Application of a counting technique to determine certain and uncertain geographic regions of emission sources

Application of a counting technique to determine certain and uncertain geographic regions of emission sources

Ecological Modelling 192 (2006) 627–636 Application of a counting technique to determine certain and uncertain geographic regions of emission sources...

831KB Sizes 0 Downloads 15 Views

Ecological Modelling 192 (2006) 627–636

Application of a counting technique to determine certain and uncertain geographic regions of emission sources Sandy Owega, Greg J. Evans ∗ , Badi-Uz-Zaman Khan, Robert E. Jervis, Mike Fila Department of Chemical Engineering and Applied Chemistry, University of Toronto, 200 College Street, Toronto, Ont., Canada M5S 3E5 Received 26 March 2004; received in revised form 31 May 2005; accepted 6 July 2005 Available online 10 October 2005

Abstract Potential source contribution functions (PSCF) are frequently used to determine possible emission sources of airborne pollutants. A counting technique was developed to correct for regions erroneously identified by conventional PSCF. While conventional PSCF uses empirically determined weighting factors to correct for these erroneous regions, the proposed technique counts trajectories passing through a location of interest and in adjacent regions to objectively determine its identification as a source region. Conventional PSCF plots and modified version using this new counting technique were created for total PM2.5 and nitrate PM2.5 mass concentration data collected between July 2001 and September 2002. The modified PSCF eliminated erroneous potential emission sources regions and also identified potential emission sources during episodic events. These episodic source regions were not observed with conventional PSCF plots. The PSCF plots of total PM2.5 mass concentration displayed geographic regions where power plants in Southern Ontario, Michigan and Ohio were located. Similarly, the PSCF plots of nitrate PM2.5 mass concentrations displayed areas in Canada and the United States related to agricultural regions. Two episodic emission sources corresponding with forest fires in Quebec and the Prairie Provinces were also identified by the modified PSCF. Interestingly, the Prairie forest fires were identified as a nitrate source, unlike the Quebec forest fires. © 2005 Elsevier B.V. All rights reserved. Keywords: Geographic emission sources; Episodic events; Nitrate-coated particulate matter; Atmospheric aerosol; Trajectory analysis

1. Introduction Identification of the types of emission sources contributing to a receptor site has been accomplished with a variety of ‘factor analysis’ techniques, where the ∗ Corresponding author. Tel.: +1 416 978 1821; fax: +1 416 978 8605. E-mail address: [email protected] (G.J. Evans).

0304-3800/$ – see front matter © 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolmodel.2005.07.025

observed variables can be explained largely or entirely in terms of a much smaller number of variables called factors. Examples of these include positive matrix factorization (PMF), UNMIX and principle components analysis (PCA) (Brown et al., 2004). The data is typically provided as temporal trends of chemical, mass and number concentrations. Research published by Hopke et al. in 1989 described a method for determining the location of these sources. This method was called

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

628

Nomenclature d iφ jφ k mij

nij

PSCFij

r s ti

tj

Wij x y

arithmetic standard deviations of counts in a given row or column number of cells in row i with at least one count number of cells in column j with at least one count number of counts in cell (i, j) number of times those air masses traveled through a location (i, j) and elevated pollutant levels observed at the receptor number of times those air masses traveled through the map grid cell (i, j) and arrived at the receptor a probabilistic estimate of how often air passing over a geographic location and arriving at a receptor site was polluted number of rows on the map grid number of columns on the map grid the latitudinal threshold number of counts for cell i that must be exceeded to ensure display of PSCFij the longitudinal threshold number of counts for cell i that must be exceeded to ensure display of PSCFij weighting factor used to compensate for PSCF values based on small nij number of counts in row i number of counts in column j

the potential source contribution function (PSCF), a function that identifies the geographic location of an emission source. It can be defined as the conditional probability that an air mass passing through a location of interest, and arriving at a receptor site, will be polluted (Zeng and Hopke, 1989). The effects of photochemistry and meteorology on environmental pollutants have since been investigated with the PSCF model (Cheng et al., 1993; Gao et al., 1993; Zeng and Hopke, 1994). An inherent problem with the PSCF model is the identification of potential geographic locations of an emission source that may in fact not exist, and produce an unrealistic attribution of source regions. Minimizing the uncertainty by down-weighting the PSCF at each

geographic location has often been performed (Polissar et al., 2001a). This was typically accomplished by introducing an empirically determined weighting factor. A geographic location through which air masses traveled frequently was not weighed down as heavily as a geographic location through which air masses traveled infrequently (Polissar et al., 2001b). To further evaluate the uncertainty in the PSCF plots, a boot-strapping technique was developed (Hopke et al., 1994). This method provided a PSCF error estimate by calculating an average and a corresponding standard deviation for every PSCF at each geographic location. Other methods include the evaluation of the relative standard deviation (R.S.D.) at each geographic location and the subsequent display of only those PSCF values that were greater than a user-selected R.S.D. (Tsai et al., 2004). This approach has left room for improvement. In this paper, we propose a new technique that, unlike earlier work, objectively differentiates between certain and uncertain PSCF values. A counting approach was chosen to accurately distinguish between erroneous and actual source regions. This technique was applied to environmental data measured in Toronto between July 2001 and September 2002. Comparison between this and conventional techniques was performed to validate the former approach.

2. Potential source contribution function (PSCF) model The air mass trajectories during the sampling period were calculated, with a Toronto elevation of 500 m using the HYbrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model. HYSPLIT is the newest version of a complete system that computes air mass trajectories for complex dispersion and deposition simulations (Draxler, 1992). Each back trajectory, representing 1 h in the sampling period, described the location of an air mass every hour backwards in time for 5 days. Hence, 120 air mass locations or “endpoints” defined 1 h of a day, and 1 day had 2880 endpoints. Since the sampling period was 440 days, approximately 1.3 million endpoints were mapped on the PSCF. The PSCF of a geographic location described how often air passing over it and arriving at a receptor site was polluted. Geographic locations that were frequently associated with polluted air were taken to rep-

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

629

resent potential sources. The PSCF was a ratio, defined as: mij PSCFij = (1) nij where m is the number of times air masses with elevated pollutant levels observed at the receptor site traveled through the map grid cell (i, j), and n is the number of times air masses traveled through the map grid cell (i, j) and arrived at the receptor site during the sampling period. Elevated pollutant levels were defined as concentrations greater than the sum of its average and its standard deviation. Each map grid cell was a 0.1◦ × 0.1◦ latitude–longitude square, while the average number of endpoints for each cell was approximately 0.2. For conventional PSCF analysis, a weighting factor Wij was necessary to reduce the uncertainty in displaying an erroneous map grid cell (i, j). Typically, a weighting factor was multiplied by the PSCF value in a map grid cell (i, j), where the weighting factor and the range for endpoints were determined empirically. The four ranges and corresponding weighting factors chosen for the scaling were determined empirically:   1.00 100 < nij       0.70 65 < n ≤ 100   ij (2) Wij =  0.50 40 < nij ≤ 65        0.20 nij ≤ 40 An important limitation of applying a weighting factor technique pertains to episodic events that may last for only a few days. During these events, some map grid cells could contribute to poor air quality at a receptor site, and would never contribute to the receptor’s air quality after the episode. These cells represent high PSCF values, which may represent potential sources, but have a large associated uncertainty. These PSCF values should not be decreased with weighting factors, as is performed by this weighting technique.

Fig. 1. The counting technique.

(i, j). This approach has been applied to data generated by electron momentum spectrometers in earlier publications (Graham et al., 1993; Goruganthu et al., 1988). The treatment of the uncertainty associated with map grid cells was based on the number of counts that each map grid cell (i, j) contained. Uncertain map grid cells occurred when the number of times an air mass that traveled through a map grid cell (i, j) during the entire sampling period was small. Hence, a greater uncertainty should be attributed to these map grid cells than others. Consequently, during elevated episodes, these map grid cells (i, j) would display a large PSCF, an uncertain and likely erroneous result. A new method for basing the confidence associated with a PSCF value of a cell upon its number of counts was developed as illustrated in Fig. 1. On a grid of size r rows × s columns, let the number of counts in cell (i, j) be k, the total number of counts in all the cells for a given latitudinal domain be x, and the total number of counts in all the cells for a given longitudinal domain be y. Define ti =

x + 1.5di iφ

(3)

y + 1.5dj jφ

(4)

and 3. Elimination of uncertain PSCF regions

tj =

To identify map grid cells with uncertain PSCF values, a counting technique was employed, and further developed. The counts in this work refers to the number of trajectories traveling through a map grid cell

where iφ and jφ denote the number of cells in row i and column j, respectively, with at least one count, and di and dj are the arithmetic standard deviations of counts in row i and column j, respectively.

630

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

The PSCF values of only those map grid cells whose number of counts k exceeded both of their thresholds ti and tj are displayed on a PSCF map using this method. Thus, the confidence associated with the PSCF value of a potential source grid cell depended not just on its own number of counts, but also on those of the remaining cells in its latitudinal and longitudinal domains. Note that the portions of the distribution that did not exhibit counts were excluded from the analysis. Thus areas that never contributed to a receptor’s air quality were discarded.

4. Environmental data The mass concentrations of particulate matter with aerodynamic diameter less than 2.5 ␮m (PM2.5 ) were monitored by a Series 1400a TEOM and a 8400 RealTime Nitrate Analyzer (Rupprecht and Patashnick Co. Inc., Albany, NY). The total PM2.5 mass concentrations were measured at the Gage Institute in the University of Toronto, while those of nitrate PM2.5 were measured at the Wallberg Building in the University of Toronto. The locations were less than 100 m apart in downtown Toronto, just outside the core of high-rise office towers. The sampling period for the TEOM measurements was between July 1, 2001 and September 13, 2002. The sampling period for the real-time nitrate analyzer measurements was between July 1, 2001 and September 15, 2002.

5. Results 5.1. PSCF plots without weighting factors Fig. 2 illustrates the PSCF results for the total PM2.5 and the nitrate PM2.5 mass concentrations, without any down-weighting performed on the PSCF values of cells containing a small number of trajectories. The PSCF plots only displayed those map grid cells (i, j) where the PSCF values were greater than 0.01. The total PM2.5 mass concentration PSCF plot in Fig. 2A displayed three questionable regions. The first region was located at the Pacific Ocean and stretched north until Alaska. The second region was located over Baffin Island. The third region stretched from Florida to California, where several hot spots were displayed.

The nitrate PM2.5 mass concentration PSCF plot in Fig. 2B displayed four questionable regions. The first region was located at north-western British Columbia over the Pacific Ocean, the Yukon and Alaska. The second region was located at northern Baffin Island and Greenland. The third region was positioned west of Texas, and the last region was in the Pacific Ocean, west of Washington State. 5.2. PSCF plots with weighting factors Fig. 3 illustrated the PSCF plots of Fig. 2 with the scaling using the chosen weighting factors. The total PM2.5 mass concentration PSCF plot in Fig. 3A displayed three tracks that had elevated mass concentrations of particles arriving in Toronto. Two of these tracks started in Michigan, and traveled above and below Lake St. Clair. Track 1 traveled through Sarnia, and ended in Toronto, while Track 2 went above the north shore of Lake Erie to arrive in Toronto. Track 3 started in Ohio, traveled over Lake Erie, through Hamilton into Toronto. The nitrate PM2.5 mass concentration PSCF plot in Fig. 3B identified south-western Ontario, extending into Michigan, as a potential source contributor. 5.3. PSCF plots treated with counting technique In Fig. 4, the hot spots evident without downweighting to the PSCF plot, as illustrated in Fig. 2, were not observed. The erroneous areas identified in Fig. 2 were eliminated, and potential sources identified in Canada and the United States were intuitively more reasonable. The total PM2.5 mass concentration PSCF plot in Fig. 4A illustrated the similar three tracks found in Fig. 3A, but with higher PSCF values than in Fig. 3A. This observation illustrated one advantage of the counting technique. The PSCF values down-weighted in Fig. 3A were retained in Fig. 4A. Consequently, the largest PSCF value of 63% in Fig. 4A was not downweighted to 44%, as observed in Fig. 3A. Thus, the information regarding the relative importance of each source region is not lost in the counting technique, unlike in the down-weighting technique. The nitrate PM2.5 PSCF plot in Fig. 4B was similar to that of Fig. 3B. Again, the down-weighting effect was observed in Fig. 3B. Subtle differences are evident between Figs. 3 and 4. While Fig. 3A displays poten-

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

Fig. 2. PSCF plots of (A) total PM2.5 , and (B) nitrate PM2.5 data without any treatment.

631

632

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

Fig. 3. PSCF plots of (A) total PM2.5 , and (B) nitrate PM2.5 data with weighting factors.

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

Fig. 4. PSCF plots of (A) total PM2.5 , and (B) nitrate PM2.5 data using the counting technique to display significant map grid cells.

633

634

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

tial sources in only northern Ohio, Fig. 4A also shows sources in central Ohio. Similarly, potential sources in the Prairie Provinces were also displayed in Fig. 4A. These sources, while absent in Fig. 3B, were also evident in Fig. 4B. The significance of these differences is next discussed.

6. Discussion Clearly, the marked regions in Fig. 2 were suspect and should not have been displayed because the associated uncertainty was large. Even though salts could travel from the Pacific Ocean to arrive in Toronto, an 80% PSCF value cannot be justified. For the same reason, Baffin Island and Greenland can be intuitively disqualified, even though PM2.5 sources are located in the Arctic Circle. However, for the regions in the middle of the United States, interpretation of the PSCF plot was more difficult because the number of trajectories that traveled through these regions was not as small as the other erroneous regions. Both Figs. 3 and 4 demonstrate the enhancement in reducing the uncertainty in the PSCF plots. Gener-

ally, the PSCF value became more meaningful closer to Toronto. The PSCF values in Fig. 3A and B suggest that the potential sources were primarily in Canada. Fig. 5 (Environment Canada, 2003) indicates the locations of coal- and oil-fired power plants in the Great Lakes region. Note that both Figs. 3A and 4A illustrate the area near Lake Erie as an important potential source in Track 3. Not surprisingly, the Nanticoke power plant near Port Dover is located in this area. However, only Fig. 4A indicated central Ohio as a potential source region. Several coal- and oil-fired power plants are located in the high PSCF regions within Ohio (Tremoulet, 1999). Additionally, Tracks 1 and 2 identified Sarnia and Detroit as important potential sources of Toronto PM2.5 . Coincidentally, the Lambton power plant is located in Sarnia, while electrical utilities are present in Detroit and surrounding areas. The PSCF plot in Fig. 4A also indicated emissions from Quebec, close to James Bay. These emissions have been attributed to the forest fires in Quebec in July 2002. Even though these fires were observed for only 3 days in Toronto, this region exhibited a similar PSCF

Fig. 5. Geographic locations of PM2.5 emission sources in the Great Lakes Region.

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

value as the Michigan and Toronto regions according to the total PM2.5 mass concentration PSCF plot in Fig. 4A. These fires were not observed in Fig. 3A, indicating the difficulty of recognizing episodic events with the down-weighting technique. Another forest fire event was observed in the Prairie Provinces in early June 2002. This event was also observed in Fig. 4A but not in Fig. 3A. The potential sources that could contribute to the elevated nitrate levels include agriculture and livestock (Environment Canada, 2003). Ammonia, produced by the hydrolysis of urea from livestock could react with acidic compounds in the atmosphere to produce ammonium nitrate (Pain and Jarvis, 1999). Significant areas highlighted in the nitrate PM2.5 mass concentration PSCF plot of Fig. 4B, were agricultural and farming regions in Canada and Michigan. The total acreage of farmland in south-western Ontario was greater than that of Michigan (USDA, 1997; OMAF, 2002). However, Fig. 4A indicated a higher PSCF PM2.5 mass concentration in Michigan than south-western Ontario. We suggest that the particles originating from Michigan were coated with nitrate during their transport through south-western Ontario. Hence, the geographic origins of PM2.5 were highlighted in the nitrate PM2.5 mass concentration PSCF plot, not the locations of the farms and livestock regions where suggested coating of the PM2.5 occurred. The nitrate PM2.5 mass concentration PSCF plot in Fig. 3B instead suggested that the particle emission source was primarily Ontario. This matter needs to be studied further as there is no reason to assume one explanation is better than the other. Finally, ammonium nitrate in Toronto occurs as a result of condensation onto particles at lower temperatures (Tan et al., 2002). Consequently nitrate PM2.5 source regions tended to occur north of the total PM2.5 source regions. Furthermore, the Prairie fires event was more evident in Fig. 4B than in Fig. 4A, while it was completely absent in Fig. 3B. This suggested that the forest fire particles originating in the Prairies were coated with nitrate during their transport to Toronto. The coating likely occurred over south-western Ontario. This hypothesis is supported by the lack of high PSCF values for the Quebec area in Fig. 4B. Since the Quebec forest fire particles did not travel over farmlands till their arrival in Toronto, no nitrate coating would be expected.

635

7. Conclusion The counting technique described in this paper provided an objective method to illustrate potential source regions of pollutants measured in Toronto. A significant improvement was apparent upon comparing conventional PSCF plots with those created with the counting technique, and also upon comparison with known locations of power plants and agricultural regions in southern Ontario, Michigan and Ohio. Unlikely potential emission sources regions were eliminated using the counting technique. Potential emission sources during the episodic events of Quebec and Prairie forest fires were also identified with this technique.

Acknowledgements The authors thank the Natural Sciences and Engineering Research Council (NSERC), Environment Canada, and Toxic Substance Research Initiative, a research program held jointly by Health Canada and Environment Canada, for funding to construct the University of Toronto Facility for Aerosol Characterization.

References Brown, S.G., Rafner, H.G. Shields E., 2004. Apportionment of air toxics data using principal component analysis. Workshop on Air Toxics Data Analysis, Rosemont, IL. Accessed May 2005. . Cheng, M.D., Hopke, P.K., Zeng, Y., 1993. A receptor-oriented methodology for determining source regions of particulate sulfate observed at Dorset. Ont. J. Geophys. Res. 98, 16839–16849. Draxler, R.R., 1992. Hybrid single-particle Lagrangian integrated trajectories (HY-SPLIT): Version 3.0—Users guide and model dispersion, NOAA Tech. Memo. ERL ARL-195, p. 26 and appendices. Environment Canada, 2003. National Pollutant Release Inventory (NPRI) Data Search. Accessed May 2003. . Gao, N., Cheng, M.D., Hopke, P.K., 1993. Potential source contribution function-analysis and source apportionment of sulfur species measured at Rubidoux, CA during the southern California air-quality study, 1987. Anal. Chim. Acta 277, 369–380. Goruganthu, R., Coplan, M., Moore, J., Tossell, J., 1988. (e, 2e) Momentum spectroscopic study of the interaction of CH3 and CF3 groups with the carbon–carbon triple bond. J. Chem. Phys. 89, 25–33.

636

S. Owega et al. / Ecological Modelling 192 (2006) 627–636

Graham, L.A., Desjardins, S.J., Bawagan, A.D.O., 1993. Coincidence electron-scattering experiments—the statistic of coincidence counting. Can. J. Chem. 71, 216–226. Hopke, P.K., Li, C.L., Ciszek, W., Landsberger, S., 1994. The use of bootstrapping to estimate conditional probability fields for source locations of airborne pollutants. InCINC’94—The First International Chemometrics InterNet Conference: General Session. Accessed May 2003. . Ontario Minstry of Agriculture and Food (OMAF), 2002. Farm land area classified by use of land, by county, 2001 (acres). Accessed July 2003.. Pain, B., Jarvis, S., 1999. Ammonia emissions from Agriculture, Institute of Grassland and Environmental Research. Accessed May 2003. . Polissar, A.V., Hopke, P.K., Poirot, R.L., 2001a. Atmospheric aerosol over Vermont: chemical composition and sources. Environ. Sci. Technol. 35, 4604–4621. Polissar, A.V., Hopke, P.K., Harris, J.M., 2001b. Source regions for atmospheric aerosol measured at Barrow. Alaska. Environ. Sci. Technol. 35, 4214–4226.

Tan, P.V., Evans, G.J., Tsai, J., Owega, S., Fila, M., Malpica, O., Brook, J.R., 2002. On-line analysis of urban particulate matter focusing on elevated wintertime aerosol concentrations. Environ. Sci. Technol. 36, 3512–3518. Tremoulet, G. (Ed.), 1999. Ash Library, University of Kentucky Center for Applied Energy Research International Ash Utilization Symposium. Accessed May 2003. . Tsai, J., Owega, S., Evans, G.J., Jervis, R., Tan, P., Fila, M., Malpica, O., 2004. Chemical composition and source apportionment of Toronto summertime urban fine aerosol (PM2.5 ). J. Radioanal. Nucl. Chem. 259, 193–197. United States Department of Agriculture (USDA), 1997. Census of Agriculture Volume 1: National, State and County Tables. Accessed July 2003. . Zeng, Y., Hopke, P.K., 1989. A study of the sources of acid precipitation in Ontario, Canada. Atmos. Environ. 23, 1499– 1509. Zeng, Y., Hopke, P.K., 1994. Comparison of the source locations and seasonal patterns for acidic species in precipitation and ambient particles in south Ontario, Canada. Sci. Total Environ. 143, 245–260.