004, 38, 4501–4507. Henry, R. C. Locating and quantifying the impact of local sources of air pollution. Atmos. Environ. 2008, 42, 358–363. Wittig, B.; Anderson, N.; Khlystov, A. Y.; Pandis, S. N.; Davidson, C. I.; Robinson, A. L. Pittsburgh Air Quality Study overview and preliminary findings. Atmos. Environ. 2004, 38, 3107–3125. Tang, W.; Raymond, T.; Wittig, B.; Davidson, C. I.; Pandis, S.; Robinson, A. L.; Crist, K. Spatial variations in PM2.5 during the Pittsburgh Air Quality Study. Aerosol Sci. Technol. 2004, 38 (S2), 80–90. U.S. Environmental Protection Agency. Emission Inventory for 2002; 2007; ftp://ftp.epa.gov/EmisInventory/2002finalnei/2002_ final_v3_2007_summaries/point/2002_nei_v3_facility_sums_ sep10_2007.zip. U.S. Environmental Protection Agency. National Ambient Air Quality Standard for PM2.5; Washington, DC, 2008; http:// www.epa.gov/air/criteria.html. Fisher, N. J.; Lewis, T.; Embleton, B. J. J. Statistical Analysis of Spherical Data; Cambridge University Press: Cambridge, England, 2007. St. John, J. C.; Chameides, W. L. Climatology of ozone exceedances in the Atlanta metropolitan area: 1-h vs 8-h standard and the role of plume recirculation air pollution episodes. Environ. Sci. Technol. 1997, 31, 2797–2804.

ES801548Z

VOL. 43, NO. 7, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2411