FINAL REPORT
September 24, 2003
EIGHT-SITE SOURCE APPORTIONMENT OF PM2 5 SPECIATION TRENDS DATA
Contract No. 68-D-02-061
Work Assignment 1-05
for
Vickie Presnell, Project Officer
Doug Solomon, Work Assignment Manager and
Ellen Baldridge, Alternate Work Assignment Manager
Office of Air Quality Planning and Standards
Emissions, Monitoring, and Analysis Division
U.S. ENVIRONMENTAL PROTECTION AGENCY
Research Triangle Park, North Carolina 27711
Prepared by:
Basil W. Coutant, Christopher H. Holloman, and Kristen E. Swinton
BATTELLE
505 King Avenue
Columbus, Ohio 43201-2693
and
Hilary R. Hafner
SONOMA TECHNOLOGY, INC.
1360 Redwood Way, Suite C
Petaluma, California 94954-1169
Eight-Site SA Speciation Trends Final Report September 24, 2003
-------
September 24, 2003
Ms. Vickie Presnell
Project Officer
Office of Air Quality Planning & Standards
Emissions, Monitoring and Analysis Division (C304-02)
Office of Air Quality Planning and Standards
U.S. Environmental Protection Agency
Research Triangle Park, North Carolina 27711
Dear Ms. Presnell:
EPA Contract No. 68-D-02-061, Work Assignment 1-05
Please find enclosed one copy of the final report "Eight-Site Source Apportionment of PM2.5
Speciation Trends Data." This report covers the source apportionment and back trajectory
analyses done for Bronx, St. Louis, Houston, Washington, D.C., Milwaukee, Birmingham,
Charlotte, and Indianapolis. It incorporates EPA comments on the April 30, 2003, draft and
additional analyses. This serves as a deliverable under the subject work assignment.
If you have any questions, please call me at 614/424-7487 or Basil Coutant at 614/424-6538.
Sincerely,
Steven M. Bortnick, Ph.D.
Senior Research Scientist
Statistics and Data Analysis Systems
BWC/SMB:llj
Enclosure
cc: Otelia Newsome (epaco) [letter only]
Doug Solomon (wam) [i copy]
Ms. Ellen Baldridge (Alternate WAM) [lcopy]
-------
BATTELLE DISCLAIMER
This report is a work prepared for the United States Environmental Protection
Agency by Battelle Memorial Institute. In no event shall either the United States
Environmental Protection Agency or Battelle Memorial Institute have any
responsibility or liability for any consequences of any use, misuse, inability to
use, or reliance upon the information contained herein, nor does either warrant or
otherwise represent in any way the accuracy, adequacy, efficacy, or applicability
of the contents hereof.
Eight-Site SA Speciation Trends Final Report
ii
September 24, 2003
-------
TABLE OF CONTENTS
Page
ACKNOWLEDGEMENTS vii
EXECUTIVE SUMMARY viii
1.0 INTRODUCTION AND BACKGROUND 1
2.0 DATA 1
2.1 Sources of the Data 1
2.2 Site Selection and Site Characteristics 3
2.2.1 Birmingham, Alabama 3
2.2.2 Bronx, New York 3
2.2.3 Charlotte, North Carolina 4
2.2.4 Houston, Texas 4
2.2.5 Indianapolis, Indiana 4
2.2.6 Milwaukee, Wisconsin 4
2.2.7 St. Louis, Missouri 5
2.2.8 Washington, D.C 5
2.3 Species Selection 5
2.4 Data Screening 6
2.5 Local Meteorological Data 6
3.0 SOURCE APPORTIONMENT PROCEDURES 7
3.1 Preliminary Procedures 7
3.2 Overview of PMF 10
3.3 Species Modeled 11
3.4 Analyses of the Residuals 12
4.0 IDENTIFYING THE SOURCES 16
4.1 Automated Matching of the Source Apportionment Output 16
4.2 Guidelines for Assigning Preliminary Identifications 17
4.3 Final Identifications 18
5.0 RESULTS OF THE SOURCE APPORTIONMENT ANALYSIS 20
5.1 The Birmingham, Alabama, Site 21
5.2 The Bronx, New York, Site 22
5.3 The Charlotte, North Carolina, Site 23
5.4 The Houston, Texas, Site 24
5.5 The Indianapolis, Indiana, Site 26
5.6 The Milwaukee, Wisconsin, Site 27
5.7 The St. Louis, Missouri, Site 28
5.8 The Washington, D.C., Site 29
6.0 BACK TRAJECTORY ANALYSIS AND ANALYSIS WITH LOCAL
METEOROLOGICAL DATA 30
6.1 Pollution Roses 30
6.2 Temperature and Pressure Comparisons 32
6.3 Day of Week and Season Summaries of the Source Strength 37
6.4 Back T rajectory Analyses 40
6.5 Comparisons with NOx and S02 Utility Plant Inventory Data 45
Eight-Site SA Speciation Trends Final Report iii September 24, 2003
-------
7.0 SITE-BY-SITE RESULTS 57
7.1 Birmingham, Alabama 58
7.2 Bronx, New York 59
7.3 Charlotte, North Carolina 60
7.4 Houston, Texas 62
7.5 Indianapolis, Indiana 63
7.6 Milwaukee, Wisconsin 64
7.7 St. Louis, Missouri 65
7.8 Washington, D.C 66
8.0 INTER-SITE ANALYSES 68
9.0 CONCLUSIONS 68
10.0 REFERENCES 73
APPENDIX A: DATA MANIPULATION A-1
APPENDIX B: MATCHING THE SOURCE APPORTIONMENT OUTPUT TO
SPECIATE PROFILES B-1
APPENDIX C: NUMERICAL SOURCE PROFILES C-1
APPENDIX D: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR BIRMINGHAM, ALABAMA D-1
APPENDIX E: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR BRONX, NEW YORK E-1
APPENDIX F: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR CHARLOTTE, NORTH CAROLINA F-1
APPENDIX G: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR HOUSTON, TEXAS G-1
APPENDIX H: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR INDIANAPOLIS, INDIANA H-1
APPENDIX I: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR MILWAUKEE, WISCONSIN 1-1
APPENDIX J: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR ST. LOUIS, MISSOURI J-1
APPENDIX K: GRAPHICAL REPRESENTATION OF THE SOURCE APPORTIONMENT
RESULTS FOR WASHINGTON, D.C K-1
APPENDIX L: THE USE OF THE BIC FOR SELECTION OF THE NUMBER OF SOURCES L-1
APPENDIX M: RESPONSE TO REVIEW COMMENTS ON THE APRIL 30, 2003,
DRAFT REPORT M-1
List of Tables
Table 2.1 Prevalent Species and the Percent of the Data Above the MDL 2
Eight-Site SA Speciation Trends Final Report
iv
September 24, 2003
-------
Table 2.2 Dates Modeled for Each of the Eight Sites 3
Table 2.3 Nearest NOAA Meteorological Station 7
Table 3.1 Summary of the PMF Q Values 13
Table 3.2 Model Error Estimates 16
Table 5.1 Summary of Source Identification for Birmingham, Alabama 21
Table 5.2 Summary of Source Identification for Bronx, New York 22
Table 5.3 Summary of Source Identification for Charlotte, North Carolina 23
Table 5.4 Summary of Source Identification for Houston, Texas 24
Table 5.5 Summary of Source Identification for Indianapolis, Indiana 26
Table 5.6 Summary of Source Identification for Milwaukee, Wisconsin 27
Table 5.7 Summary of Source Identification for St. Louis, Missouri 28
Table 5.8 Summary of Source Identification for Washington, D.C 29
Table 6.1 Pollution Rose Summaries 31
Table 6.2 Summary of Source Strength Correlation with Temperature 33
Table 6.3 Summary of Source Strength Correlation with Pressure 35
Table 6.4 Summary of High Source Strength Periods for Weekdays and Weekends 37
Table 6.5 Summary of High Source Strength Periods for Seasons 39
Table 6.6 Summary of the Back Trajectory Plots 43
Table 7.1 Summary of the Birmingham, Alabama, Results 58
Table 7.2 Summary of the Bronx, New York, Results 59
Table 7.3 Summary of the Charlotte, North Carolina, Results 60
Table 7.4 Summary of the Houston, Texas, Results 62
Table 7.5 Summary of the Indianapolis, Indiana, Results 63
Table 7.6 Summary of the Milwaukee, Wisconsin, Results 64
Table 7.7 Summary of the St. Louis, Missouri, Results 65
Table 7.8 Summary of the Washington, D.C., Results 67
Table 9.1 Summary of the Mean Apportioned Mass Across Sites 72
List of Figures
Figure 3.1 Aluminum versus Silicon Concentrations in the Houston Area 9
Figure 3.2 Iron versus Silicon in the St. Louis Area 9
Figure 3.3 Calcium versus Iron in the St. Louis Area 10
Figure 3.4 Q-Q Plot of the Scaled Residuals of the FRM Data for St. Louis 14
Figure 3.5 Q-Q Plot of the Scaled Residuals of the Speciation Monitor Mass Data for
St. Louis 14
Figure 6.1 Utility Plant Emissions of S02 46
Figure 6.2 Utility Plant Emissions of NOx 46
Figure 6.3 Nitrate Source Region Plot for Source 1, Ammonium Nitrate at
Birmingham, Alabama 48
Figure 6.4 Nitrate Source Region Plot for Source 6, Ammonium Nitrate, at
Bronx, New York 48
Figure 6.5 Nitrate Source Region Plot for Source 6, Ammonium Nitrate, at
Charlotte, North Carolina 49
Figure 6.6 Nitrate Source Region Plot for Source 5, Marine Ammonium Nitrate source, at
Houston, Texas 49
Figure 6.7 Nitrate Source Region Plot for Source 2, Ammonium Nitrate, at
Indianapolis, Indiana 50
Figure 6.8 Nitrate Source Region Plot for Source 5, Ammonium Nitrate, at
Milwaukee, Wisconsin 50
Figure 6.9 Nitrate Source Region Plot for Source 5, Ammonium Nitrate, at
St. Louis, Missouri 51
Figure 6.10 Nitrate Source Region Plot for Source 3, Ammonium Nitrate and Salt, at
Washington, D.C 51
Figure 6.11 Sulfate Source Region Plot for Source 7, Coal Combustion (Ni), at
Eight-Site SA Speciation Trends Final Report
v
September 24, 2003
-------
Birmingham, Alabama 52
Figure 6.12 Sulfate Source Region Plot for Source 1, Coal Combustion, at
Bronx, New York 52
Figure 6.13 Sulfate Source Region Plot for Source 2, Coal Combustion, at
Charlotte, North Carolina 53
Figure 6.14 Sulfate Source Region Plot for Source 7, Coal Combustion (Ni), at
Houston, Texas 53
Figure 6.15 Sulfate Source Region Plot for Source 7, Coal Combustion 1, at
Indianapolis, Indiana 54
Figure 6.16 Sulfate Source Region Plot for Source 8, Coal Combustion 2 (Ni), at
Indianapolis, Indiana 54
Figure 6.17 Sulfate Source Region Plot for Source 1, Coal Combustion (Ni), at
Milwaukee, Wisconsin 55
Figure 6.18 Sulfate Source Region Plot for Source 8, Industrial Diesel and Sulfate Mix, at
Milwaukee, Wisconsin 55
Figure 6.19 Sulfate Source Region Plot for Source 3, Coal Combustion, at
St. Louis, Missouri 56
Figure 6.20 Sulfate Source Region Plot for Source 2, Coal Combustion, at
Washington, D.C 56
Eight-Site SA Speciation Trends Final Report
vi
September 24, 2003
-------
ACKNOWLEDGEMENTS
The authors wish to sincerely thank the following people for providing a local
perspective for each of the sites.
Aaron Childs, Indiana Public Works - Air Pollution Control Division
Robert D. Day, Washington, D.C., Department of Health
Randy Dillard, Jefferson County (AL) Health Department
Dirk Felton, New York State, Department of Environmental Conservation
Jeff Francis, Mecklenburg County (NC) Air Quality
David Krask, Washington, D.C., Department of Health
Calvin Ku, Missouri Department of Natural Resources
Ed Michel, Texas Commission on Environmental Quality
Ed Miller, Wisconsin Department of Natural Resources
Terry Rowles, Missouri Department of Natural Resources
Ram Tangirala, Washington, D.C., Department of Health
The authors further wish to acknowledge the valuable input and comments from the
following individuals:
Katherine Brehme, DynCorp
Stephanie Dickinson, formerly of Battelle
Thomas Kelly, Battelle
Donna Kenski, LADCO
Rich Poirot, VT DEC
Chet Spicer, Battelle
Brandon Wood, Battelle
Finally, we wish to thank Brian Orndorff and Jim Szykman both of EPA for running the
HYSPLIT model to generate the back trajectories.
Eight-Site SA Speciation Trends Final Report
vii
September 24, 2003
-------
-------
EXECUTIVE SUMMARY
This source apportionment and back trajectory study analyzes speciated PM2.5 data from
eight of EPA's Trends Sites located in Birmingham, Alabama; Bronx, New York;
Charlotte, North Carolina; Houston, Texas; Indianapolis, Indiana; Milwaukee, Wisconsin;
St. Louis, Missouri; and Washington, D.C. Unlike previous studies of IMPROVE and
CASTNET data, these sites are in urban areas that are expected to include strong local effects as
well as effects from long-range transport. The results of both the source apportionment and back
trajectory analyses are consistent with this expectation.
This report covers the results and methods used to apportion the data into the major
sources of the PM2.5. It also covers the methods used to identify those sources on the basis of the
apportioned chemical characteristics. The methods applied are somewhat different from the
methods used in previous source apportionment work of IMPROVE and CASTNET sites. The
screening criteria used were much less stringent to allow more data to be used, since the data
cover a significantly shorter time period. At the same time, the model fitting criteria were more
stringent to protect against inappropriate model results. One important consequence of the
differences in the methods is that the methods used here identify relatively infrequent sources,
such as fireworks, while the data screening frequently used in source apportionment studies are
in part designed to exclude those sources. The development of the back trajectories is
documented and slightly extends the methods developed in previous studies. The extra step in
the analysis of the back trajectories ensures that the scales are comparable across sites. Analyses
of the source strengths with respect to various meteorological data are also included as a part of
developing an understanding of the sources.
While the combination of source apportionment techniques, local meteorological
analysis, and back trajectory methods provide a very useful means of understanding the PM2.5
sources, there are some limitations:
• Sufficient data are needed with a sufficient number of measured species that are
observed at levels above the MDL. The data available did not allow the mobile
sources to be apportioned into separate diesel and non-diesel components.
• The wind and pollution roses are based on low-level winds from "nearby" weather
stations. These can be highly variable within an urban area. Even co-located wind
information can be misleading if interpreted too literally.
• The back trajectory methods require careful interpretation and need to have as many
reality checks as possible. They are based on modeling back trajectories of air
packets that start at 500 m above the site and use gridded meteorological data that
have a three-hour time resolution and 80 km grid cells. Confounding factors, such as
sources and data that are dependent on meteorological conditions, can lead to
incorrect conclusions. Further, local sources may be missed entirely by these
methods because of the spatial resolution of the data.
Eight-Site SA Speciation Trends Final Report
viii
September 24, 2003
-------
• Different sites have differing time periods over which speciated data were available.
As a result, some sites may have more warm seasons or cold seasons represented than
other sites. This unequal representation of seasons may result in overstatement of the
contribution from a seasonal source when that source's season is overrepresented.
Hence, it is necessary to use a weight-of-evidence approach to understanding the results with as
many independent checks of the conclusions as possible and careful checks on the modeling.
The methods applied within the study started with receptor modeling and careful
checking of the data, the residuals, and the internal consistency of the profiles. Preliminary
source identifications were based first on the chemical composition of the profiles. These were
then balanced against the relative contribution of the source to the various species and the time
series output. Second, a reality check was made by contacting local monitoring personnel to
discuss their expectations of the sources of the PM measured at the receptor. In the case of
Bronx, this led to using a different dataset because of additional QA review that had been given
to the final data used. It also resulted in an extensive list of possible sources for some sites that
was supplemented by examining other inventories. Third, back trajectories were used to identify
source locations for sources that are 3 to 72 hours upwind. Pollution roses were used to identify
source directions from local winds. Attempts were made to verify that local point sources are at
least approximately in the directions indicated. Other analyses using the local meteorological
data are included for further confirmation. The final source identifications are based on all the
available information. Because there were attempts to confirm the results at various points, the
process was not linear.
For each site, the PM2.5 was apportioned into six to eight sources. While the species were
chosen to be consistent across the sites, the number of sources used in the modeling was allowed
to vary between sites. Eight may be a limit of the model for the amount of data that were
available. There were several commonly identified sources. Each of these source categories was
expected to affect the receptor:
• For each site, a coal combustion source was identified with a mean mass of between
4.5 and 7.7 |ig/m3. These sources are the main sources of sulfur/sulfate for each site.
They also include selenium that is associated with coal burning. Some of these
sources also have enhanced nickel content compared to the coal combustion profiles
found at rural sites. This may mean that some oil burning has been apportioned to
these sources. However, it may not. There is a preliminary indication from transport
analyses that some of the trace metals may be preferentially removed from the PM2.5
fraction resulting in relatively lower concentrations further from the source. The back
trajectory analyses for these sources are somewhat mixed. The back trajectory
analysis corresponds well to the utility plants in the Midwest, Southeast, and eastern
seashore. To some extent in St. Louis, and to a greater extent in Houston, the high
concentrations of sulfate are partially related to the effects of high pressure systems
located to the north and east of the site.
• For each site, a mobile source was identified with a mean mass of 2.5 to 6.5 (ig/rn3.
For Houston, in addition to the main mobile source with a mass of 5.2 (ig/m3, there
Eight-Site SA Speciation Trends Final Report
ix
September 24, 2003
-------
was a source with a mean mass of 1.0 (ig/m3 that may be mobile related. This
additional source is high in OC (organic carbon, usually associated mobile sources)
and with significant amounts of Mn (sometimes associated with off-road diesel from
the additive MMT). However, this source could be grain dust with a Mn-based
antifungal coating from the ship channel. Further refinement of the carbon sources
would benefit all of the sites, but particularly the Houston site. Finally, the profile for
the mobile source in St. Louis contains an unusually high amount of lead (for current
mobile sources) that is likely related to a historical problem with lead in the area.
• Each site also had a small crustal dirt source with a mean mass between 0.3 (ig/rn3
and 1.5 (ig/m3. The 1.5 (ig/m3 source is for Washington, D.C., which also contains
diesel components and is probably tied to a large road construction project under way
during the period modeled. For St. Louis, the crustal material may be supplemented
by point sources such as cement manufacturing.
• Houston had a very small nitrate source that was associated with a marine profile.
The other sites had nitrate sources that ranged from 1.2 to 5.0 (ig/m3. For the sites
other than Houston, the back trajectories indicate Midwestern source regions that
would be associated with agricultural ammonia emissions. Illinois, in particular,
stands out among the source regions. This should be expected, since Illinois has both
NOx utility emissions and the farming regions for sources of ammonia.
• Bronx, Charlotte, Houston, and Indianapolis each had small sea marine and industrial
salt sources. The largest is for Indianapolis, but the source profile shows signs of
nitrate substitution for the chlorine during transport.
• A source clearly dominated by fireworks was found for Birmingham, Charlotte,
Houston, Indianapolis, Milwaukee, and Washington, D.C. These sources are all very
similar in size (-0.5 (ig/m3 ) except for Birmingham, which is twice as large as the
others (1.2 (ig/m3 ). Because of the similarities in the source profiles to vegetative
burning, these sources should include any vegetative burning in the areas. The
source name, "Vegetative burning and fireworks," was chosen to reflect the more
frequent of the two sources.
• Sources that appear to be related to industrial activity were found in Birmingham,
Bronx, Milwaukee, Houston, and St. Louis.
• Both Bronx and Charlotte had oil combustion sources with masses of 1.2 (ig/m3 and
1.9 (ig/m3 respectively.
• Charlotte and St. Louis had zinc sources with each having masses of 0.9 (ig/m3. The
pollution rose for the St. Louis zinc source is consistent with a local zinc refinery. In
addition, St. Louis had a copper smelting (0.6 (ig/m3 ) and steel production
(0.8 (ig/m3) source.
Eight-Site SA Speciation Trends Final Report
x
September 24, 2003
-------
• Finally, there was a huge spike in the PM2.5 mass on July 7, 2002, in
Washington, D.C., that is associated with Canadian forest fires. This source is
apportioned over 1 (ig/rn3 of the 16.6 (ig/rn3 of mass observed during the modeled
period. The Indianapolis site was also affected by these fires, but to a much lesser
extent.
As indicated above, the back trajectory analyses and wind/pollution roses for the sites
yield source location information for the apportioned source categories. There had been some
concern that the back trajectories would not work for nitrate sources, but rather just show an
association with cool air from the north. The multiple sites within this study show that while this
might be true to some extent, comparisons of the back trajectory contour maps of the various
non-marine nitrate sources show a very common pattern of association. The nitrate sources are
associated with the Midwest farming regions.
The comparisons of the coal combustion source regions with the SO2 utility emissions
did not work as well as expected. For some of the sites, the Bronx site for instance, the back
trajectories do yield the expected source region associations with large utility emissions of SO2,
namely the Ohio River Valley and the borders of Ohio, West Virginia, and Pennsylvania.
Further complicating the analysis for the sulfate sources is that some seem to be related more to
high pressure systems (as evidenced by the clockwise swirl of many of the back trajectories for
the high source days). With additional data, it should be expected that the tools would separate
the coal combustion sources into separate meteorological regimes, as in the case of Indianapolis
and other IMPROVE sites.
The various analyses are generally self-consistent, consistent among analysis types,
consistent with expectations for the sites, and consistent from site-to-site. Taken together they
show that a monitoring and modeling combination provides an effective means of understanding
the source categories affecting urban areas. The coal combustion sources account for about
one-third of the PM2.5. The next largest portion is either from nitrate or mobile sources. All
three of these source categories show transport components. Additional study of the mobile
sources could be beneficial through the addition of VOCs, speciated PM carbon data, or finer
carbon fractions in the source apportionment. After the three main sources, the smaller sources
are more site-specific except for crustal dust. The ability to separate and identify these is likely
to be data dependent. Up to eight sources that can include marine influences, metal production,
general industrial, and oil combustion are within the range of resolvability with approximately
one year of speciation data at current levels of technology. Additional source resolution should
be possible with longer data streams or additional carbon species.
Any mention of explicit sources within the source identifications is included only as an
example of a local source with characteristics similar to what the study has found. Additional
analysis would be needed to relate an effect at the receptor to an explicit source.
Eight-Site SA Speciation Trends Final Report
xi
September 24, 2003
-------
1.0 INTRODUCTION AND BACKGROUND
EPA has promulgated a National Ambient Air Quality Standard (NAAQS) for PM2.5, fine
particulate matter, which the agency has determined is needed to protect the public's health. Due
to this increased awareness of PM2.5 health related problems and regional haze problems, there
has been a dramatic increase in ambient air quality monitoring of PM2.5 mass and its chemical
composition. Concentrations of PM2.5 exist in the ambient air as a composition of chemical
species originating from natural and manmade emissions that may be transported thousands of
kilometers from their origins. Development of efficient emission control strategies to lower
PM2.5 ambient concentrations to below the health standards (e.g., a three-year annual average of
15 ug/m3) can be aided by determining the relationship between the various types of emissions
sources and elevated levels of PM2.5 at ambient monitoring sites. This study uses Positive Matrix
Factorization (PMF) as the main tool for identifying potential sources. Moreover, the output of
the source apportionment is combined with an air mass history analysis (ensemble back
trajectory analyses) to associate the location and transport distances of the air mass with
dominant sources. Additional comparisons with local meteorological data are also included to
support the source identification process. A primary purpose of this project is to develop a better
scientific foundation for performing source apportionment and the associated conditional
ensemble back trajectory analyses by applying these methods to data available from the PM2.5
Chemical Speciation network. A second major purpose for this study is to provide initial data
analysis for understanding PM2.5 transport through multi-site analyses. This report focuses on
the first aspect.
2.0 DATA
The source apportionment results presented in this report are based on speciated PM2.5
measurements. The daily measurements are from integrated 24-hour collection periods using
filter-based methods. Specifically, the PM2.5 speciation sites use X-Ray Fluorescence (XRF),
Ion Chromatography (IC), and Thermal-Optical Analysis (TOR) analyses done on Teflon, nylon,
and quartz filters, respectively. Generally, 50+ parameters are measured; however, some of
those are never detected at some sites. The sections below discuss the data sources, data issues,
site selection, and species selection.
2.1 Sources of the Data
The initial data for the project were for Bronx, St. Louis, and Houston and came from the
AQS database, http://www.epa.gov/ttnairs 1/airsaqs/index.htm, in January 2002. This was
supplemented with data from the New York Department of Environmental Conservation website,
http://www.dec.state.ny.us/website/dar/baqs/ pm25mon.html. A summary of this data is
provided in Table 2.1. This table shows the species examined, the number of days with
non-missing data, and the percentage of days when the observation was above the MDL. The
results shown for the Bronx site are based on the data from this website (the data run through
January 2002 rather than September 2001). For consistency among the results throughout the
project, the species modeled were based on this initial summary. AQS data for Milwaukee and
Washington, D.C., were obtained in September 2002 and the AQS data for Birmingham,
Charlotte, and Indianapolis were added in January 2003. The uncertainty estimates for all the
Eight-Site SA Speciation Trends Final Report
1
September 24, 2003
-------
sites are based in part on the co-located data within the original AIRS database (commonly
referred to as the Mini-Trends sites).
Table 2.1 Prevalent* Species and the Percent of the Data Above the MDL
Species
Bronx
Houston
St. Louis (City)
Days
Percent >MDL
Days
Percent >MDL
Days
Percent >MDL
Aluminum
161
46.58
131
47.33
111
56.76
Ammonium
156
99.36
123
95.93
113
100.00
Arsenic
161
14.91
131
60.31
111
73.87
Barium
161
27.95
131
49.62
111
45.95
Bromine
161
64.60
131
90.84
111
93.69
Calcium
161
100.00
131
100.00
111
100.00
Chlorine
161
29.81
131
50.38
111
60.36
Chromium
161
29.81
131
44.27
111
66.67
Cobalt
161
14.29
131
0.76
111
0.00
Copper
161
78.26
131
90.84
111
100.00
Elemental Carbon
156
100.00
125
97.60
98
100.00
Gallium
161
14.29
131
13.74
111
9.01
Iron
161
100.00
131
99.24
111
100.00
Lead
161
59.01
131
82.44
111
98.20
Magnesium
161
24.22
131
22.14
111
15.32
Manganese
161
42.86
131
87.79
111
95.50
Nickel
161
100.00
131
70.99
111
66.67
Nitrate
157
100.00
123
99.19
113
100.00
OCX
156
100.00
22
100.00
28
100.00
OCX2
0
NA
40
100.00
44
100.00
Organic Carbon
156
100.00
125
100.00
98
100.00
Phosphorus
161
19.88
131
0.00
111
3.60
PM2.5
157
100.00
131
100.00
111
100.00
Potassium
161
100.00
131
94.66
111
100.00
Potassium Ion
157
63.69
123
90.24
113
67.26
Selenium
161
27.33
131
32.82
111
61.26
Silicon
161
100.00
131
98.47
111
100.00
Sodium
161
67.08
131
70.23
111
54.05
Sodium Ion
157
97.45
16
93.75
113
91.15
Strontium
161
10.56
131
12.98
111
22.52
Sulfate
157
100.00
123
100.00
113
100.00
Sulfur
161
100.00
131
94.66
111
100.00
Tantalum
161
44.10
131
38.17
111
30.63
Tin
161
49.07
131
58.78
111
68.47
Titanium
161
96.27
131
97.71
111
99.10
Vanadium
161
98.14
131
76.34
111
46.85
Zinc
161
100.00
131
93.89
111
100.00
* Only species that were above the MDL at least ten percent of the time for at least
one site are shown.
Eight-Site SA Speciation Trends Final Report
2
September 24, 2003
-------
The time periods covered by the data at each site are not the same. Some monitor
readings begin in 2000 while others are not recorded until 2001. Also, some monitor readings at
some sites end in 2001 while others end in 2002. Table 2.2 summarizes the time periods over
which monitor readings were recorded at each of the eight sites.
Table 2.2 Dates Modeled for Each of the Eight Sites
Site
Start Date
End Date
Days
Modeled
Sampling
Frequency
Birmingham, AL
1/13/2001
8/9/2002
186
1 -in-3 day
Bronx, NY
9/3/2000
1/29/2002
160
1 -in-3 day
Charlotte, NC
1/13/2001
8/6/2002
143
1 -in-3 day
Houston, TX
8/17/2000
7/7/2001
121
1 -in-3 day/daily
Indianapolis, IN
12/20/2000
8/6/2002
155
1 -in-3 day
Milwaukee, Wl
12/14/2000
9/8/2002
172
1 -in-3 day
St. Louis, MO
8/4/2000
7/12/2001
112
1 -in-3 day
Washington, DC
4/7/2001
8/6/2002
124
1 -in-3 day
2.2 Site Selection and Site Characteristics
The Bronx, St. Louis, and Houston sites were selected from among the urban speciation
Trends Network to be representative of urban sites around the nation. These sites were also
chosen to supplement the source apportionment work being done in other studies, namely the
source apportionment of various IMPROVE and CASTNET sites in the northeast for
MARAMA; work done in Portland, Oregon; and application to six Midwestern sites by LADCo.
The remaining five sites were chosen to ensure coverage in the eastern portion of the
United States.
2.2.1 Birmingham. Alabama
The Birmingham site (010730023) is located in an urban neighborhood in a heavily
industrialized area of the city. A U.S. Pipe Plant is located 1/4 mile east and northeast of the site.
A Sloss Industries Coke Plant and a Slag Wool Plant are located 3/4 mile to the north and 1 mile
northeast, respectively. Finally, an American Cast Iron Pipe Plant is located about 2 miles
west-southwest of the site. Diesel trains and equipment are located south, southeast, east, and
northeast of the site. The nearest major roadway is about 30 meters away. Natural gas is the
main fuel for heating, and coal is the main fuel for electricity for the area.
2.2.2 Bronx. New York
The Bronx site (360050083) is located in the middle of the Bronx, a heavily populated
urban area. There are local sources that could potentially have a significant effect on the site.
These include mobile emissions, fuel oil (particularly in the winter), two oil-fired power plants,
street cleaning, and marine influence.
Eight-Site SA Speciation Trends Final Report
3
September 24, 2003
-------
2.2.3 Charlotte. North Carolina
The Charlotte site (371190041) is located on the campus of Garinger High School, at
1100 Eastway Drive. The area surrounding the school is primarily residential, but contains some
commercial land uses that would be associated with densely populated residential areas
(convenience stores, restaurants, and other small businesses) near intersections along the main
thoroughfares. The area also contains some light industrial land uses within relatively close
proximity.
Probably the largest nearby source is a concrete plant approximately 1.24 miles
north-northwest of the site. School buses would be a diesel source as they service the school and
are parked at the school. The buses are parked approximately 650 feet from the monitoring site.
There has been some construction at the school within the past 2 years. A major renovation of
the main school building was performed during the summer of 2001.
Fuels for heating are primarily gas and oil, but also include electric and some wood.
Electricity in Mecklenburg County is generated primarily by coal and nuclear fuels.
2.2.4 Houston. Texas
The Houston site chosen was the Aldine Road site (482010024). This site is not as
heavily impacted by the ship channel as other sites in the Houston area and, hence, should be
more representative of other urban areas around the nation. It was expected to be affected by
sources that would be associated with an urban area. In particular, mobile emissions should be
significant.
2.2.5 Indianapolis, Indiana
The Indianapolis site (180970078) is in a residential area that is northeast of the central
core of the city. The area is highly populated. The site is in a parking lot next to a police station
and a city park. There is some light industry in the area including a printing operation to the
south of the site. The main fuels are natural gas and oil-burning home heating furnaces.
Electricity is provided by power plants in the southern part of the city and state.
2.2.6 Milwaukee. Wisconsin
The Milwaukee site (550790026) is located on a wooden stand 4 feet off the ground on
the Southeast Region Headquarters' parking lot at 2300 North Dr. Martin Luther King Jr. Drive.
It is about 100 feet from the street. This street is a major north-south artery with high levels of
motor vehicle traffic. In addition, the site lies about 150 feet north of North Avenue, a major
east-west artery with traffic comparable to Dr. Martin Luther King Jr. Drive. The intersection of
these two major roads lies approximately 125 feet southwest of the monitor, so high PM2.5
contributions from cars idling, stopping, and accelerating are expected. A building separates the
monitor from the intersection. Finally, Interstate 43 (a north-south roadway) lies about
1,000 feet west of the monitor site. This roadway is subject to high motor vehicle traffic
Eight-Site SA Speciation Trends Final Report
4
September 24, 2003
-------
especially at certain times of the day. The surrounding area is primarily commercial and
residential. Natural gas is the most widely used fuel for cooking and heating.
2.2.7 St. Louis. Missouri
The St. Louis site is the Blair Street site (295100085). This site is located near the
intersection of several highways, so mobile emissions should be a major component. In fact,
major interstate highways and major traffic arteries are located within some peripheral areas of
the site. The interstate highways extend to the west of the site. There are several municipal
incinerators, a zinc smelter, a very large lead smelter, a steel mill, cement manufacturing, and
limestone quarrying in the area.
2.2.8 Washington, D.C.
The Washington, D.C., site (110010043) is the McMillan Site. It is located within a
fenced property that surrounds the McMillan Reservoir (a water storage facility for the
District of Columbia). The trailer is in the middle of a large field approximately 50 to 70 yards
east of the lake shore. Approximately 2.6 miles to the south is the U.S. Capitol.
There is a small municipal parking lot directly to the southwest of the trailer where
approximately 10 to 20 diesel vehicles owned by the Department of Public Works are parked. If
all these vehicles start up at the same time, a local microscale diesel event might be produced.
However, there is an R&P TEOM operating at the McMillan Site (30-minute time resolution),
and it has not seen any extreme peaks of mass.
North Capitol is the closest major street, which can have over 40,000 vehicles per day.
There are numerous highways serving the area.
The main fuels for the area are fuel oil and natural gas: (a) inside D.C., mostly fuel oil,
natural gas, and a small amount of coal, and (b) outside the District and within a 50-mile radius
are five coal-fired power generation facilities. Four facilities are to the southwest and southeast,
and one facility is to the northwest of the McMillan site.
There are steel and aluminum facilities 30 to 40 miles to the northwest in
Frederick County, Maryland.
The data may also be affected by a major highway construction project approximately
15 miles to the southwest.
2.3 Species Selection
The species modeled affect the results in two important ways. First, in order for PMF to
find a source, that source should be a significant contributor to at least one of the species being
fit. In fact, this was part of the criteria used in deciding how many sources should be used in the
modeling. Second, in order to identify a source in the output, tracer species, characteristic
species, or characteristic ratios between species are needed. Balancing both of these is the fact
that a sufficient amount of data above the MDL is needed to obtain meaningful results with
Eight-Site SA Speciation Trends Final Report
5
September 24, 2003
-------
respect to the species. However, it is not known at what point useful results are possible. This is
because the modeling is based on all of the species simultaneously with down-weighting of the
below MDL data. Frequently, species can be usefully included even when more than 50 percent
of the values are below the MDL, and it may be that the minimum necessary for useful inclusion
is an absolute value rather than a percentage.
Table 2.1 shows the species that were considered from the speciation samplers. To be
able to make comparisons across sites, the same species were used at all sites. Generally, any
species with at least 35 percent of the observations above the MDL was included. Selenium was
felt to be essential for identifying coal burning and, hence, was included despite having many
observations below the MDL. In addition, at the Bronx and St. Louis sites, co-located FRM
mass measurements were available. Specifically, the species used with PMF were PM2.5 (both
from the speciation monitor and a co-located FRM when available), sulfate, nitrate, ammonium,
Al, As, Ba, Br, Ca, CI, Cr, Cu, Elemental Carbon (EC), Fe, Pb, Mn, Ni, Organic Carbon (OC),
K, K+, Se, Si, Na, S, Sn, Ta, Ti, V, and Zn. The inclusion of both the mass measurements and
both the sulfur and sulfate measurements effectively doubles the weight given to these species
and provides a means for evaluating the error in the apportionment.
2.4 Data Screening
Screening for outliers generally takes one of two forms. First, data consistency checks,
such as a comparison of the reconstructed mass to the measured mass, can sometimes be applied
to identify inconsistent data. Alternatively, outliers can also be identified by comparisons across
time. This latter form requires a long series of measurements to build up a basis for the criteria
that reject data unusual for the site. This will typically remove the effects of infrequent sources,
such as fireworks, from the data. In the MARAMA source apportionment study (Coutant, 2002),
this type of screening was mainly used to eliminate data where the EC or OC measurements were
not consistent with trace metal measurements. Similar screening was attempted with this
project's data. However, unlike the MARAMA study, unusual trace metal-to-carbon ratios were
not clustered in the sense that an unusual Cu/EC ratio would not correspond to an unusual Fe/EC
ratio. Hence, these data were included in the source apportionment analysis. It may be that the
source make-up is much more varied around these urban sites or that data from several years are
required to effectively screen the data in this manner (or both). Since some of the more common
"consistency" checks, such as a bound on the anion/cation ratio, are usually based on a historical
record for the site, the only comprehensive data screening used was based on the measured mass
versus a reconstructed mass (=the sum of the nitrate, sulfate, ammonium, organic carbon, and
elemental carbon masses plus IMPROVE's soil concentration. (See Appendix A.) The only
exception to this was four days within the Washington, D.C., data when Sulfate/(3*sulfur) >1.5
or <0.5, or CI >0.6 jig/nr. These same conditions remove two unusual nitrate values and an
unusual EC value. The data for these four days were sufficiently different that the PMF model
was treating them as a separate source (or sources depending on the number modeled).
2.5 Local Meteorological Data
Local meteorological data were obtained for each site from the NOAA archives.
Table 2.3 indicates the site location and the distance to the nearest NOAA MET station with
sufficient data to use in the analysis.
Eight-Site SA Speciation Trends Final Report
6
September 24, 2003
-------
Table 2.3 Nearest NOAA Meteorological Station
Site
Site
Lat.
Site
Long.
Nearest Available Meteorological Station
WBAN
Number
MEt Station
Name
MET Station Location
Distance
(miles)
Birmingham, AL
33.55
-86.82
13876
Birmingham, AL
International Airport
25.6
Bronx, NY
40.87
-73.88
94741
Teterboro, NJ
Teterboro Airport1
25.9
Charlotte, NC
35.24
-80.79
13881
Charlotte, NC
Douglas International Airport
14.8
Houston, TX
29.90
-95.33
53910
Houston, TX
Hooks Memorial Airport
9.6
Indianapolis, IN
39.81
-86.11
53842
Indianapolis, IN
Eagle Creek Airpark
21.8
Milwaukee, Wl
43.06
-87.91
4840
Fond Du Lac, Wl
Fond Du Lac County Airport2
33.6
St. Louis, MO
38.66
-90.20
53904
St. Charles, MO
St. Charles Smart Airport
7.4
Washington, DC
38.92
-77.01
13743
Washington, DC
Ronald Reagan National Airport3
27.5
The latitude and longitude coordinates for the site suggest that this airport is closer than LaGuardia Airport.
The latitude and longitude coordinates for the site suggest that this airport is closer than Milwaukee Mitchell
Airport.
Second nearest used because of MET station data problems.
3.0 SOURCE APPORTION PROCEDURES
The goal is to apportion the mass concentrations into components attributable to the most
significant sources. To do this, it is assumed that individual sources will contribute to the species
mass concentrations at the receptor with fixed proportions between the various species. This
should be at least approximately true for most species and sources considered in this study. With
this assumption, if the data could be measured without error, then the data matrix would have a
rank equal to the number of sources. With the additional assumption that there are sufficient
periods for each source when it makes no significant contribution to the receptor mass of any
species, there is a unique decomposition of the data into a matrix of profiles and a matrix of
relative contributions. Because of the measurement error, the tools can detect only sources with
a significant contribution to one or more of the fitting species.
3.1 Preliminary Procedures
The first step in source apportionment is to examine plots of the data. Scatter plots of
concentrations of one species versus another were examined as a part of the site selection. These
plots show important information about the data. Plots that are nearly linear (see Figure 3.1 of
aluminum versus silicon for the Houston site) indicate that the significant sources produce these
species in the same ratio. It is likely that there is only one major source of the pair. The source
apportionment results apportion all of the aluminum and about 75 percent of the silicon to a
single source. Wedge-shaped plots indicate at least two major sources of the pair of species (see
Figures 3.2 and 3.3 of silicon versus iron and calcium versus iron, respectively, from the
St. Louis site). The edges of the plots are produced from the two major sources of the species
pair with the most disparate ratios between the two species. The source apportionment results
for St. Louis include a source with 50 percent of the iron and about 10 percent of the observed
Eight-Site SA Speciation Trends Final Report
7
September 24, 2003
-------
silicon and another source that is apportioned 75 percent of the silicon and about 20 percent of
the iron. If there are only two major sources, then frequently the highest concentrations from one
source do not coincide with the highest concentrations from the other and, hence, the middle
section of the wedge has few high concentration points (see Figure 3.2). This would be expected
if two different wind directions were needed for the receptor to be affected by sources. On the
other hand, if the wedge is filled, then either the receptor is affected by more than two sources or
there is some correlation in the times when the highest concentrations occur (see Figure 3.3).
Considerations such as these give the first indication of which species will be useful in the source
apportionment fitting and a lower bound for the number of sources that affect the receptor.
The next step in analyzing the data is to use source apportionment techniques to identify
the number and types of sources at each site. For this purpose, we use two source apportionment
tools: UNMIX and PMF. The application of these tools requires the specification of several
technical options. We note that while we describe, in detail, the reasons for some options
chosen, other options were chosen based on limited past experience with simulated data. The
choices made in these cases have not been independently verified.
UNMIX was used as a preliminary source apportionment tool. It provides additional
diagnostics to aid in determining how many sources should be included in the solutions. For
each site, five or more sources are indicated by the preliminary diagnostics for each site. Also,
by the very nature of what UNMIX does, it will not find a solution with any given number of
sources unless there is numerical evidence within the data for at least that many sources. Since
six source solutions were found with UNMIX for each of the sites, six is a good minimum for
these sites. However, because PMF will fit more species and sources, PMF can be easier for the
analyst to interpret.
Both tools require complete data for the species being fit to use the data from a given day.
In other words, both tools require a value for each species on each day being modeled. To
increase the number of available days for both tools, values less than the minimum detection
level were replaced with one-half that level. Previous work with synthetic data indicates that this
can increase the mean apportioned mass of species. (The total apportioned mass is not
constrained by the measured mass of species.) Missing data were filled in with the species mean
times the ratio of the daily PM2.5 value to the mean PM2.5 value. See Appendix A for additional
details about the data handling procedures.
Eight-Site SA Speciation Trends Final Report
8
September 24, 2003
-------
Aluminum Versus Silicon at the Houston Site
1.2
1.1
1.0
0.9
0.8
™E
S5
0.7
E
D
0.6
C
E
0.5
D
<
0.4
0.3
0.2
0.1
0.0
Figure 3.1
Silicon ^g/m
Aluminum versus Silicon Concentrations in the Houston Area.
Iron Versus Silicon at the St. Louis Site
1.2
1.1
1.0
0.9
0.8
™E
0.7
S5
0.6
c
0
—
0.5
0.4
0.3
0.2
0.1
0.0
Silicon ^g/m
Figure 3.2 Iron versus Silicon in the St. Louis Area.
Eight-Site SA Speciation Trends Final Report
9
September 24, 2003
-------
Calcium Versus Iron at the St. Louis Site
0.7
0.6
0.5
™E
0.4
£
m 0.3
CO
O
0.2
0.1
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
Iron Mg/rri3
Figure 3.3 Calcium versus Iron in the St. Louis Area.
3.2 Overview of PMF
The project used PMF as the main tool for the source apportionment. PMF uses the
monitoring data (including uncertainty estimates) and a user estimate of how many sources
significantly affect the receptor as input. The output includes two types of information for each
source: a profile and a time series of each source's strength at the receptor. There are several
equivalent ways of scaling the output, and these slightly change the interpretation of the output.
In this report, a source profile is a list of the mean species concentrations from the source at the
receptor. The corresponding time series, or relative source contributions, is a list of
multiplicative factors that indicate how much above or below the mean the source strength was
for a given day. With this representation, the profile list has concentration units and the relative
contributions are unitless ratios. The output in Appendix D is in this form. Alternatively, a
relative source profile could be a list of the ratios of the mean species concentrations from the
source at the receptor divided by the mean total mass concentration. The associated time series
(i.e., the source mass contributions) is a list of the total mass concentration at the receptor for
each of the measured days. In this representation, the profiles are unitless ratios and the
contributions have mass concentration units.
PMF performs "constrained" maximization of a weighted object function. The main
object function is a goodness-of-fit of the predicted mass contributions for each species, where
the species are typically weighted by a measure of trust in the individual measurements. The
measure of trust was adjusted for closeness to the minimum detection level, filling in for missing
Eight-Site SA Speciation Trends Final Report 10
September 24, 2003
-------
values, as well as for sampling error. The results are constrained to be non-negative (although
small negative values can occur) by adding penalty functions to the object function.
3.3 Species Modeled
The species used with PMF were PM2.5 (both from the speciation monitor and a
co-located FRM when available), sulfate, nitrate, ammonium, Al, As, Ba, Br, Ca, CI, Cr, Cu, EC,
Fe, Pb, Mn, Ni, OC, K, K+, Se, Si, Na, S, Sn, Ta, Ti, V, and Zn.
Further, PMF requires estimates for the uncertainty of the measurements as input so that
species with individual measurements can be weighted appropriately. This was obtained from
the co-located speciation data available within the AQS data (the "Mini Trends" data). These
data were collected as part of the initial design of the network just prior to the period modeled
(February 2000 to August 2000). A standard error in the measurements of each species was
estimated and used in the PMF modeling. Because a day-to-day variation in the uncertainties
can adversely affect PMF results, the only day-to-day variation in the uncertainties is for
instances of below MDL data and missing data. (See Appendix A for explicit details.)
PMF has other technical options used in the optimization that can influence the output.
These options are set in a text file (called an initialization file) that is read in at the time of
program execution. These files were generated in SAS and initially deviated from each other
only as required by the data to indicate appropriate files and file sizes. The technical options that
may be of most interest are:
1. The program was set to search for 5 to 10 source solutions at all sites. Analysis of
the solutions led us to use the 6 to 8 source solutions. A statistical algorithm was
implemented for the selection of the number of sources for Birmingham,
Charlotte, Indianapolis, Milwaukee, and Washington, D.C. This algorithm is
based on the Bayesian Information Criterion (BIC) that is frequently used for time
series model selection (Wei, 1990). Application of this algorithm yields
consistent results with the previous methods and reduces the effort needed for
model selection. Details of the use of BIC for selection of the number of sources
are included in Appendix L.
2. The program was run in its robust mode (as recommended by the software
developer).
3. The "outlier" sensitivity was set to 6. The usual range is 4 to 8, with 8 being the
least sensitive to outliers and 4 the most sensitive.
4. The "Fpeak" value was set between 0 and 0.5. This parameter can "rotate" the
solution toward either more zeros in the profile matrix or more zeros in the
contribution matrix with strength of the rotation set by the absolute value of the
parameter. The value of 0 gives no rotation. We attempted to follow the
procedure suggested by Phil Hopke (Willis, 2000) and used with the MARAMA
source apportionment study. The Fpeak parameter was increased to a point just
before the diagnostics showed a marked increase in the chi-squared
Eight-Site SA Speciation Trends Final Report
11
September 24, 2003
-------
goodness-of-fit value. However, compared to previous source apportionment of
IMPROVE data, the modeling was quite sensitive to this parameter (i.e., the
Q value increased dramatically), and it was decided to use only the zero setting.
The sensitivity of the Q value to the Fpeak parameter indicates little rotational
ambiguity in the solutions. The sensitivity may also be a function of the number
of days with data.
5. The program was run from at least six different random starting points and the
best fitting solution was used. (The software documentation recommends using
multiple starting points, but makes no recommendation on how many.)
The settings chosen above are based on previous experience with the PMF modeling
software. The software has not been systematically tested to identify the best setting. These
choices may not be optimal.
3.4 Analyses of the Residuals
The modeling procedures included analyses of the residuals and errors. Some of these
were done for model selection purposes (see Appendix L) and were used for assessment of the
model goodness-of-fit and estimation of the modeling errors. This section discusses the
assessment of model goodness-of-fit and the estimation of the modeling errors.
The assessment of the model goodness-of-fit starts with the assessment of the Q values
report by PMF. The Q value is the sum over all days and species of the squared residuals (daily
measured species concentration minus the total over all sources of the model estimate of the
species concentration) divided by the square of the species' uncertainty estimate. Table 3.1
shows the Q value for the model chosen at each site. It also shows the number of species
modeled, the number of days modeled, and the number of sources in the model. The
"theoretical" expected value for Q can be expressed in terms of the model characteristics. These
values range from 0.9 to 10 times the theoretical value, but the theoretical value is based on
assumptions that are not quite satisfied by the model and assumes that there are no "outliers"
identified by the model that are treated differently when running PMF in robust mode. Deleting
the outliers from consideration reduces the observed Qs to be at most 3 times the theoretical
value.
Eight-Site SA Speciation Trends Final Report
12
September 24, 2003
-------
Table 3.1
Summary of the PMF Q Values
Site
Species
Modeled
Days
Sources
Q
Birmingham, Al
30
186
7
55392
Bronx, NY
30
160
7
9456
Charlotte, NC
30
143
8
3899
Houston, TX
301
121
7
8974
Indianapolis, IN
30
155
8
7662
Milwaukee, Wl
30
172
8
14576
St. Louis, MO
30
112
7
32862
Washington, DC
30
128
6
4611
1 Including the duplicate mass measurement.
The Q values give a good summary across all of the data. Quantile-quantile or Q-Q plots
were used to examine the modeling errors at the species level. Examples are shown in
Figures 3.4 and 3.5. These are formed by sorting the scaled residuals (the residual divided by the
uncertainty) and plotting them against the quantiles of a normal distribution with mean 0 and
standard deviation 1. Ideally, these points should lie on the y = x line. They will lie on a line
with a different slope if the uncertainties are uniformly underestimated or overestimated.
Deviations from a straight line indicate that either modeling assumptions are incorrect, or the
model chosen for the site is incorrect.
Eight-Site SA Speciation Trends Final Report
13
September 24, 2003
-------
Q-Q Plot for the FRM Mass in St. Louis
fci
w
T3
a
-i o 1
Standard Normal Quantiles
Figure 3.4 Q-Q plot of the Scaled Residuals of the FRM Data for St. Louis.
Q-Q Plot for the Speciation Mass in St. Louis
&
T3
13 2
a
1 -2
-1 0 1
Standard Normal Quantiles
Figure 3.5 Q-Q plot of the Scaled Residuals of the Speciation Monitor Mass Data for
St. Louis.
Eight-Site SA Speciation Trends Final Report
14
September 24, 2003
-------
Figures 3.4 and 3.5 show that the model assumptions hold fairly well in St. Louis.
However, the FRM uncertainties have been slightly overestimated while the uncertainties of the
mass from the speciation monitor are slightly underestimated. Other sites and species also have
fairly straight Q-Q plots with varying slopes.
Modeling error can also be assessed by examining the difference between the apportioned
values for the FRM mass and the mass from the speciation monitor (except in Houston, which
did not have a co-located FRM) and the difference between three times the sulfur concentration
(the apportioned XRF sulfur mass) and the sulfate concentration (the apportioned IC sulfate
mass). The two mass values should differ only by measurement error as should the sulfur-sulfate
pair under the assumption that all of the sulfur is present in the form of sulfate. The differences
give a direct means of estimating the errors in the apportioned masses of the species (assuming
that the other species are similar).
For each site, four summary values are shown in Table 3.2. The first is error estimated
from the relative differences in the apportionment of the two total mass values. The second is
from the relative differences between the three-times-the-sulfur and the sulfate apportionment.
The third is a weighted average of these two (weighting the mass twice as much as the sulfur
based error estimate). The final column is an estimate of the relative error of the mean of the
apportioned FRM mass and the speciation mass. (The mean is the mass value shown throughout
the report for each source.) The standard errors listed in Sections 5 and 7 are based on the last
column in Table 3.2. The mass based errors were calculated using:
where n is the number of sources, FRM, is the FRM mass apportioned to the ith source, and SPM,
is the apportioned speciation monitor mass for the ith source. The sulfur based errors were
calculated using:
where n is the number of sources, Sulfur,¦ is the sulfur mass apportioned to the ith source, and
Sulfatei is the apportioned sulfate mass for the ith source.
mass based error =
(Eq. 1.)
sulfur based error =
(Eq. 2.)
Eight-Site SA Speciation Trends Final Report 15
September 24, 2003
-------
Table 3.2
Model Error Estimates
Site
Mass
Based
Sulfur
Based
Combined
Mass
CV
Bronx, NY
73%
43%
63%
45%
Birmingham, Al
25%
52%
34%
24%
Charlotte, NC
56%
78%
61%
43%
Houston, TX
NA
84%
84%
84%
St. Louis, MO
15%
48%
30%
21%
Milwaukee, Wl
71%
68%
67%
47%
Washington, DC
60%
87%
66%
47%
Indianapolis, IN
39%
60%
45%
32%
4.0 IDENTIFYING THE SOURCES
The source apportionment output yields a chemical profile for each source (or source
category) and a time series for the mass. (See Appendices D through K for the graphical output.)
While the profile is unique for the source, it does not explicitly identify the source. Two main
methods were employed to identify the sources from the PMF output. Both of these methods
were applied to each source identified at each of the eight sites. First, an automated method was
used to match the output with source profiles in the speciate database,
http://www.epa.gov/ttnchiel/software/speciate/index.html. The matching algorithm produces up
to ten possible source matches with specific sources from the speciate database. The second
"method" is informed opinion. Using the automated matching, past experience, and discussions
with local individuals, most of the profiles can be identified with specific source categories.
Some "rules of thumb" are given below as a guide on how to start that process. This is followed
by evaluation of the back trajectories and comparisons with the local meteorological data to
check for consistency.
4.1 Automated Matching of the Source Apportionment Output
The first step in identifying the profiles is with an automated matching algorithm. The
algorithm is based on a weighted regression between the source apportionment profile output and
the source profiles in speciate. The fit is derived from the mean-squared-error (MSE), and it can
be interpreted as approximately the average across the species of the percent difference between
the mean source mass and a corresponding speciate source mass relative to the mean species
mass observed at the receptor. Smaller values indicate a better match between the profiles.
However, small values can occur when a speciate profile includes only species for which the
source is not a significant contributor to the receptor mass. Generally, 0 to 10 percent is a very
good fit, 10 to 15 percent is good, 15 to 20 percent is a marginally good fit, and greater than
20 percent indicates a poor match to the speciate profile. See Appendix B for details of the
matching algorithm. The first use of the algorithm is as a check on the modeling results, in
particular, the number of sources. Physically meaningful results will have fairly consistent
source assignments, e.g., a list of various dirt and road dust profiles. Once the modeling is done,
the consistent source assignments become the initial source assignments unless the source is
Eight-Site SA Speciation Trends Final Report 16
September 24, 2003
-------
flagged as secondary, in which case the source is also flagged with the dominant secondary
species.
4.2 Guidelines for Assigning Preliminary Identifications
The preliminary identifications are based on review of the various output and some
additional local information. The following are general characteristics associated with sources
that can be used to aid in source identifications:
• Crustal: This is a source consisting of silicon, aluminum, iron, and other trace metals
• Residual oil: This is a source high in sulfate with vanadium and nickel.
• Mobile/Secondary OC: Anything characterized by high OC, some EC (strictly less
than the OC), very little sulfate, and some metals (particularly, Ba from brake pads).
• Sea salt: a source with high sodium content, Mg and Mn. Usually contains
secondary formations also.
• Vegetative burning: A source with significant amounts of K and with OC > EC.
• Incinerator: A source of OC, EC, sulfate, and trace metals without V.
• Industrial non-oil/non-coal: A source high in sulfate without Se, V, or Ni.
• Road sand: A wintertime silicon source.
• Industrial: A source of sodium with a mix of sulfate, OC, EC, and metals.
• Coal-Fired Power plant: A large sulfate source with Se and frequently Ti.
• Diesel: An EC-OC source with EC > OC + sulfate + trace metals. Mn from the
additive MMT should no longer be included in on-road diesel sources, but may be
included in off-road sources.
• Smelters: These are the sources of trace metals, particularly Pb, Zn, Sr, Cu, and/or Ti
without much OC or EC.
• Wood smoke: Wintertime vegetative burning.
• Road salt: A sodium source with mobile components (Na, OC, EC + metals, low
sulfate).
• Fireworks: A source high in OC with a significant amount of K and a high source
strength on or after July 4 and/or January 1. May also have significant amounts of Cu
and other trace metals.
Eight-Site SA Speciation Trends Final Report 17
September 24, 2003
-------
4.3 Final Identifications
The final identifications are a merging of all the various analyses and review by source
apportionment experts and local representatives, and represent our best current understanding of
the sources. This section discusses the primary characteristics of the sources identified. In
particular, some of the coal combustion sources are flagged with "(Ni)," which is discussed
below.
• Ammonium nitrate - As the name implies, the "source profiles" for this category are
dominated by ammonium and nitrate. This is an example where the source
apportionment source profile cannot be matched with an emission source. The
problem comes from the fact that the relative amount of nitrate due to an emission
source will generally not be constant with respect to the other species because of the
semi-volatile nature of nitrate. Hence, the tools will separate the species into its own
source category. This separation would not matter as much if there was only one
major emission source. Unfortunately, ammonium nitrate is formed from a
combination of ammonia (with a large portion coming from agricultural sources) and
NOx (with substantial portions from both utilities and mobile sources). Some of the
profiles contain coal burning tracers and some of the preliminary transport analyses
seem to indicate a relationship to coal burning, but these only reveal that coal burning
is part of the source. Apportionment of these species may be possible by restricting
the analyses to periods with cooler weather.
• Canadian fires - In July 2002, there were major fires in Canada. The plume from
these fires can be seen in satellite photos and the source is clearly tied to this event. It
would be expected that any wood smoke during the rest of the year would also be
apportioned to this source, but the source is so strongly dominated by the single event
that it is difficult to tell.
• Coal combustion- This is the major source of sulfate for all sites (and, hence, the
major source). Differences in fuel sources and distances to the source contribute to
the site-to-site variations in the profiles. In the case of Indianapolis, the source was
split into two sources that are similar to what has been found at various IMPROVE
sites. The split is consistent with two extremes in the atmospheric formation of the
sulfate with the one portion related to a cold weather pattern (a wintertime peak) and
the other associated with a warmer weather. The coal combustion source is also a
major source of Se, a coal burning tracer.
• Coal combustion (Ni) - This is a variation of the coal combustion profile
characterized by a Ni/Se ratio that is greater than 1. Several of the sources marked as
having an enhanced Ni content also have higher amounts of V compared to the other
coal combustion sources. Ni and V have both traditionally been used as oil burning
tracers, so this may indicate that some oil combustion is being apportioned to these
coal combustion sources. However, both elements are also present in fly ash. The
amounts can vary due to the operating conditions of the power plant. Additionally,
the amount of Ni in the fly ash has been shown to vary with the particle size.
Eight-Site SA Speciation Trends Final Report
18
September 24, 2003
-------
Moreover, both Ni and V oxides can act as catalysts in the conversion of SO2 into
sulfate, enhancing particle growth. These factors may lead to deposition rates or
particle growth beyond the 2.5 cut point that vary with the Ni (and V) content.
Hence, the Ni enhancement relative to the Ni content in coal burning profiles at other
sites may have nothing to do with oil combustion. Additional analyses are planned
that may help resolve this issue.
• Crustal - All sites are apportioned a crustal source. The profiles match the profiles
found in Speciate quite well.
• Industrial sources - These are expected to vary considerably from site to site. Most
of the time they probably represent a mix of a strong local set of industrial emissions
and small amounts of any similar sources/mixes that happen to be in the region. In
Houston, the wind data suggest a relationship with the industries in the ship channel.
In Bronx, the back trajectory analyses suggest a regional mixture of sources from
along the east coast.
• Marine and industrial salts- These sources have sea spray components (including
trace metals) and source regions that extend into the ocean. The back trajectories
suggest that there are inland sources also. This leads to the industrial salt
characterization. It is likely that neither category is large enough or distinctive
enough for the tools to separate.
• Mobile sources- These include both gas and diesel mobile sources. It may be
possible to separate the gas and diesel sources with speciated carbon data or a
surrogate such as the IMPROVE carbon fractions. Neither of these was available for
the sites studied here. The sources in this study are the dominant sources of organic
carbon and, hence, are expected to be mostly associated with gasoline combustion.
Local mobile sources would generally be expected to be stronger during the week
compared with weekends. However, the delays in transport would obscure that
relationship if a significant portion is not local.
• Oil combustion- Two oil combustion sources were identified. They are carbon
sources and sulfate, which are also the major sources of Ba, Ni, and V.
• Road construction- This was identified for the Washington, D.C., site. The source
profile is a mix of crustal components and diesel mobile (EC dominant). The source
is stronger during weekdays and lasts for several months.
• Smelting and steel production - These are characterized by their metal content and
distinguished from incinerators by the lack of carbon. The profiles may also show
power production components either due to direct coal burning or coal burning by the
electrical source that varies with production. In St. Louis the local wind pattern
associates the source strengths with known local sources.
Eight-Site SA Speciation Trends Final Report
19
September 24, 2003
-------
• Vegetative burning and fireworks- The July 4th source event clearly dominates the
source strength pattern. Both source categories are high in organic carbon and are
major sources of potassium. The fireworks are probably responsible for the copper
and other trace metal components. However, the other similarities in the profiles and
indications of small amounts of source activity during other times of the year suggest
that vegetative burning is included in this source category.
• Zinc and other sources identified by species- These are each characterized by being a
major contributor of a specific species or containing an unusual amount of the
species. In St. Louis, there is a zinc refinery in a direction indicated by the local wind
data and, hence, this zinc source is identified. However, zinc is also found in
incinerator and recycling emissions, and these may be included in that profile and the
zinc source found in Birmingham. The other sources only identified by species are a
lead source for Birmingham and a chlorine source for Milwaukee.
5.0 RESULTS OF THE SOURCE APPORTIONMENT ANALYSIS
This section presents the source identifications for each site. See Appendix C for the
numerical results. See Appendices D through K for graphical representations of the source
apportionment output and source strength analyses for each site. Any mention of explicit sources
within the source identifications is included only as an example of a local source with the
characteristics similar to what the study has found. Additional analysis would be needed to
relate an effect at the receptor to an explicit source.
Note that while the main reasoning behind making source category assignments is based
on the characteristics of the source profile, the additional supporting analyses are also
considered. Hence, some of the comments in this section refer to analyses described in
Section 6.
Eight-Site SA Speciation Trends Final Report
20
September 24, 2003
-------
5.1 The Birmingham, Alabama, Site
Table 5.1 Summary of Source Identification for Birmingham, Alabama
Source
Number
Identification
Apportioned
Mass (SE) |jg/m3
Notes and Profile Comments
1
Ammonium Nitrate
1.84 (0.45)
The Se is indicative of a coal-NOx
relationship.
2
Crustal
1.27 (0.31)
3
Mobile Sources
6.51 (1.59)
Expected; OC>EC indicates gasoline rather
than diesel dominance; WD>WE
4
Vegetative Burning and
Fireworks
1.15 (0.28)
It is assumed that if the main event is
removed, that the remainder is vegetative
burning.
5
Lead Source
0.71 (0.17)
Dominated by a single event.
6
Zinc Source
0.79 (0.19)
Possible sources include recycling plants,
smelters, and incinerators.
7
Coal Combustion (Ni)
7.27 (1.77)
The sulfate and Se content associates this
with coal burning. See Section 4.3
regarding enhanced Ni content.
Notes:
• Additional analyses are needed to refine the source labels for Sources 5 and 6.
Eight-Site SA Speciation Trends Final Report
21
September 24, 2003
-------
5.2 The Bronx, New York, Site
Table 5.2 Summary of Source Identification for Bronx, New York
Source
Number
Identification
Apportioned
Mass (SE) (jg/m3
Notes and Profile Comments
1
Coal Combustion
5.29 (2.36)
Key species include NH4, OC, S04, mass.
This is consistent with the regional
background/transport sources observed in
all SA analyses done in the Northeast.
2
Oil Combustion
1.22 (0.54)
Key species include EC, OC, CI, V, Ni, V,
and Ni, winter peak lead to fuel oil
combustion.
3
Marine and Industrial
Salts
0.30 (0.13)
Key species include Na, K, CI, several
metals. There is some indication of general
industrial sources.
4
Mobile Sources with Tire
Wear
2.49 (1.11)
Key species include Na, OC>EC, several
metals. Possible mobile source profile
including tire wear.
5
Industrial
1.82 (0.81)
Key species include Zn, Ca, Se, Ni, Pb,
OC>EC. Winter peak. Note that the sulfur
and V contributions are low while Zn, Pb,
Cu, and Ca are enhanced.
6
Ammonium Nitrate
4.09 (1.82)
Key species include K, N03, NH4, mass.
This is consistent with a regional nitrate
signature.
7
Crustal
0.97 (0.43)
Key species include K, Al, Ca, Si, Ti. Most
likely from street cleaning and agricultural
transport.
Notes:
• The coal combustion profile and the summer peak for Source 1 are consistent with the
regional background/transport sources observed in all SA analyses done in the
Northeast. This is consistent with the observation that all of the area sulfate values
fluctuate in unison across the area in contrast to the OC and EC values that vary
spatially.
• The marine and industrial profile has indications of general industrial sources, namely
the presence of several metals and its source region.
• The fuel oil and oil combustion profiles have similarities. The oil combustion profile
may contain some traces of coal combustion or another industrial source as well (note
the Se and other metals).
• The nitrate profile is consistent with northeast regional nitrate formation and
transport.
Eight-Site SA Speciation Trends Final Report
22
September 24, 2003
-------
5.3 The Charlotte, North Carolina, Site
Table 5.3 Summary of Source Identification for Charlotte, North Carolina
Source
Number
Identification
Apportioned
Mass (SE) |jg/m3
Notes and Profile Comments
1
Vegetative Burning and
Fireworks
0.48 (0.21)
It is assumed that if the main event is
removed, that the remainder is vegetative
burning.
2
Coal Combustion
5.71 (2.47)
Se is associated with this source linking it to
coal combustion.
3
Crustal
0.57 (0.24)
4
Oil Combustion
1.87 (0.81)
Ba may be a useful tracer for power plants.
5
Marine and Industrial
Salts
0.08 (0.04)
Winds support this conclusion.
6
Ammonium Nitrate
1.21 (0.52)
The Se in the factor associates it with coal
combustion.
7
Smelting
0.67 (0.29)
Copper, Zinc, and EC typical of smelting /
metal production.
8
Mobile Sources
3.87 (1.68)
We expect mobile sources, however, the
weekday pattern does not support it.
Notes:
• It was noted by local contacts after the source apportionment modeling was done that
the speciation PM2.5 concentration values are generally higher than the PM2.5 FRM
data. The MetOne SASS generally yields higher values than the FRM, but during the
first six months of sampling, the SASS values were determined to be much higher
because of a problem with the filter cassettes. After this problem was corrected in
July 2001, the mass value difference was reduced. The apportioned mass listed for
the sources is based on the average of the two apportioned masses as with the other
sites. The standard error listed is based on the difference in the apportioned values
and, hence, it may be slightly larger because of this problem. It should be noted that
the filter problem would have affected the other species as well as the mass and that
the modeling results are based on the inclusion of those data.
Eight-Site SA Speciation Trends Final Report
23
September 24, 2003
-------
5.4 The Houston, Texas, Site
The data from this site are the least consistent. There are many days (-10 percent of the
days) where a reconstructed mass is more than 1.3 times the measured mass. These days were
not used in the analysis. The PM2.5 and FRM values are all the same for this site. The duplicate
values were included in the apportionment so that the species would be weighted the same as
they are at the other sites.
Table 5.4 Summary of Source Identification for Houston, Texas
Source
Number
Identification
Apportioned
Mass (SE) |jg/m3
Notes and Profile Comments
1
Crustal
0.77 (0.65)
Usual crustal elements.
2
Vegetative Burning and
Fireworks
0.49 (0.41)
The peak is for July 4. The July 5 estimate
is about half of the July 4 value. The
wintertime portion may be consistent with
wood smoke.
3
Industrial
0.87 (0.73)
The chlorine content associates this with
local industrial sources.
4
Mobile Sources
5.19 (4.38)
This site is in a residential neighborhood
with freeways to the north.
5
Marine Ammonium Nitrate
0.29 (0.24)
This could be a marine influenced profile
from the gulf or bay on which sodium nitrate
has formed as the air parcels pass over the
emissions sources. That would explain the
absence of ammonium and sulfur.
6
Mobile Mn Source or
Grain Dust
1.04 (0.88)
The Mn signature may indicate off-road
diesel. Possible transport from across the
Gulf? Or it could be grain dust with a Mn
anti-fungal coating with other ship channel
sources.
7
Coal Combustion (Ni)
5.54 (4.68)
The Se associates this with coal
combustion. (See Section 4.3 regarding
enhanced Ni content.)
Notes:
• The vegetative burning and fireworks source may also include wood smoke during
the winter.
• It is hard to be any more specific about the industrial source. The industrial mixture
in the Houston Ship Channel is very broad, and so perhaps it is not unusual that we
cannot more precisely identify the source type.
• The nitrate source profile could be a marine influenced profile from the gulf or bay in
which sodium nitrate has formed as the air parcels pass over the emissions sources.
That would explain the absence of ammonium and sulfur.
Eight-Site SA Speciation Trends Final Report
24
September 24, 2003
-------
• There is a coal burning power plant to the southwest of the site, Texas Lignite. The
coal combustion source also includes material from transport.
Eight-Site SA Speciation Trends Final Report
25
September 24, 2003
-------
5.5 The Indianapolis, Indiana, Site
Table 5.5 Summary of Source Identification for Indianapolis, Indiana
Source
Number
Identification
Apportioned
Mass (SE) (jg/m3
Notes and Profile Comments
1
Vegetative Burning and
Fireworks
0.69 (0.22)
It is assumed that if the main event is
removed, that the remainder is vegetative
burning.
2
Ammonium Nitrate
3.58 (1.15)
3
Canadian Fires
0.25 (0.08)
Coincides with transport from a large,
known fire event.
4
Marine and Industrial
Salts
0.47 (0.15)
Note the substitution of chloride with nitrate
during transport from the Gulf.
5
Crustal
0.51 (0.16)
6
Mobile Sources
3.21 (1.03)
Expected mobile sources. Note that
OC>EC indicates gasoline rather than
diesel dominance, however the day of week
pattern is not supportive.
7
Coal Combustion 1
1.64 (0.53)
EC, Se and winter similar to findings from
Poirot.
8
Coal Combustion 2 (Ni)
7.03 (2.26)
See Section 4.3 regarding enhanced Ni
content.
Notes:
• The marine and industrial salt source is probably enhanced by secondary material in
transport.
Eight-Site SA Speciation Trends Final Report
26
September 24, 2003
-------
5.6 The Milwaukee, Wisconsin, Site
Table 5.6 Summary of Source Identification for Milwaukee, Wisconsin
Source
Number
Identification
Apportioned
Mass (SE) |jg/m3
Notes and Profile Comments
1
Coal Combustion (Ni)
4.54 (2.14)
See Section 4.3 regarding enhanced Ni
content.
2
Mobile Sources
1.53 (0.72)
OC>EC indicates gasoline rather than
diesel dominance, however the day of week
pattern is not supportive.
3
Crustal
0.12 (0.06)
4
Chlorine Sources
2.66 (1.26)
May be from industrial sources.
5
Ammonium Nitrate
4.07 (1.92)
6
Crustal Related Events
0.19 (0.09)
Mainly from three events.
7
Vegetative Burning and
Fireworks
0.35 (0.17)
It is assumed that if the main event is
removed, that the remainder is vegetative
burning.
8
Industrial Diesel and
Sulfate Mix
0.93 (0.44)
Notes:
• The source identification of Sources 4 and 8 are the least certain of all of the sources.
These are the only two sources among all eight sites for which Sonoma and Battelle
could not agree to a likely source category label.
Eight-Site SA Speciation Trends Final Report
27
September 24, 2003
-------
5.7 The St. Louis, Missouri, Site
Table 5.7 Summary of Source Identification for St. Louis, Missouri
Source
Number
Identification
Apportioned
Mass (SE) |jg/m3
Notes and Profile Comments
1
Zinc Refinery
0.85 (0.18)
Big River Zinc Corporation is located 5-10
miles to the SE.
2
Smelting (Copper)
0.59 (0.12)
Cerro Copper Products Company is located
5-10 miles to the SE.
3
Coal Combustion
5.74 (1.21)
Consistent with power generation. Does not
show a seasonal trend.
4
Steel Production
0.76 (0.16)
Granite City Steel may contribute to high Fe
levels.
5
Ammonium Nitrate
5.02 (1.06)
NOx from power plants. Power plant to the
southeast.
6
Crustal
1.43 (0.30)
High Ca, K relative to typical crustal.
Possibility cement plant or limestone
quarrying, but peaks probably coincide with
agricultural activity.
7
Mobile Sources
2.92 (0.62)
High Pb possible because of residue (in
road dust) from old Pb smelter emissions
and hauling w/o tarps.
Notes:
• Up to 1.1 micrograms of iron per cubic meter of air have been measured. The iron
concentration is above 0.5 micrograms per cubic meter 10 percent of the time.
• The mobile source is not a major source of lead, but does have an unusual amount of
lead compared to other mobile sources around the nation.
Eight-Site SA Speciation Trends Final Report
28
September 24, 2003
-------
5.8 The Washington, D.C., Site
Table 5.8 Summary of Source Identification for Washington, D.C.
Source
Number
Identification
Apportioned
Mass (SE) |jg/m3
Notes and Profile Comments
1
Vegetative Burning and
Fireworks
0.53 (0.25)
It is assumed that if the main event is
removed, that the remainder is vegetative
burning.
2
Coal Combustion
7.70 (3.62)
3
Ammonium Nitrate and
Salt
1.23 (0.58)
Has NaCI and may have some substitution
of chloride with nitrate. Possibly a mix with
road salt.
4
Mobile Sources
4.72 (2.22)
Local and transported pollutants: gasoline
dominant (OC>EC), however, the day of
week pattern is not as expected. May also
include power plant combustion, note Se,
Ni, V, and sulfate.
5
Canadian Fires
1.11 (0.52)
Coincides with transport from large known
fire event.
6
Road Construction
1.47 (0.69)
Crustal component with diesel influence.
Note EC, metals, and Mn plus day of week
pattern (WD>WE).
Notes:
• The road construction source contains a majority of the crustal material. The profile
shows a mix of crustal and diesel components.
• The Canadian fire source is a known event that was observed on the eastern coast in
July 2002.
Eight-Site SA Speciation Trends Final Report
29
September 24, 2003
-------
6.0 BACK TRAJECTORY ANALYSIS AND ANALYSIS WITH LOCAL
METEOROLOGICAL DATA
The source strength output from the PMF model can be combined with meteorological
data to yield information about possible source locations. Two very different methods were used
to obtain this information. They rely on very different assumptions about the pathway from the
source to the receptor. Most of the output in this section is graphical and is contained in
Appendices D through K. Tables 6.1 through 6.5 contain summaries based on examining the
various graphs.
6.1 Pollution Roses
For each of the sites, hourly local wind data for the nearest national weather station were
downloaded from the NOAA website (http://nndc.noaa.gov/7home.shtml). (See Table 2.3.)
These data were used to produce pollution roses for each source at each site. (See Appendices D
through K.) The pollution roses show the mean source strength relative to the overall source
strength by direction and wind category: 1 to 5 mi/hr, 5 to 10 mi/hr, and 10+ mi/hr. See
Table 6.1 for a summary of the information from the comparisons with the local, low-level
winds.
The summaries and the graphs need to take into consideration that the weather station
data are collected from 10 m towers and are representative of low-level, local winds. The
analyses assume a straight-line path consistent with these low-level winds and, hence, are best
suited to local sources. Local point sources should show a clear directional preference that may
be associated with a particular wind range also. Similarly, local area sources should show a
preference with a continuous range of directions. Sources such as dust that are related to wind
speed should show preference for particular wind speeds. However, distant sources may also be
associated with particular higher wind speeds and directions. Finally, nitrate sources need to be
considered carefully, because nitrate formation is enhanced in colder weather. Hence, a direction
could be associated with a nitrate source simply because it is associated with colder weather.
Eight-Site SA Speciation Trends Final Report
30
September 24, 2003
-------
Table 6.1 Pollution Rose Summaries
Site
Source
Direction
Speed
Birmingham
Ammonium Nitrate
Uniform
Uniform
Birmingham
Crustal
Easterly
Decreases with speed
Birmingham
Mobile Sources
Uniform
strongest < 1 mph
Birmingham
Vegetative Burning and Fireworks
NE, E, SE, S, SW
strongest < 1 mph
Birmingham
Lead Source
E, SE
Decreases with speed
Birmingham
Zinc Source
NE, E, SE
Uniform
Birmingham
Coal Combustion (Ni)
N, NE, E, SE, S, SW
Uniform
Bronx
Coal Combustion
SE, S, SW
Decreases with speed
Bronx
Oil Combustion
SW, W, NW
strongest < 1 mph
Bronx
Marine and Industrial Salts
NE, E, SE
Uniform
Bronx
Mobile Sources with Tire Wear
Easterly
Uniform
Bronx
Industrial
Westerly
Uniform
Bronx
Ammonium Nitrate
SE, S, SW, W
Decreases with speed
Bronx
Crustal
SE
Decreases with speed
Charlotte
Vegetative Burning and Fireworks
N, NW, SE, S
Uniform
Charlotte
Coal Combustion
all except W
Uniform
Charlotte
Crustal
Northerly and Southerly
Uniform
Charlotte
Oil Combustion
Westerly
Uniform
Charlotte
Marine and Industrial Salts
Uniform
Uniform
Charlotte
Ammonium Nitrate
all except W
Uniform
Charlotte
Smelting
SE, S, SW, W, NW
Decreases with speed
Charlotte
Mobile Sources
Southerly
Decreases with speed
Houston
Crustal
SW
strongest < 5 mph
Houston
Vegetative Burning and Fireworks
W
strongest < 5 mph
Houston
Industrial
SE, S
Uniform
Houston
Mobile Sources
Northerly
strongest < 10 mph
Houston
Marine Ammonium Nitrate
NW, SE
Uniform
Houston
Mobile Mn Source or Grain Dust
NW, SE
Decreases with speed
Houston
Coal Combustion (Ni)
Easterly
Uniform
Eight-Site SA Speciation Trends Final Report
31
September 24, 2003
-------
Table 6.1 Pollution Rose Summaries (continued)
Site
Source
Direction
Speed
Indianapolis
Vegetative Burning and Fireworks
NW
strongest 1-5 mph
Indianapolis
Ammonium Nitrate
E, W
Uniform
Indianapolis
Canadian Fires
NW, NE
strongest < 1 mph
Indianapolis
Marine and Industrial Salts
Southerly
Uniform
Indianapolis
Crustal
SW
Uniform
Indianapolis
Mobile Sources
Uniform
Decreases with speed
Indianapolis
Coal Combustion 1
NE, E, SE, S, SW
Decreases with speed
Indianapolis
Coal Combustion 2 (Ni)
NE, E, SE, S, SW
Decreases with speed
Milwaukee
Coal Combustion (Ni)
Southerly
strongest 1-5 mph
Milwaukee
Mobile Sources
S, SW, W, NW
strongest < 5 mph
Milwaukee
Crustal
Southerly
Uniform
Milwaukee
Chlorine Sources
Southerly
Uniform
Milwaukee
Ammonium Nitrate
Southerly
strongest 5-10 mph
Milwaukee
Crustal Related Events
SE, S, SW, W, NW
Uniform
Milwaukee
Vegetative Burning and Fireworks
NE, SW
Uniform
Milwaukee
Industrial Diesel and Sulfate Mix
NE, S, SW
strongest 1-10 mph
St. Louis
Zinc Refinery
N, NE, E, SE
Decreases with speed
St. Louis
Smelting (Copper)
Easterly
Uniform
St. Louis
Coal Combustion
NE, E, SE, S, SW
Decreases with speed
St. Louis
Steel Production
Easterly
strongest 5-10 mph
St. Louis
Ammonium Nitrate
Northerly
Uniform
St. Louis
Crustal
S, SW
Decreases with speed
St. Louis
Mobile Sources
NE, E, SE, S, SW
Decreases with speed
Washington
Vegetative Burning and Fireworks
NW, N
strongest 5-10 mph
Washington
Coal Combustion
N, NE, E, SE, S, SW
strongest < 5 mph
Washington
Ammonium Nitrate and Salt
Easterly
Decreases with speed
Washington
Mobile Sources
NE, E, SE, S, SW
strongest < 5 mph
Washington
Canadian Fires
N, SW
strongest < 5 mph
Washington
Road Construction
NE, E, SE, S
strongest < 5 mph
6.2 Temperature and Pressure Comparisons
The local meteorological data were used to compare the source strength with the
temperature and pressure. The temperature comparison was made seasonally, and the pressure
comparison is over the entire modeling period. An open question is how strongly the nitrate
source strengths are associated with temperature. While the source strengths are rarely related to
the pressure, it was felt to be a good check because high pressure systems tend to concentrate the
pollution. Hence, a strong correlation would indicate that the source strength is being driven by
the meteorological conditions rather than increased source activity and/or favorable wind
directions, which would violate the assumptions made in the back trajectory and pollution rose
analyses. Tables 6.2 and 6.3 summarize the temperature and pressure correlations, respectively.
Eight-Site SA Speciation Trends Final Report
32
September 24, 2003
-------
Table 6.2 Summary of Source Strength Correlation with Temperature
Site
Source
Temperature Correlation
Birmingham
Ammonium Nitrate
Winter (-), Spring (—)
Birmingham
Crustal
Birmingham
Mobile Sources
Birmingham
Vegetative Burning and Fireworks
Birmingham
Lead Source
Fall (+)
Birmingham
Zinc Source
Fall (++)
Birmingham
Coal Combustion (Ni)
Winter and Spring (++), Summer (+)
Bronx
Coal Combustion
Winter (+), Fall and Spring (++), Summer (+++)
Bronx
Oil Combustion
Bronx
Marine and Industrial Salts
Spring (-)
Bronx
Mobile Sources with Tire Wear
Spring (+)
Bronx
Industrial
Fall (-), Winter (+)
Bronx
Ammonium Nitrate
Spring (-)
Bronx
Crustal
Fall, Winter, and Spring (++), Summer (+)
Charlotte
Vegetative Burning and Fireworks
Fall (-), Spring (++)
Charlotte
Coal Combustion
Spring (+++), Summer (+)
Charlotte
Crustal
Winter (-), Summer (+)
Charlotte
Oil Combustion
Fall (+), Winter (++)
Charlotte
Marine and Industrial Salts
Fall (+)
Charlotte
Ammonium Nitrate
Winter and Spring (--), Summer (-)
Charlotte
Smelting
Winter (-)
Charlotte
Mobile Sources
Spring (++)
Houston
Crustal
Fall (+), Spring (++)
Houston
Vegetative Burning and Fireworks
Winter (-)
Houston
Industrial
Fall (+++), Summer (-)
Houston
Mobile Sources
Fall (--), Winter and Spring (-)
Houston
Marine Ammonium Nitrate
Fall (-), Winter (-)
Houston
Mobile Mn Source or Grain Dust
Spring and Winter (+)
Houston
Coal Combustion (Ni)
Fall, Winter, and Summer(+)
Indianapolis
Vegetative Burning and Fireworks
Spring and Summer (+)
Indianapolis
Ammonium Nitrate
Winter (-), Spring (--)
Indianapolis
Canadian Fires
Indianapolis
Marine and Industrial Salts
Fall (++), Summer (-)
Indianapolis
Crustal
Winter and Summer (+), Spring (++)
Indianapolis
Mobile Sources
Spring (++)
Indianapolis
Coal Combustion 1
Winter (-)
Indianapolis
Coal Combustion 2 (Ni)
Winter (+), Fall and Summer (++), Spring (+++)
Eight-Site SA Speciation Trends Final Report
33
September 24, 2003
-------
Table 6.2 Summary of Source Strength Correlation with Temperature (continued)
Site
Source
Temperature Correlation
Milwaukee
Coal Combustion (Ni)
Winter and Summer (+), Fall (+++)
Milwaukee
Mobile Sources
Fall (+), Winter Spring and Summer (+)
Milwaukee
Crustal
Winter (++), Spring and Summer (+)
Milwaukee
Chlorine Sources
Winter (-), Spring (+), Summer (++)
Milwaukee
Ammonium Nitrate
Fall and Winter (+)
Milwaukee
Crustal Related Events
Fall (++)
Milwaukee
Vegetative Burning and Fireworks
Fall (+), Spring (++)
Milwaukee
Industrial Diesel and Sulfate Mix
Winter (++), Spring and Summer (+)
St. Louis
Zinc Refinery
Fall and Summer (+)
St. Louis
Smelting (Copper)
Winter (-), Summer (+)
St. Louis
Coal Combustion
Winter (+), Spring and Summer (++), Fall (+++)
St. Louis
Steel Production
Fall (+)
St. Louis
Ammonium Nitrate
Fall and Spring (--)
St. Louis
Crustal
Fall (+), Winter (-), Spring (+++)
St. Louis
Mobile Sources
Fall (++), Winter (-)
Washington
Vegetative Burning and Fireworks
Summer (+), Winter (++)
Washington
Coal Combustion
Winter (++), Spring and Summer (+++)
Washington
Ammonium Nitrate and Salt
Spring (-), Winter (--)
Washington
Mobile Sources
Winter (-), Spring (++)
Washington
Canadian Fires
Winter (-), Spring (++)
Washington
Road Construction
Winter (+), Summer (--)
(+) = positive, R-squared 0.05 to 0.15
(++) = positive, R-squared 0.16 to 0.3
(+++) = positive, R-squared 0.31 to 0.5
(-) = negative, R-squared 0.05 to 0.15
(-) = negative, R-squared 0.16 to 0.3
(—) = negative, R-squared 0.31 to 0.5
Eight-Site SA Speciation Trends Final Report
34
September 24, 2003
-------
Table 6.3 Summary of Source Strength Correlation with Pressure
Site
Source
Pressure Correlation
Birmingham
Ammonium Nitrate
Mild positive correlation
Birmingham
Crustal
Slight positive correlation
Birmingham
Mobile Sources
Mild positive correlation
Birmingham
Vegetative Burning and Fireworks
Birmingham
Lead Source
Birmingham
Zinc Source
Birmingham
Coal Combustion (Ni)
Mild negative correlation
Bronx
Coal Combustion
Bronx
Oil Combustion
Bronx
Marine and Industrial Salts
Bronx
Mobile Sources with Tire Wear
Bronx
Industrial
Bronx
Ammonium Nitrate
Bronx
Crustal
Charlotte
Vegetative Burning and Fireworks
Charlotte
Coal Combustion
Charlotte
Crustal
Slight positive correlation
Charlotte
Oil Combustion
Slight negative correlation
Charlotte
Marine and Industrial Salts
Charlotte
Ammonium Nitrate
Mild positive correlation
Charlotte
Smelting
Charlotte
Mobile Sources
Houston
Crustal
Houston
Vegetative Burning and Fireworks
Houston
Industrial
Mild negative correlation
Houston
Mobile Sources
Slight positive correlation
Houston
Marine Ammonium Nitrate
Houston
Mobile Mn Source or Grain Dust
Houston
Coal Combustion (Ni)
Indianapolis
Vegetative Burning and Fireworks
Indianapolis
Ammonium Nitrate
Slight positive correlation
Indianapolis
Canadian Fires
Indianapolis
Marine and Industrial Salts
Indianapolis
Crustal
Indianapolis
Mobile Sources
Mild positive correlation
Indianapolis
Coal Combustion 1
Slight positive correlation
Indianapolis
Coal Combustion 2 (Ni)
Eight-Site SA Speciation Trends Final Report 35 September 24, 2003
-------
Table 6.3 Summary of Source Strength Correlation with Pressure (continued)
Site
Source
Pressure Correlation
Milwaukee
Coal Combustion (Ni)
Milwaukee
Mobile Sources
Slight positive correlation
Milwaukee
Crustal
Milwaukee
Chlorine Sources
Milwaukee
Ammonium Nitrate
Milwaukee
Crustal Related Events
Milwaukee
Vegetative Burning and Fireworks
Milwaukee
Industrial Diesel and Sulfate Mix
St. Louis
Zinc Refinery
Mild positive correlation
St. Louis
Smelting (Copper)
St. Louis
Coal Combustion
Slight negative correlation
St. Louis
Steel Production
St. Louis
Ammonium Nitrate
Mild positive correlation
St. Louis
Crustal
Slight negative correlation
St. Louis
Mobile Sources
Washington
Vegetative Burning and Fireworks
Washington
Coal Combustion
Washington
Ammonium Nitrate and Salt
Mild positive correlation
Washington
Mobile Sources
Washington
Canadian Fires
Washington
Road Construction
Mild positive correlation
Eight-Site SA Speciation Trends Final Report
36
September 24, 2003
-------
6.3 Day of Week and Season Summaries of the Source Strength
The seasonal and weekday versus weekend patterns also yield important clues to some
sources. Table 6.4 indicates the periods of high source strength for weekday versus weekends.
Table 6.5 indicates the season with the highest average.
Table 6.4 Summary of High Source Strength Periods for Weekdays and Weekends
Site
Source
High period
Birmingham
Ammonium Nitrate
Slightly more on weekends
Birmingham
Crustal
Weekday
Birmingham
Mobile Sources
Weekday
Birmingham
Vegetative Burning and Fireworks
Slightly more on weekdays
Birmingham
Lead Source
Weekday
Birmingham
Zinc Source
Weekday
Birmingham
Coal Combustion (Ni)
Slightly more on weekends
Bronx
Coal Combustion
Uniform
Bronx
Oil Combustion
Weekday
Bronx
Marine and Industrial Salts
Slightly more on weekends
Bronx
Mobile Sources with Tire Wear
Slightly more on weekdays
Bronx
Industrial
Uniform
Bronx
Ammonium Nitrate
Uniform
Bronx
Crustal
Weekend
Charlotte
Vegetative Burning and Fireworks
Weekday
Charlotte
Coal Combustion
Slightly more on weekdays
Charlotte
Crustal
Weekday
Charlotte
Oil Combustion
Uniform
Charlotte
Marine and Industrial Salts
Uniform
Charlotte
Ammonium Nitrate
Uniform
Charlotte
Smelting
Weekday
Charlotte
Mobile Sources
Slightly more on weekends
Houston
Crustal
Slightly more on weekends
Houston
Vegetative Burning and Fireworks
Weekday
Houston
Industrial
Slightly more on weekends
Houston
Mobile Sources
Uniform
Houston
Marine Ammonium Nitrate
Weekend
Houston
Mobile Mn Source or Grain Dust
Uniform
Houston
Coal Combustion (Ni)
Weekend
Indianapolis
Vegetative Burning and Fireworks
Weekday
Indianapolis
Ammonium Nitrate
Uniform
Indianapolis
Canadian Fires
Weekend
Indianapolis
Marine and Industrial Salts
Slightly more on weekends
Indianapolis
Crustal
Slightly more on weekdays
Indianapolis
Mobile Sources
Uniform
Indianapolis
Coal Combustion 1
Slightly more on weekdays
Indianapolis
Coal Combustion 2 (Ni)
Uniform
Eight-Site SA Speciation Trends Final Report
37
September 24, 2003
-------
Table 6.4 Summary of High Source Strength Periods for Weekdays and Weekends
(continued)
Site
Source
High period
Milwaukee
Coal Combustion (Ni)
Slightly more on weekends
Milwaukee
Mobile Sources
Slightly more on weekends
Milwaukee
Crustal
Slightly more on weekdays
Milwaukee
Chlorine Sources
Slightly more on weekends
Milwaukee
Ammonium Nitrate
Slightly more on weekdays
Milwaukee
Crustal Related Events
Uniform
Milwaukee
Vegetative Burning and Fireworks
Slightly more on weekends
Milwaukee
Industrial Diesel and Sulfate Mix
Weekday
St. Louis
Zinc Refinery
Uniform
St. Louis
Smelting (Copper)
Weekday
St. Louis
Coal Combustion
Weekend
St. Louis
Steel Production
Weekday
St. Louis
Ammonium Nitrate
Uniform
St. Louis
Crustal
Slightly more on weekends
St. Louis
Mobile Sources
Weekday
Washington
Vegetative Burning and Fireworks
Weekday
Washington
Coal Combustion
Weekday
Washington
Ammonium Nitrate and Salt
Slightly more on weekdays
Washington
Mobile Sources
Slightly more on weekends
Washington
Canadian Fires
Weekend
Washington
Road Construction
Weekday
Eight-Site SA Speciation Trends Final Report
38
September 24, 2003
-------
Table 6.5 Summary of High Source Strength Periods for Seasons
Site
Source
High Season
Birmingham
Ammonium Nitrate
Winter
Birmingham
Crustal
Fall and Spring
Birmingham
Mobile Sources
Fall
Birmingham
Vegetative Burning and Fireworks
Fall and Summer
Birmingham
Lead Source
Fall
Birmingham
Zinc Source
Uniform
Birmingham
Coal Combustion (Ni)
Summer
Bronx
Coal Combustion
Summer
Bronx
Oil Combustion
Winter
Bronx
Marine and Industrial Salts
Uniform
Bronx
Mobile Sources with Tire Wear
Spring, Summer, and Fall
Bronx
Industrial
Fall and Winter
Bronx
Ammonium Nitrate
Fall and Winter
Bronx
Crustal
Fall and Spring
Charlotte
Vegetative Burning and Fireworks
Summer
Charlotte
Coal Combustion
Spring and Summer
Charlotte
Crustal
Fall
Charlotte
Oil Combustion
Spring and Summer
Charlotte
Marine and Industrial Salts
Uniform
Charlotte
Ammonium Nitrate
Winter
Charlotte
Smelting
Fall and Winter
Charlotte
Mobile Sources
Fall and Winter
Houston
Crustal
Summer
Houston
Vegetative Burning and Fireworks
Summer
Houston
Industrial
Winter and Spring
Houston
Mobile Sources
Fall and Winter
Houston
Marine Ammonium Nitrate
Winter and Spring
Houston
Mobile Mn Source or Grain Dust
Fall and Winter
Houston
Coal Combustion (Ni)
Summer and Fall
Indianapolis
Vegetative Burning and Fireworks
Summer
Indianapolis
Ammonium Nitrate
Winter
Indianapolis
Canadian Fires
Winter
Indianapolis
Marine and Industrial Salts
Fall
Indianapolis
Crustal
Spring and Summer
Indianapolis
Mobile Sources
Fall and Summer
Indianapolis
Coal Combustion 1
Fall and Winter
Indianapolis
Coal Combustion 2 (Ni)
Spring and Summer
Eight-Site SA Speciation Trends Final Report
39
September 24, 2003
-------
Table 6.5 Summary of High Source Strength Periods for Seasons (continued)
Site
Source
High Season
Milwaukee
Coal Combustion (Ni)
Spring and Summer
Milwaukee
Mobile Sources
Fall and Summer
Milwaukee
Crustal
Uniform
Milwaukee
Chlorine Sources
Fall and Summer
Milwaukee
Ammonium Nitrate
Winter
Milwaukee
Crustal Related Events
Uniform
Milwaukee
Vegetative Burning and Fireworks
Fall and Summer
Milwaukee
Industrial Diesel and Sulfate Mix
Fall
St. Louis
Zinc Refinery
Summer, Fall, and Winter
St. Louis
Smelting (Copper)
Summer, Fall, and Winter
St. Louis
Coal Combustion
Summer
St. Louis
Steel Production
Spring, Summer, and Fall
St. Louis
Ammonium Nitrate
Winter
St. Louis
Crustal
Spring, Summer, and Fall
St. Louis
Mobile Sources
Fall
Washington
Vegetative Burning and Fireworks
Summer
Washington
Coal Combustion
Spring and Summer
Washington
Ammonium Nitrate and Salt
Winter
Washington
Mobile Sources
Fall and Summer
Washington
Canadian Fires
Summer
Washington
Road Construction
Fall
6.4 Back Trajectory Analyses
Using NOAA's HYSPLIT model, packets of air can be tracked back in time over long
distances. EPA did the HYSPLIT modeling for each of the sites. (See Appendix H for details.)
The HYSPLIT output was then analyzed in conjunction with the source strength output from the
PMF model to aid in determining possible source locations.
For each source at a site, the back trajectories were collected into three groups: days
when the source's strength was high (the days with the largest 20 percent source strength), low
(the lowest 20 percent), and medium. The conceptual model is based on the assumption that on
high source strength days the air must pass over the source. Likewise, on the majority of the low
source strength days, the path most likely did not pass over the source location. The analysis
tries to find areas that are associated with sources by considering where the various back
trajectories from the high strength days cross.
Several different methods have been proposed to quantify these ideas. They start by
superimposing a grid over the area being modeled and then considering the number of times or
probability a back trajectory path crosses into each grid cell for the various categories of source
days. The two methods most used in this report are referred to as the incremental probability
field and the source contribution function.
Eight-Site SA Speciation Trends Final Report
40
September 24, 2003
-------
For each grid cell, the incremental probability is defined as the difference between the
probability that a back trajectory from the receptor on high days passes through the grid cell and
the probability that a back trajectory from the receptor (from any day) passes through the grid
cell. For this study, high days for a source were defined to be the days with sources strengths in
the top 20 percent of the source strengths for the source.
For each grid cell, the source contribution function is defined as the conditional
probability that a back trajectory crosses the grid cell, given that the trajectory is from a high
day. This is equivalent to the ratio of the proportion of the time that a back trajectory from the
receptor on high days passes through the grid cell to the probability that a back trajectory from
the receptor (from any day) passes through the grid cell. As above, high days for a source were
defined to be the days with sources strengths in the top 20 percent of the source strengths for the
source.
The back trajectory data from HYSPLIT contain the estimated latitude and longitude of
an air packet at hourly intervals (referred to as end points). These end points are used to estimate
the probabilities by counting the number of end-points in the grid cell and dividing by the total
number of end points. This means that the probabilities are computed in such a way that they are
naturally weighted by the amount of time the trajectory spends in the grid cell.
The plots shown in Appendices D through K differ from the standard source contribution
and incremental probability plots. The difference is that they have been rescaled so that plots
from different sites can be compared. The values plotted in the source contribution plots have
been replaced with a lower end point for a Clopper-Pearson confidence interval for the
conditional probability. If the cells were not related to each other spatially, then grids without a
source would have a probability of about 20 percent and grids with a source would have a value
greater than 20 percent. As with the incremental probability, there is a random chance of
observing values more than 20 percent. So instead of plotting the raw estimate, a
Clopper-Pearson lower bound on the estimate is plotted. Hence, values above 20 percent are
significantly above 20 percent. The incremental probabilities are divided by the expected
standard error of the difference of the probabilities. Anything greater than two standard errors in
absolute value is often considered good evidence that the value is not due to random chance.
These changes essentially replace the raw plot with a weight of evidence and automatically
remove grids that have large estimates from an insignificant number of trajectories. The plots
can also be compared across sites with different amounts of data, since it is essentially the
significance level being compared.
The preference between the two types of contour plots is mostly personal. Some prefer
the incremental probability because it indicates both likely source regions and unlikely source
regions. The current version of the source contribution ties in more directly with a
weight-of-evidence presentation since it is based on a p-value. Some also prefer its less cluttered
results. At this point, there is no mathematical or statistical preference for one over the other and
each could be modified to have the advantages of the other.
For each site and source, three sets of maps are shown in Appendices D through K. The
first map shows contours of the source contribution function. The second shows the contours of
Eight-Site SA Speciation Trends Final Report
41
September 24, 2003
-------
the incremental probability metric. The third shows the paths of the various back trajectories.
This third figure is source-specific in that the trajectories are colored according to the type of
source day to which the trajectory corresponds. The blue trajectories are from the days with the
lowest 20 percent of the source strengths, the red are from the 20 percent worst days, and the
green are all the rest.
Unfortunately, the spatial relationships can complicate the natural interpretation of the
incremental probability and source contribution plots. A grid may not contain the source, but
may be located along the most common route from the source to the receptor. (The marine
source for Indianapolis may be an example. Not only are gulf coast regions highlighted, but also
several regions in between.)
In fact, just as with the wind roses, certain source categories are not well suited for this
analysis. Consider a source like crustal dust. The source is "located" virtually everywhere on
land, but may require particular winds to create a strong source-day at the receptor. Inland areas
may seem not to be associated with a high source day because air from an inland area may be
associated with winds that are too low. At the same time, a grid cell over the ocean could be
associated with the source, because air passing over the grid cell is associated with strong winds.
Nitrate sources are another example that could be misleading, because cold temperatures are
required for the formation of particulate nitrate. (Actually, the nitrate sources appear to be
associated with area sources of ammonia.) Combined sources are also problematic. Consider a
marine sulfate source. The formation of sodium sulfate may require sodium from the ocean and
a sulfur source on land. The apparent source location may be grid cells that are associated with
the combination rather than the true source locations. Finally, if the major source within the
source category is located within the receptor grid (or even within a few grid cells), the source
contribution function could appear to be less than 20 percent everywhere. Table 6.6 summarizes
the conclusions drawn from the back trajectory analyses. Finally, since the analysis is based on
80 km grid cells, local sources may not be indicated.
Eight-Site SA Speciation Trends Final Report
42
September 24, 2003
-------
Table 6.6 Summary of the Back Trajectory Plots
Site
Source
Location
Birmingham
Ammonium Nitrate
Southern AL and MS, Gulf of Mexico, Eastern TX, Northern LA,
Southern AR, Central TN and KY, IN, Southern IL, Northwestern IA,
Northern MN
Birmingham
Crustal
Southeastern AL, GA, Atlantic Ocean, SC, NC, Gulf of Mexico,
Southern VA
Birmingham
Mobile Sources
Northern GA, Western SC, Central NC, Atlantic Ocean, Central AL
and LA, Northern IA, Southern MN, Eastern OK
Birmingham
Vegetative Burning and
Fireworks
Northern GA, Eastern TN, Southern and Western VA, Southwestern
WV, Eastern KY, OH, Southern IL, Southwestern AR, Eastern OK,
Gulf of Mexico
Birmingham
Lead Source
LA, MS, AR, Southern AL and GA, Northern FL, Southern SC, Central
NC, Atlantic Ocean
Birmingham
Zinc Source
Southern AR and AL, Southeastern LA, Eastern GA, SC, NC,
Southern VA, Atlantic Ocean, Southern FL
Birmingham
Coal Combustion (Ni)
Northern GA, SC, NC, Central TN and KY, Western WV, Southeastern
MO, IN, Central IL, Northern FL, Atlantic Ocean, Gulf of Mexico
Bronx
Coal Combustion
PA, OH, Northern WV and VA, Northern IN, Southeastern Wl, Atlantic
Ocean
Bronx
Oil Combustion
PA, Northern NJ, MD, Northeastern WV, Eastern TN
Bronx
Marine and Industrial Salts
Southwestern PA, Southeastern OH, Northern VA, WV, NJ, MD,
Atlantic Ocean, Southern Wl, Northern Ml
Bronx
Mobile Sources with Tire Wear
Northern and Central VA, Rl, Eastern MA, Western OH, Central IN,
Southwestern Wl, Southern SC
Bronx
Industrial
PA, Western NY, Northern VA and WV, MD, Northern DE, NJ, OH,
Western Ml, Southern IL, Wl
Bronx
Ammonium Nitrate
PA, NJ, Northern MD, Southeastern OH, Northwestern IN, IL,
Southern Wl, Canada, Southwestern VA, Northwestern NC
Bronx
Crustal
PA, OH, MD, VA, Atlantic Ocean, WV, Northern NJ, Southern Wl
Charlotte
Vegetative Burning and
Fireworks
Northeastern VA, MD, Central WV, Eastern NC, SC, Southern GA,
Western FL, Central LA, Gulf of Mexico
Charlotte
Coal Combustion
Western VA, Eastern KY, WV, Western OH, Nothern NJ, NYC, CT,
Western PA, Northeastern IN, SC, GA, Southeastern LA, Northern FL,
Gulf of Mexico
Charlotte
Crustal
Southern NC, Eastern SC and GA, Central FL, Atlantic Ocean, NJ,
NYC, Eastern PA, AR
Charlotte
Oil Combustion
SC, GA, Central FL, Atlantic Ocean, Northern NJ, NYC, Eastern MS,
Southeastern LA
Charlotte
Marine and Industrial Salts
Atlantic Ocean, Southern GA, Gulf of Mexico, Northern KY,
Southeastern IN, Southwestern OH
Charlotte
Ammonium Nitrate
SC, Atlantic Ocean, Eastern NC and VA, MD, TN, KY, Eastern KS,
Southern MO, Central FL
Charlotte
Smelting
VA, Western NC, Northern TN, KY, Western OH, Eastern IN, Western
PA, Central IL, AR, Northeastern Wl
Charlotte
Mobile Sources
NC, SC, GA, Central FL
Eight-Site SA Speciation Trends Final Report
43
September 24, 2003
-------
Table 6.6 Summary of the Back Trajectory Plots (continued)
Site
Source
Location
Houston
Crustal
East Central TX, Eastern AR, Southern AL, Gulf of Mexico
Houston
Vegetative Burning and
Fireworks
East Central TX, LA, MS, Western AL, Western TN, Eastern GA,
Central SC. Southern IL
Houston
Industrial
East Central TX, Gulf of Mexico, Central FL, Southeastern LA
Houston
Mobile Sources
Central TX, Southern TX, LA, MS, Western AL, Western KY and TN,
Southern IL and IN, Central TN and KY
Houston
Marine Ammonium Nitrate
Central TX, Southern TX, Gulf of Mexico, Central FL
Houston
Mobile Mn Source or Grain Dust
East Central TX, Gulf of Mexico, Central FL, FL Panhandle
Houston
Coal Combustion (Ni)
Southwestern IN, MS, Western AL, Southern GA, Central SC, Central
FL, Gulf of Mexico, Southeastern LA
Indianapolis
Vegetative Burning and
Fireworks
Central IN, Western KY and TN, Northern GA, Southern AL and MS,
AR, Northern LA, Canada
Indianapolis
Ammonium Nitrate
IL, MO, Western IN, Western OH, Canada, Southern Wl, Eastern and
Southern IA, Western AR, Eastern OK, Southeastern ND, Southern
MN
Indianapolis
Canadian Fires
KY, Eastern TN, Central GA, MO, Western IA, Eastern SD and ND,
Eastern AR, Western PA
Indianapolis
Marine and Industrial Salts
Central KY and TN, Eastern IN, Western OH, Northern Wl, Eastern
MN, Northern LA, Southern AR, Eastern TX, MS, Western AL, FL
panhandle, Gulf of Mexico
Indianapolis
Crustal
AR, LA, MS, AL , Gulf of Mexico, Eastern TX, Eastern OK, Southern
MO, Western TN and KY, Central GA, Southern OH, Canada
Indianapolis
Mobile Sources
KY, Northern TN, Western NC, Northern GA, Northern LA, AR,
Southeastern NE, IN, Great Lakes, Southwestern Ml, Eastern Wl,
Canada
Indianapolis
Coal Combustion 1
Eastern KY and TN, Southern IN and IL, MO, Northeastern KS,
Southeastern NE, Southern IA
Indianapolis
Coal Combustion 2 (Ni)
KY, TN, Northern GA, Western NC, AL, MS, Northern LA, Eastern AR,
Southeastern TX, Gulf of Mexico, Southern MO, Southern IL and IN
Milwaukee
Coal Combustion (Ni)
IL, IN, IA, MO, Eastern NE and KS, Northwestern OK, Western KY
and TN, Northern MS, Central AL, Central LA, Western OH, Southern
Ml
Milwaukee
Mobile Sources
Southern AL, Northern MS, Eastern MO, Northern IL, Wl, Canada,
Southeastern MN
Milwaukee
Crustal
Northern and Western OH, Northern IN, IL, Central MO, Eastern KS
and OK, Eastern TX, Southern AL, Canada, Central IA, Eastern SD,
Southern MN, Western TN
Milwaukee
Chlorine Sources
Eastern KS and OK, Central IA, Great Lakes, Canada, Northern Ml,
Northeastern ND, Central MS
Milwaukee
Ammonium Nitrate
Eastern NE and KS, Northeastern OK, MO, IA, IL, IN, Western OH,
Southern Ml, Western KY and TN
Milwaukee
Crustal Related Events
IN, Western OH, Eastern IL, Southern Ml, Central KY and TN, Great
Lakes, Canada, Southern LA, Northern MO, Eastern KS,
Southwestern IA
Milwaukee
Vegetative Burning and
Fireworks
IL, Eastern MO and AR, MS, Southern AL, LA, Eastern KS,
Northeastern OK, Canada, Southern Ml, Western TN and KY
Milwaukee
Industrial Diesel and Sulfate Mix
IL, IN, Western OH, Western KY and TN, MO, Eastern KS,
Northeastern NE, Central IA, Canada, Central LA
Eight-Site SA Speciation Trends Final Report
44
September 24, 2003
-------
Table 6.6 Summary of the Back Trajectory Plots (continued)
Site
Source
Location
St. Louis
Zinc Refinery
IL, KY, Central WV, Western TN, Eastern KS, Northern LA, Southern
AR, Southeastern NE
St. Louis
Smelting (Copper)
KY, Western TN, AR, Southern MO, Eastern KS and OK, Eastern TX,
Southeastern NE
St. Louis
Coal Combustion
KY, TN, Southern IL, Southeastern MO, AR, LA, WV, Southwestern
MS
St. Louis
Steel Production
KY, Western TN, Northern MS, Central AR, WV, Northeastern KS,
Eastern TX, Southern IL and IN, Northern KS, Southern NE
St. Louis
Ammonium Nitrate
IL, Central KY and TN, IN, Southwestern OH, Northeastern KS,
Southeastern NE
St. Louis
Crustal
Southern MO, Western KY and TN, Northern MS, AR, LA, Eastern OK
St. Louis
Mobile Sources
Eastern KS and NE, MO, Southern LA
Washington
Vegetative Burning and
Fireworks
Central Ml, DE, MD, Southern NJ, NC, Atlantic Ocean, SC, Southern
AR, Central MO
Washington
Coal Combustion
NC, SC, VA, WV, Eastern KY and TN, OH, IN, Eastern IL,
Southwestern and Northern PA, Southern NY, Southern AR, Western
GA, Atlantic Ocean
Washington
Ammonium Nitrate and Salt
Eastern PA, Central NY, MD, DE, Southern NJ, Central TN, KY,
Southwestern WV, Northwestern OH, Central and Southern IL,
Canada
Washington
Mobile Sources
VA, NC, SC, Atlantic Ocean, Southern MD, DE, WV, Central KY,
Central and Western TN, Eastern GA, Central AL, Western IL
Washington
Canadian Fires
Central VA, Southern MD and DE, SC, Central KY, Western IL,
Northeastern MO, Southern AR, Central AL, Eastern IA
Washington
Road Construction
IN, Southwestern OH, Eastern IL, Northern KY, Central TN, Central
NC, Eastern VA and MD, DE, Northern NY
6.5 Comparisons with NOx and S02 Utility Plant Inventory Data
Sulfate is generally formed in the atmosphere from SO2 (which is why the source is often
referred to as secondary sulfate). Since the major sources of SO2 emissions are utility plants,
which are fairly well inventoried, the sulfate source locations should be compared to the utility
plant SO2 emissions as a check on the source identifications. Similarly, much of the nitrate is
formed from NOx reactions in the atmosphere with utility plants being a major source of NOx-
Hence, the nitrate source locations should also be compared with utility plant NOx emissions
inventories (although we do not expect the correlation to be as good because (a) nitrate is
semi-volatile, (b) there are other significant sources of NOx, and (c) the nitrate formation is also
dependent on NH3 emissions). Figures 6.1 and 6.2 show plots of the utility plant emissions of
SO2 and NOx across the nation.
The emissions inventories are not weighted for their impact on the receptor sites.
Smaller, nearby sources could be contributing much more to the receptor than a large, distant
source. Visual comparisons only serve as a reality check on the source identifications. Also note
that since the inventories are from the same utility plants, the locations are the same for both the
NOx and SO2 emissions plots.
Eight-Site SA Speciation Trends Final Report
45
September 24, 2003
-------
S02 tons/year
tons/yr 0 ¦ 500
1000 ¦ 5000
¦ 10000 ¦ 50000
Figure 6.1 Utility Plant Emissions of S02.
NOX tons/year
tons/yr 0 ¦ 500
1000 ¦ 5000
¦ 10000 ¦ 50000
Figure 6.2 Utility Plant Emissions of NOx-
Eight-Site SA Speciation Trends Final Report
46
September 24, 2003
-------
Comparisons of the contour maps of the various non-marine nitrate sources show a
common pattern, namely Midwest farming regions. Illinois, in particular, stands out. It has both
NOx utility emissions and the farming regions for sources of ammonia. See Figures 6.3
through 6.10 for nitrate source regions identified at the eight sites.
The comparisons of the sulfate source regions with the SO2 utility emissions did not work
as well as expected. For some of the sites, the Bronx site for instance, the back trajectories do
yield the expected source region associations with large utility emissions of SO2, namely the
Ohio River Valley and the borders of Ohio, West Virginia, and Pennsylvania. However,
complicating the analysis for the sulfate sources is that some seem to be related more to high
pressure systems (as evidenced by the clockwise swirl of many of the back trajectories for the
high source days). See Figures 6.11 through 6.20 for sulfate source regions identified at the
eight sites.
Eight-Site SA Speciation Trends Final Report
47
September 24, 2003
-------
Birmingham, AL Source 1
Source Contribution Function for High Source Strength
Figure 6.3 Nitrate Source Region Plot for Source 1, Ammonium Nitrate, at
Birmingham, Alabama.
Bronx, NX Source 6
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.4 Nitrate Source Region Plot for Source 6, Ammonium Nitrate, at
Bronx, New York.
Eight-Site SA Speciation Trends Final Report 48 September 24, 2003
-------
Charlotte, NC, Source 6
Source Contribution Function for High Source Strength
Lower Bound 0.00 ¦ 0.20
¦ 0.25 ¦ 0.30
Figure 6.5 Nitrate Source Region Plot for Source 6, Ammonium Nitrate, at
Charlotte, North Carolina.
Houston, TX, Source 5
Source Contribution Function for High Source Strength
Lower Bound 0.00 ¦ 0.20
¦ 0.25 ¦ 0.30
Figure 6.6 Nitrate Source Region Plot for Source 5, Marine Ammonium Nitrate, at
Houston, Texas.
Eight-Site SA Speciation Trends Final Report
49
September 24, 2003
-------
Indianapolis, IN, Source 2
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.7 Nitrate Source Region Plot for Source 2, Ammonium Nitrate, at
Indianapolis, Indiana.
Milwaukee, Wl, Source 5
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.8 Nitrate Source Region Plot for Source 5, Ammonium Nitrate, at
Milwaukee, Wisconsin.
Eight-Site SA Speciation Trends Final Report 50 September 24, 2003
-------
ST Louis, MO, Source 5
Source Contribution Function for High Source Strength
*
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.9 Nitrate Source Region Plot for Source 5, Ammonium Nitrate, at
St. Louis, Missouri.
Washington D.C., Source 3
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.10 Nitrate Source Region Plot for Source 3, Ammonium Nitrate and Salt, at
Washington, D.C.
Eight-Site SA Speciation Trends Final Report
51
September 24, 2003
-------
Birmingham, AL Source 7
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.11 Sulfate Source Region Plot for Source 7, Coal Combustion (Ni), at
Birmingham, Alabama.
Bronx, NX Source 1
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.12 Sulfate Source Region Plot for Source 1, Coal Combustion, at
Bronx, New York.
Eight-Site SA Speciation Trends Final Report
52
September 24, 2003
-------
Charlotte, NC, Source 2
Source Contribution Function for High Source Strength
Lower Bound 0.00 ¦ 0.20
¦ 0.25 ¦ 0.30
Figure 6.13 Sulfate Source Region Plot for Source 2, Coal Combustion, at
Charlotte, North Carolina.
Houston, TX, Source 7
Source Contribution Function for High Source Strength
Lower Bound 0.00 ¦ 0.20
¦ 0.25 ¦ 0.30
Figure 6.14 Sulfate Source Region Plot for Source 7, Coal Combustion (Ni), at
Houston, Texas.
Eight-Site SA Speciation Trends Final Report
53
September 24, 2003
-------
Indianapolis, IN, Source 7
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.15 Sulfate Source Region Plot for Source 7, Coal Combustion 1, at
Indianapolis, Indiana.
Indianapolis, IN, Source 8
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.16 Sulfate Source Region Plot for Source 8, Coal Combustion 2 (Ni), at
Indianapolis, Indiana.
Eight-Site SA Speciation Trends Final Report
54
September 24, 2003
-------
Milwaukee, Wl, Source 1
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.17 Sulfate Source Region Plot for Source 1, Coal Combustion (Ni), at
Milwaukee, Wisconsin.
Milwaukee, Wl, Source 8
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.18 Sulfate Source Region Plot for Source 8, Industrial Diesel and Sulfate Mix, at
Milwaukee, Wisconsin.
Eight-Site SA Speciation Trends Final Report
55
September 24, 2003
-------
ST Louis, MO, Source 3
Source Contribution Function for High Source Strength
Lower Bound 0.00 I 0.20
¦ 0.25 ¦ 0.30
Figure 6.19 Sulfate Source Region Plot for Source 3, Coal Combustion, at
St. Louis, Missouri.
Washington D.C., Source 2
Source Contribution Function for High Source Strength
Lower Bound 0.00 ¦ 0.20
¦ 0.25 ¦ 0.30
Figure 6.20 Sulfate Source Region Plot for Source 2, Coal Combustion, at
Washington, D.C.
Eight-Site SA Speciation Trends Final Report
56
September 24, 2003
-------
7.0 SITE-BY-SITE RESULTS
This section repeats the main results found elsewhere in the report, but it is organized by
site for convenience.
Eight-Site SA Speciation Trends Final Report
57
September 24, 2003
-------
7.1 Birmingham, Alabama
The Birmingham site (010730023) is located in an urban neighborhood in a heavily industrialized area of the city. The FRM
mass sampler had a mean mass of 18.3 [ig/nr during the period modeled. A U.S. Pipe Plant is located 1/4 mile east and northeast of the
site. A Sloss Industries Coke Plant and a Slag Wool Plant are located 3/4 mile to the north and 1 mile northeast, respectively. Finally, an
American Cast Iron Pipe Plant is located about 2 miles west-southwest of the site. Diesel trains and equipment are located south,
southeast, east, and northeast of the site. The nearest major roadway is about 30 meters away. Natural gas is the main fuel for heating,
and coal is the main fuel for electricity. The data available range from January 13, 2001, to August 9, 2002, so while summer and winter
sources should be approximately equally represented, fall sources will be underrepresented and spring sources will be overrepresented.
Table 7.1 Summary of the Birmingham, Alabama, Results
Source
Mass
>,
Mg/m
Comments
Day of week
High Season
Pollution
Rose
Back Trajectory Location
Ammonium Nitrate
1.8
(0.4)
The Se is indicative of a coal-
NOx relationship.
Slightly more
on weekends
Winter
Uniform
Southern AL and MS, Gulf of Mexico, Eastern TX,
Northern LA, Southern AR, Central TN and KY, IN,
Southern IL, Northwestern IA, Northern MN
Crustal
1.3
(0.3)
Weekday
Fall and Spring
Easterly
Southeastern AL, GA, Atlantic Ocean, SC, NC, Gulf
of Mexico, Southern VA
Mobile Sources
6.5
(1.6)
Expected; OC>EC indicates
gasoline rather than diesel
dominance; WD>WE
Weekday
Fall
Uniform
Northern GA, Western SC, Central NC, Atlantic
Ocean, Central AL and LA, Northern IA, Southern
MN, Eastern OK
Vegetative Burning
and Fireworks
1.2
(0.3)
It is assumed that if the main
event is removed, that the
remainder is vegetative
burning.
Slightly more
on weekdays
Fall and
Summer
NE, E, SE,
S, SW
Northern GA, Eastern TN, Southern and Western
VA, Southwestern WV, Eastern KY, OH, Southern
IL, Southwestern AR, Eastern OK, Gulf of Mexico
Lead Source
0.7
(0.2)
Dominated by a single event.
Weekday
Fall
E, SE
LA, MS, AR, Southern AL and GA, Northern FL,
Southern SC, Central NC, Atlantic Ocean
Zinc Source
0.8
(0.2)
Possible sources include
recycling plants, smelters,
incinerators.
Weekday
Uniform
NE, E, SE
Southern AR and AL, Southeastern LA, Eastern
GA, SC, NC, Southern VA, Atlantic Ocean,
Southern FL
Coal Combustion (Ni)
7.3
(1.8)
The sulfate and Se content
associates this with coal
burning. (See general note
regarding enhanced Ni
content.)
Slightly more
on weekends
Summer
N, NE, E,
SE, S, SW
Northern GA, SC, NC, Central TN and KY, Western
WV, Southeastern MO, IN, Central IL, Northern FL,
Atlantic Ocean, Gulf of Mexico
Eight-Site SA Speciation Trends Final Report
58
September 24, 2003
-------
7.2 Bronx, New York
The Bronx Garden site (360050083) is located in the middle of the Bronx, a heavily populated urban area. The FRM mass
sampler had a mean mass of 15.0 jig/nr during the period modeled. There are local sources that could potentially have a significant
effect on the site. These include mobile emissions, fuel oil (particularly in the winter), two oil-fired power plants, street cleaning, and
marine influence. The data available range from September 3, 2000, to January 29, 2002, so fall sources may be overrepresented.
Table 7.2 Summary of the Bronx, New York, Results
Source
Mass
(SE)
Mfl/m
Comments
Day of week
High
Season
Pollution
Rose
Back Trajectory Location
Coal Combustion
5.3
(2.4)
Key species include NH4, OC, S04,
mass. This is consistent with the
regional background/transport
sources observed in all SA
analyses done in the Northeast.
Uniform
Summer
SE, S, SW
PA, OH, Northern WV and VA, Northern IN,
Southeastern Wl, Atlantic Ocean
Oil Combustion
1.2
(0.5)
Key species include EC, OC, CI, V,
Ni. V and Ni, winter peak lead to
fuel oil combustion.
Weekday
Winter
SW, W,
NW
PA, Northern NJ, MD, Northeastern WV, Eastern TN
Marine and Industrial
Salts
0.3
(0.1)
Key species include Na, K, CI,
several metals. There is some
indication of general industrial
sources.
Slightly more
on weekends
Uniform
NE, E, SE
Southwestern PA, Southeastern OH, Northern VA,
WV, NJ, MD, Atlantic Ocean, Southern Wl, Northern
Ml
Mobile Sources with
Tire Wear
2.5
(1.1)
Key species include Na, OC>EC,
several metals. Possible mobile
source profile including tire wear.
Slightly more
on weekdays
Spring,
Summer,
and Fall
Easterly
Northern and Central VA, Rl, Eastern MA, Western
OH, Central IN, Southwestern Wl, Southern SC
Industrial
1.8
(0.8)
Key species include Zn, Ca, Se, Ni,
Pb, OC>EC. Winter peak. Note
that the sulfur and V contributions
are low while Zn, Pb, Cu, and Ca
are enhanced.
Uniform
Fall and
Winter
Westerly
PA, Western NY, Northern VA and WV, MD, Northern
DE, NJ, OH, Western Ml, Southern IL, Wl
Ammonium Nitrate
4.1
(1.8)
Key species include K, NO3, NH4,
mass. This is consistent with a
regional nitrate signature.
Uniform
Fall and
Winter
SE, S,
SW, W
PA, NJ, Northern MD, Southeastern OH, Northwestern
IN, IL, Southern Wl, Canada, Southwestern VA,
Northwestern NC
Crustal
1.0
(0.4)
Key species include K, Al, Ca, Si,
Ti. Most likely from street cleaning
and agricultural transport.
Weekend
Fall and
Spring
SE
PA, OH, MD, VA, Atlantic Ocean, WV, Northern NJ,
Southern Wl
Eight-Site SA Speciation Trends Final Report
59
September 24, 2003
-------
7.3 Charlotte, North Carolina
The Charlotte site (371190041) is located on the campus of Garinger High School. The FRM mass sampler had a mean mass
of 15.2 [ig/nr during the period modeled. The area surrounding the school is primarily residential but contains some commercial land
uses that would be associated with densely populated residential areas (convenience stores, restaurants, and other small businesses)
near intersections along the main thoroughfares. The area also contains some light industrial land uses within relatively close
proximity. The data available range from January 13, 2001, to August 6, 2002, so while summer and winter sources should be
approximately equally represented, fall sources will be underrepresented and spring sources will be overrepresented.
Probably the largest nearby source is a concrete plant approximately 1.24 miles north-northwest of the site. School buses
would be a diesel source as they service the school and are parked at the school. The buses are parked approximately 650 feet from
the monitoring site. There has been some construction at the school within the past two years. A major renovation of the main school
building was performed during the summer of 2001. Fuels for heating are primarily gas and oil, but also include electric and some
wood. Electricity in Mecklenburg County is generated primarily by coal and nuclear fuels.
Table 7.3 Summary of the Charlotte, North Carolina, Results
Source
Mass
>,
Mf|/m
Comments
Day of week
High Season
Pollution
Rose
Back Trajectory Location
Vegetative Burning
and Fireworks
0.5
(0.2)
It is assumed that if the main
event is removed, that the
remainder is vegetative
burning.
Weekday
Summer
N, NW, SE,
S
Northeastern VA, MD, Central WV, Eastern NC,
SC, Southern GA, Western FL, Central LA, Gulf of
Mexico
Coal Combustion
5.7
(2.5)
Sulfate and Se are
associated with this source
linking it to coal combustion.
Slightly more
on weekdays
Spring and
Summer
all except W
Western VA, Eastern KY, WV, Western OH,
Northern NJ, NYC, CT, Western PA, Northeastern
IN, SC, GA, Southeastern LA, Northern FL, Gulf of
Mexico
Crustal
0.6
(0.2)
Weekday
Fall
Northerly
and
Southerly
Southern NC, Eastern SC and GA, Central FL,
Atlantic Ocean, NJ, NYC, Eastern PA, AR
Oil Combustion
1.9
(0.8)
Ba may be a useful tracer for
power plants.
Uniform
Spring and
Summer
Westerly
SC, GA, Central FL, Atlantic Ocean, Northern NJ,
NYC, Eastern MS, Southeastern LA
Marine and Industrial
Salts
0.1
(0.0)
Winds support this
conclusion.
Uniform
Uniform
Uniform
Atlantic Ocean, Southern GA, Gulf of Mexico,
Northern KY, Southeastern IN, Southwestern OH
Eight-Site SA Speciation Trends Final Report
60
September 24, 2003
-------
Source
Mass
>,
Mf|/m
Comments
Day of week
High Season
Pollution
Rose
Back Trajectory Location
Ammonium Nitrate
1.2
(0.5)
The Se in the factor
associates it with coal
combustion.
Uniform
Winter
all except W
SC, Atlantic Ocean, Eastern NC and VA, MD, TN,
KY, Eastern KS, Southern MO, Central FL
Smelting
0.7
(0.3)
Copper, Zinc, and EC typical
of smelting/metal production.
Weekday
Fall and
Winter
SE, S, SW,
W, NW
VA, Western NC, Northern TN, KY, Western OH,
Eastern IN, Western PA, Central IL, AR,
Northeastern Wl
Mobile Sources
3.9
(1.7)
We expect mobile sources,
however the weekday pattern
does not support it.
Slightly more
on weekends
Fall and
Winter
Southerly
NC, SC, GA, Central FL
Eight-Site SA Speciation Trends Final Report
61
September 24, 2003
-------
7.4 Houston, Texas
The Houston site chosen was the Aldine Road site (482010024). The PM2.5 mass from the speciation sampler had a mean
mass of 14.2 jig/nr during the period modeled. This site is not as heavily impacted by the ship channel as other sites in the Houston
area and, hence, should be more representative of other urban areas around the nation. It was expected to be affected by sources that
would be associated with an urban area. In particular, mobile emissions should be significant. The data available range from
August 17, 2000, to July 7, 2001, so all sources should be represented approximately equally.
Table 7.4 Summary of the Houston, Texas, Results
Source
Mass
(SEk
|jg/m
Comments
Day of week
High Season
Pollution
Rose
Back Trajectory Location
Crustal
0.8
(0.7)
Usual crustal elements.
Slightly more
on weekends
Summer
SW
East Central TX, Eastern AR, Southern AL,
Gulf of Mexico
Vegetative Burning and
Fireworks
0.5
(0.4)
The peak is for July 4. The July 5
estimate is about half of the July 4
value. The wintertime portion may
be consistent with wood smoke.
Weekday
Summer
w
East Central TX, LA, MS, Western AL,
Western TN, Eastern GA, Central SC.
Southern IL
Industrial
0.9
(0.7)
The chlorine content associates
this with local industrial sources.
Slightly more
on weekends
Winter and
Spring
SE, S
East Central TX, Gulf of Mexico, Central FL,
Southeastern LA
Mobile Sources
5.2
(4.4)
This site is in a residential
neighborhood with freeways to the
north.
Uniform
Fall and Winter
Northerly
Central TX, Southern TX, LA, MS, Western
AL, Western KY and TN, Southern IL and IN,
Central TN and KY
Marine Ammonium
Nitrate
0.3
(0.2)
This could be a marine influenced
profile from the gulf or bay on
which sodium nitrate has formed as
the air parcels pass over the
emissions sources. That would
explain the absence of ammonium
and sulfur.
Weekend
Winter and
Spring
NW, SE
Central TX, Southern TX, Gulf of Mexico,
Central FL
Mobile Mn Source or
Grain Dust
1.0
(0.9)
The Mn signature may indicate
off-road diesel or it could be grain
dust with an Mn anti-fungal coating
with other ship channel sources.
Uniform
Fall and Winter
NW, SE
East Central TX, Gulf of Mexico, Central FL,
FL Panhandle
Coal Combustion (Ni)
5.5
(4.7)
The Se associates this with coal
combustion. (See Section 4.3
regarding enhanced Ni content.)
Weekend
Summer and
Fall
Easterly
Southwestern IN, MS, Western AL, Southern
GA, Central SC, Central FL, Gulf of Mexico,
Southeastern LA
Eight-Site SA Speciation Trends Final Report
62
September 24, 2003
-------
7.5 Indianapolis, Indiana
The Indianapolis site (180970078) is in a residential area that is northeast of the central core of the city. The FRM mass
sampler had a mean mass of 16.6 jig/nr during the period modeled. The area is highly populated. The site is in a parking lot next to a
police station and a city park. There is some light industry in the area including a printing operation to the south of the site. The main
fuels are natural gas and oil burning home heating furnaces. Electricity is provided by power plants in the southern part of the city and
state. The data available range from December 20, 2000, to August 6, 2002, so while summer and winter sources should be
approximately equally represented, fall sources will be underrepresented and spring sources will be overrepresented.
Table 7.5 Summary of the Indianapolis, Indiana, Results
Source
Mass
(SE)>,
[iglm
Comments
Day of week
High
Season
Pollution
Rose
Back Trajectory Location
Vegetative Burning
and Fireworks
0.7
(0.2)
It is assumed that if the main
event is removed, that the
remainder is vegetative burning.
Weekday
Summer
NW
Central IN, Western KY and TN, Northern GA,
Southern AL and MS, AR, Northern LA, Canada
Ammonium Nitrate
3.6
(1.1)
Uniform
Winter
E, W
IL, MO, Western IN, Western OH, Canada, Southern
Wl, Eastern and Southern IA, Western AR, Eastern
OK, Southeastern ND, Southern MN
Canadian Fires
0.3
(0.1)
Coincides with transport from
large known fire event.
Weekend
Winter
NW, NE
KY, Eastern TN, Central GA, MO, Western IA,
Eastern SD and ND, Eastern AR, Western PA
Marine and Industrial
Salts
0.5
(0.1)
Note the substitution of chloride
with nitrate during transport from
the Gulf.
Slightly more
on weekends
Fall
Southerly
Central KY and TN, Eastern IN, Western OH,
Northern Wl, Eastern MN, Northern LA, Southern
AR, Eastern TX, MS, Western AL, FL panhandle,
Gulf of Mexico
Crustal
0.5
(0.2)
Slightly more
on weekdays
Spring and
Summer
SW
AR, LA, MS, AL , Gulf of Mexico, Eastern TX,
Eastern OK, Southern MO, Western TN and KY,
Central GA, Southern OH, Canada
Mobile Sources
3.2
(1.0)
Expect mobile sources. Note
that OC>EC indicates gasoline
rather than diesel dominance,
however the day of week pattern
is not supportive.
Uniform
Fall and
Summer
Uniform
KY, Northern TN, Western NC, Northern GA,
Northern LA, AR, Southeastern NE, IN, Great Lakes,
Southwestern Ml, Eastern Wl, Canada
Coal Combustion 1
1.6
(0.5)
EC, Se and wintertime peak
similar to findings from Poirot.
Slightly more
on weekdays
Fall and
Winter
NE, E,
SE, S, SW
Eastern KY and TN, Southern IN and IL, MO,
Northeastern KS, Southeastern NE, Southern IA
Coal Combustion 2 (Ni)
7.0
(2.3)
See Section 4.3 regarding the
enhanced Ni content.
Uniform
Spring and
Summer
NE, E,
SE, S, SW
KY, TN, Northern GA, Western NC, AL, MS,
Northern LA, Eastern AR, Southeastern TX, Gulf of
Mexico, Southern MO, Southern IL and IN
Eight-Site SA Speciation Trends Final Report
63
September 24, 2003
-------
7.6 Milwaukee, Wisconsin
The Milwaukee site (550790026) is located on a wooden stand 4 feet off the ground on Southeast Region Headquarters
parking lot. It is about 100 feet from Dr. Martin Luther King Jr. Drive, which is the closest street. In addition, North Avenue, the
intersection of North Avenue and Dr. Martin Luther King Jr. Drive, and Interstate 43 are possible local mobile sources. The
surrounding area is primarily commercial and residential. The FRM mass sampler had a mean mass of 13.4 jig/nr during the period
modeled. Natural gas is the most widely used fuel for cooking and heating. The data available range from December 14, 2000, to
September 8, 2002, so while summer and winter sources should be approximately equally represented, fall sources will be
underrepresented and spring sources will be overrepresented.
Table 7.6 Summary of the Milwaukee, Wisconsin, Results
Source
Mass,
M9 lm
Comments
Day of week
High
Season
Pollution
Rose
Back Trajectory Location
Coal Combustion (Ni)
4.5
(2.1)
See Section 4.3 regarding
the enhanced Ni content.
Slightly more
on weekends
Spring and
Summer
Southerly
IL, IN, IA, MO, Eastern NE and KS, Northwestern OK,
Western KY and TN, Northern MS, Central AL, Central
LA, Western OH, Southern Ml
Mobile Sources
1.5
(0.7)
OC>EC indicates gasoline
rather than diesel
dominance, however the
day of week pattern is not
supportive.
Slightly more
on weekends
Fall and
Summer
S, SW, W,
NW
Southern AL, Northern MS, Eastern MO, Northern IL,
Wl, Canada, Southeastern MN
Crustal
0.1
(0.1)
Slightly more
on weekdays
Uniform
Southerly
Northern and Western OH, Northern IN, IL, Central MO,
Eastern KS and OK, Eastern TX, Southern AL, Canada,
Central IA, Eastern SD, Southern MN, Western TN
Chlorine Sources
2.7
(1.3)
May be from industrial
sources.
Slightly more
on weekends
Fall and
Summer
Southerly
Eastern KS and OK, Central IA, Great Lakes, Canada,
Northern Ml, Northeastern ND, Central MS
Ammonium Nitrate
4.1
(1.9)
Slightly more
on weekdays
Winter
Southerly
Eastern NE and KS, Northeastern OK, MO, IA, IL, IN,
Western OH, Southern Ml, Western KY and TN
Crustal Related
Events
0.2
(0.1)
Mainly from three events.
Uniform
Uniform
SE, S, SW,
W, NW
IN, Western OH, Eastern IL, Southern Ml, Central KY
and TN, Great Lakes, Canada, Southern LA, Northern
MO, Eastern KS, Southwestern IA
Vegetative Burning
and Fireworks
0.4
(0.2)
It is assumed that if the
main event is removed, that
the remainder is vegetative
burning.
Slightly more
on weekends
Fall and
Summer
NE, SW
IL, Eastern MO and AR, MS, Southern AL, LA, Eastern
KS, Northeastern OK, Canada, Southern Ml, Western
TN and KY
Industrial Diesel and
Sulfate Mix
0.9
(0.4)
Weekday
Fall
NE, S, SW
IL, IN, Western OH, Western KY and TN, MO, Eastern
KS, Northeastern NE, Central IA, Canada, Central LA
Eight-Site SA Speciation Trends Final Report
64
September 24, 2003
-------
7.7 St. Louis, Missouri
The St. Louis site is the Blair Street site (295100085). This site is located near the intersection of several highways, so mobile
emissions should be a major component. The FRM mass sampler had a mean mass of 16.9 jig/nr during the period modeled. There
are several municipal incinerators, a zinc smelter, a very large lead smelter, a steel mill, cement manufacturing, and limestone
quarrying in the area. The data available range from August 4, 2000, to July 12, 2001, so all sources should be represented
approximately equally.
Table 7.7 Summary of the St. Louis, Missouri, Results
Source
Mass,
MFl/m
Comments
Day of week
High Season
Pollution
Rose
Back Trajectory Location
Zinc Refinery
0.9
(0.2)
Big River Zinc Corporation is
located 5-10 miles to the SE.
Uniform
Summer, Fall,
and Winter
N, NE, E,
SE
IL, KY, Central WV, Western TN, Eastern KS,
Northern LA, Southern AR, Southeastern NE
Smelting (Copper)
0.6
(0.1)
Cerro Copper Products
Company is located 5-10 miles
to the SE.
Weekday
Summer, Fall,
and Winter
Easterly
KY, Western TN, AR, Southern MO, Eastern KS
and OK, Eastern TX, Southeastern NE
Coal Combustion
5.7
(1.2)
Consistent with power
generation. Does not show a
seasonal trend.
Weekend
Summer
NE, E, SE,
S, SW
KY, TN, Southern IL, Southeastern MO, AR, LA,
WV, Southwestern MS
Steel Production
0.8
(0.2)
Granite City Steel may
contribute to high Fe levels.
Weekday
Spring,
Summer, and
Fall
Easterly
KY, Western TN, Northern MS, Central AR, WV,
Northeastern KS, Eastern TX, Southern IL and IN,
Northern KS, Southern NE
Ammonium Nitrate
5.0
(1.1)
NOx from power plants. Power
plant to the southeast.
Uniform
Winter
Northerly
IL, Central KY and TN, IN, Southwestern OH,
Northeastern KS, Southeastern NE
Crustal
1.4
(0.3)
High Ca, K relative to typical
crustal. Possibility cement
plant or limestone quarrying,
but peaks probably coincide
with agricultural activity.
Slightly more
on weekends
Spring,
Summer, and
Fall
S, SW
Southern MO, Western KY and TN, Northern MS,
AR, LA, Eastern OK
Mobile Sources
2.9
(0.6)
High Pb possible because of
residue (in road dust) from old
Pb smelter emissions and
hauling w/o tarps.
Weekday
Fall
NE, E, SE,
S, SW
Eastern KS and NE, MO, Southern LA
Eight-Site SA Speciation Trends Final Report
65
September 24, 2003
-------
7.8 Washington, D.C.
The Washington, D.C., site (110010043) is the McMillan Site. It is located within a fenced property that surrounds the
McMillan Reservoir (a water storage facility for the District of Columbia). The trailer is in the middle of a large field approximately
50 to 70 yards from the Lake shore, which is directly west. Approximately 2.6 miles to the south is the U.S. Capitol. The FRM mass
sampler had a mean mass of 16.6 jig/nr during the period modeled. The data available range from April 7, 2001, to August 6, 2002,
so summer sources might be overrepresented.
There is a small municipal parking lot directly to the southwest of the trailer where approximately 10 to 20 diesel vehicles
owned by the Department of Public Works are parked. If all these vehicles start up at the same time, a local microscale diesel event
might be produced. However, there is an R&P TEOM operating at the McMillan Site (30-minute time resolution), and it has not seen
any extreme peaks of mass. North Capitol is the closest major street, which can have over 40,000 vehicles per day. There are
numerous highways serving the area. The main fuels for the area are fuel oil and natural gas. Outside the District and within a
50-mile radius are five coal-fired power generation facilities. Four facilities are to the southwest and southeast, and one facility is to
the northwest of the McMillan site. There are steel and aluminum facilities 30 to 40 miles to the northwest in
Frederick County, Maryland. The data may also be affected by a major highway construction project approximately 15 miles to the
southwest.
Eight-Site SA Speciation Trends Final Report
66
September 24, 2003
-------
Table 7.8 Summary of the Washington, D.C., Results
Source
Mass,
Mg/m3
Comments
Day of week
High
Season
Pollution
Rose
Back Trajectory Location
Vegetative Burning
and Fireworks
0.5
(0.3)
It is assumed that if the main event is
removed, that the remainder is
vegetative burning.
Weekday
Summer
NW, N
Central Ml, DE, MD, Southern NJ, NC,
Atlantic Ocean, SC, Southern AR, Central
MO
Coal Combustion
7.7
(3.6)
Weekday
Spring and
Summer
N, NE, E,
SE, S, SW
NC, SC, VA, WV, Eastern KY and TN, OH,
IN, Eastern IL, Southwestern and Northern
PA, Southern NY, Southern AR, Western
GA, Atlantic Ocean
Ammonium Nitrate
and Salt
1.2
(0.6)
Has NaCI and may have some
substitution of chloride with nitrate.
Possibly a mix with road salt.
Slightly more
on weekdays
Winter
Easterly
Eastern PA, Central NY, MD, DE, Southern
NJ, Central TN, KY, Southwestern WV,
Northwestern OH, Central and Southern IL,
Canada
Mobile Sources
4.7
(2.2)
Local and transported pollutants:
gasoline dominant (OC>EC),
however the day of week pattern is
not as expected. May also include
power plant combustion, note Se, Ni,
V, and sulfate.
Slightly more
on weekends
Fall and
Summer
NE, E, SE, S,
SW
VA, NC, SC, Atlantic Ocean, Southern MD,
DE, WV, Central KY, Central and Western
TN, Eastern GA, Central AL, Western IL
Canadian Fires
1.1
(0.5)
Coincides with transport from large
known fire event.
Weekend
Summer
N, SW
Central VA, Southern MD and DE, SC,
Central KY, Western IL, Northeastern MO,
Southern AR, Central AL, Eastern IA
Road Construction
1.5
(0.7)
Crustal component with diesel
influence. Note EC, metals, and Mn
plus day of week pattern (WD>WE).
Weekday
Fall
NE, E, SE, S
IN, Southwestern OH, Eastern IL, Northern
KY, Central TN, Central NC, Eastern VA
and MD, DE, Northern NY
Eight-Site SA Speciation Trends Final Report
67
September 24, 2003
-------
8.0
INTER-SITE ANALYSES
See separate draft report entitled "Estimation of PM2.5 Transport in the Eastern United
States" dated September 2003.
9.0 CONCLUSIONS
This source apportionment and back trajectory study analyzes speciated PM2.5 data from
eight of EPA's Trends Sites located in Birmingham, Alabama; Bronx, New York;
Charlotte, North Carolina; Houston, Texas; Indianapolis, Indiana; Milwaukee, Wisconsin;
St. Louis, Missouri; and Washington, D.C. Unlike previous studies of IMPROVE and
CASTNET data, each of these is in an urban area that is expected to include strong local effects
as well as effects from long-range transport. The results of both the source apportionment and
back trajectory analyses are consistent with this expectation.
While the combination of source apportionment techniques, local meteorological
analysis, and back trajectory methods provide a very useful means of understanding the PM2.5
sources, there are some limitations:
• Sufficient data are needed with a sufficient number of measured species that are
observed at levels above the MDL. The data available did not allow the mobile
sources to be apportioned into separate diesel and non-diesel components.
• The wind and pollution roses are based on low-level winds from "nearby" weather
stations. These can be highly variable within an urban area. Even co-located wind
information can be misleading if interpreted too literally.
• The back trajectory methods require careful interpretation and need to have as many
reality checks as possible. They are based on modeling back trajectories of air
packets that start at 500 m above the site and use gridded meteorological data that
have a three-hour time resolution and 80 km grid cells. Confounding factors, such as
sources and data that are dependent on meteorological conditions, can lead to
incorrect conclusions. Further, local sources may be missed entirely by these
methods because of the spatial resolution of the data.
• Different sites have differing time periods over which speciated data were available.
As a result, some sites may have more warm seasons or cold seasons represented than
other sites. This unequal representation of seasons may result in overstatement of the
contribution from a seasonal source when that source's season is overrepresented.
Hence, it is necessary to use a weight-of-evidence approach to understanding the results with as
many independent checks of the conclusions as possible and careful checks on the modeling.
Eight-Site SA Speciation Trends Final Report
68
September 24, 2003
-------
For each site, the PM2.5 was apportioned into six to eight sources. While the species were
chosen to be consistent across the sites, the number of sources used in the modeling was allowed
to vary between sites. Eight sources may be the limit of the model for the amount of data that
were available. There were several commonly identified sources, each of which was expected to
affect the receptor. Table 9.1, at the end of this section, summarizes the sources with the
common source categories grouped together.
• For each site, a coal combustion source was identified with a mean mass of between
4.5 and 7.7 (ig/m3. These include selenium that is associated with coal burning.
Some of these sources have enhanced nickel content compared to the coal combustion
profiles found at rural sites. This may mean that some oil burning has been
apportioned to these sources. However, it may not. There is some preliminary
indication from transport analyses that some of the trace metals may be preferentially
removed from the PM2.5 fraction resulting in relatively lower concentrations further
from the source. That is to say that the presence of additional amounts of Ni (as well
as Ta and V) in the profile may only be an indication of nearby coal combustion. The
back trajectory analyses for these sources are somewhat mixed. The back trajectory
analysis corresponds well to the utility plants in the Midwest, Southeast, and eastern
seashore. To some extent in St. Louis and to a greater extent in Houston, the high
concentrations of sulfate are partially related to the effects of high pressure systems.
• For each site, a mobile source was identified with a mean mass of 2.5 to 6.5 (ig/rn3.
For Houston, in addition to the main mobile source with a mass of 5.2 (ig/m3, there
was an additional source with a mean mass of 1.0 (ig/m3 that may be mobile related.
This source is high in OC (organic carbon, usually associated mobile sources) and
with significant amounts of Mn (sometimes associated with off-road diesel from the
additive MMT). However, this source could be grain dust with a Mn-based
antifungal coating from the ship channel. Further refinement of the carbon sources
would benefit all sites, but particularly the Houston site. Finally, the profile for the
mobile source in St. Louis contains an unusually high amount of lead (for current
mobile sources) that is probably related to a historical problem with lead in the area.
• Each site also had a small crustal dirt source with a mean mass between 0.3 (ig/m3
and 1.5 (ig/m3. The 1.5 (ig/m3 source is for Washington, D.C.; it also contains diesel
components and is probably tied to a large road construction project under way
during the period modeled. For St. Louis, the crustal material may be supplemented
by point sources such as cement manufacturing.
• Houston had a very small nitrate source that was associated with a marine profile.
The other sites had nitrate sources that ranged from 1.2 to 5.0 (ig/m3. For the sites
other than Houston, the back trajectories indicate Midwestern source regions that
would be associated with agricultural ammonia emissions. Illinois, in particular,
stands out among the source regions. This should be expected, since Illinois has both
NOx utility emissions and the farming regions for sources of ammonia.
Eight-Site SA Speciation Trends Final Report
69
September 24, 2003
-------
• Bronx, Charlotte, Houston, and Indianapolis each had small marine and industrial
salt sources. The largest is for Indianapolis, but the source profile shows signs of
nitrate substitution for the chlorine during transport.
• A source clearly dominated by fireworks was found for Birmingham, Charlotte,
Houston, Indianapolis, Milwaukee, and Washington, D.C. These sources are all very
similar in size (-0.5 (ig/m3) except for Birmingham, which is twice as large as the
others (1.2 (ig/rn3 ). Because of the similarities in the source profiles to vegetative
burning, these sources should include any vegetative burning in the areas. The
source name, "Vegetative burning and fireworks," was chosen to reflect the more
frequent of the two sources.
• Sources that appear to be related to industrial activity were found in Birmingham,
Bronx, and Houston.
• Both Bronx and Charlotte had oil combustion sources with masses of 1.2 and
1.9 (ig/m3, respectively.
• Charlotte and St. Louis had zinc sources with each having masses of 0.9 (ig/m3. The
pollution rose for the St. Louis source is consistent with a local zinc refinery. In
addition, St. Louis had a copper smelting (0.6 (ig/m3 ) and steel production
(0.8 (ig/m3) source.
• Finally, there was a huge spike in the PM2.5 mass on July 7, 2002, in
Washington, D.C., that is associated with Canadian forest fires. This source is
apportioned over 1 (ig/m3 of the 16.6 (ig/m3 of mass observed during the modeled
period. The Indianapolis site was also affected by these fires, but to a much lesser
extent.
As indicated above, the back trajectory analyses and wind/pollution roses for the sites
yield source location information for the apportioned source categories. There had been some
concern that the back trajectories would not work for nitrate sources, but rather just show an
association with cool air from the north. The multiple sites within this study show that while this
might be true to some extent, comparisons of the back trajectory contour maps of the various
non-marine nitrate sources show a very common pattern of association. The nitrate sources are
associated with the Midwest farming regions.
The comparisons of the coal combustion source regions with the SO2 utility emissions
did not work as well as expected. For some of the sites, the Bronx site for instance, the back
trajectories do yield the expected source region associations with large utility emissions of SO2,
namely the Ohio River Valley and the borders of Ohio, West Virginia, and Pennsylvania.
Further complicating the analysis for the sulfate sources is that some seem to be related more to
high pressure systems (as evidenced by the clockwise swirl of many of the back trajectories for
the high source days). With additional data, it should be expected that the tools would separate
the coal combustion sources into separate meteorological regimes, as in the case of Indianapolis
and other IMPROVE sites.
Eight-Site SA Speciation Trends Final Report
70
September 24, 2003
-------
The various analyses are generally self-consistent, consistent among analysis types,
consistent with expectations for the sites, and consistent from site-to-site. Taken together, they
show that a monitoring and modeling combination provides an effective means of understanding
the source categories affecting urban areas. The coal combustion sources account for about
one-third of the PM2.5. The next largest portion is either from nitrate or mobile sources. All
three of these source categories show transport components. Additional study of the mobile
sources could be beneficial through the addition of VOCs, speciated PM carbon data, or finer
carbon fractions in the source apportionment. After the three main sources, the smaller sources
are more site-specific except for crustal dust. The ability to separate and identify these is likely
to be data dependent. Up to eight sources that can include marine influences, metal production,
general industrial, and oil combustion are within the range of resolvability with approximately
one year of speciation data at current levels of technology. Additional source resolution should
be possible with longer data streams or additional carbon species.
Eight-Site SA Speciation Trends Final Report
71
September 24, 2003
-------
Table 9.1 Summary of the Mean Apportioned Mass Across Sites
Major Source
Categories
Mean apportioned mass: ng/m3 (%total)
Birmingham
Bronx
Charlotte
Houston
Indianapolis
Milwaukee
St. Louis
Washington
Ammonium Nitrate
1.84 (9.4%)
4.09 (25.4%)
1.21 (7.5%)
3.58 (20.7%)
4.07 (28.1%)
5.02 (29.2%)
1.23 (7.4%)
Canadian Fires
0.25 (1.5%)
1.11 (6.7%)
Coal Combustion
7.27 (37.2%)
5.29 (32.9%)
5.71 (35.4%)
5.54 (39.1%)
8.67 (50.1%)
4.54 (31.3%)
5.74 (33.4%)
7.70 (46.2%)
Crustal
1.27 (6.5%)
0.97 (6.0%)
0.57 (3.5%)
0.77 (5.4%)
0.51 (3.0%)
0.31 (2.1%)
1.43 (8.3%)
1.47 (8.8%)
Industrial
1.50 (7.7%)
1.82 (11.3%)
0.87 (6.1%)
2.66 (18.4%)
Marine
0.30 (1.9%)
0.08 (0.5%)
0.29 (2.0%)
0.47 (2.7%)
Metal production
0.67 (4.2%)
2.20 (12.8%)
Mobile Source or
Grain dust
1.04 (7.3%)
Mobile sources
6.51 (33.4%)
2.49 (15.5%)
3.87 (24.0%)
5.19 (36.7%)
3.21 (18.5%)
2.46 (17.0%)
2.92 (17.0%)
4.72 (28.3%)
Oil combustion
1.22 (7.6%)
1.87 (11.6%)
Vegetative Burning
and Fireworks
1.15 (5.9%)
0.48 (3.0%)
0.49 (3.5%)
0.69 (4.0%)
0.35 (2.5%)
0.53 (3.2%)
Total mass being
apportioned (ng/m3)
19.53
16.08
16.15
14.16
17.29
14.47
17.19
16.67
Eight-Site SA Speciation Trends Final Report
72
September 24, 2003
-------
10.0 REFERENCES
Coutant, B., et al. (2002). "Source Apportionment Analysis of Air Quality Monitoring Data:
Phase 1", Final Report to the Mid-Atlantic/Northeast visibility Union and Midwest Regional
Planning Organization; Battelle, Columbus, Ohio; May.
Henry, R.C. (1997). "History and Fundamentals of Multivariate Air Quality Receptor Model,"
Chemometrics and Intelligent Laboratory Systems, 37:525-530.
Miller, Albert (1971). Meteorology Second Edition. Merrill Physical Sciences Series, Charles
E. Merrill Publishing Company, Columbus, Ohio.
Missouri Department of Natural Resources website: http ://www. dnr. state. mo .us/en v/ here .htm.
Paatero, Pentti (2000). "User's Guide for Positive Matrix Factorization Programs PMF2 and
PMF3, Part 1: Tutorial", guide distributed with purchase of software only, Copyright 1998;
University of Helsinki, Finland; last changed on February 25.
Song, Xin-Hua, Polissar, Alexandr V., and Hopke, Philip K. (2001). "Sources of Fine Particle
Composition in the Northeastern U.S.," Atmospheric Environment, 35, pp 5277-5286.
Technical memorandum to Dr. Basil Coutant, Battelle, from Hilary Main, Sonoma Technology,
Inc.; STI Ref. No. 902310; dated June 28, 2002.
Versar, Inc., (2002). "Exposure Assessment for Sources of Dioxin-Like Compounds in the
United States using the RELMAP Modeling System." Interim Report to U.S. EPA, National
Center for Environmental Assessment; Versar, Inc., Columbia, Maryland; April 30.
Wei (1990). Time Series Analysis. Addison-Wesley Publishing Company, Inc.,
Redwood City, California.
Willis, R.D. (2000). "Workshop on UNMIX and PMF as Applied to PM2.5, February 14-16,
2000, U.S. EPA, Research Triangle Park, North Carolina." Report to EPA National Exposure
Research Laboratory, Human Exposure and Atmospheric Sciences Division, Contract No. 68-
D5-0049, June.
Eight-Site SA Speciation Trends Final Report
73
September 24, 2003
------- |