United States
Environmental Protection
Agency
Environmental Sciences Research
Laboratory
Research Triangle Park NC 27711
Research and Development
EPA-600/S2-80-195 Apr. 1981
Project Summary
Modified Factor
Analysis of Selected
RAPS Aerosol Data
Stephanie C. Davis
Target transformation factor anal-
ysis (TTFA) has been applied to a
subset of the aerosol-composition
data acquired during the Regional Air
Pollution Study (RAPS) for St. Louis,
Missouri. The RAPS program collected
a large number of samples with 10
continuously operated dichotomous
samplers from March 1975 to March
1977. The purpose of the present
study was to evaluate the capability of
TTFA to resolve sources of airborne
particulate matter in a large set of
ambient-aerosol samples. Only the
samples from July and August 1976,
both fine and coarse fractions, were
examined in this study. To determine
the most appropriate way to apply
TTFA, two separate sets of data were
analyzed—all the samples collected
.during the two months at a single
station and all the samples collected
during a single week at all 10 RAPS
stations. A more detailed source res-
olution was obtained when the data
were further divided into subsets
corresponding to the fine- and coarse-
particle fractions. Because of the large
number of different sources in the St.
Louis area, superior results were ob-
tained from the examination of the
variation in aerosol composition with
time at a single location rather than
the spatial variation over multiple sites
during a shorter time period.
This Project Summary was devel-
oped by EPA's Environmental Sciences
Research Laboratory, Research Tri-
angle Park. NC. to announce key
findings of the research project that is
fully documented in a separate report
of the same title (see Project Report
ordering information at back).
Introduction
In 1971, the first National Ambient
Air Quality Standards were established,
setting limits on permissible levels of
total suspended particulate matter.
Because these regulations concentrated
only on the absolute amounts of sus-
pended particulate matter, control efforts
were directed primarily toward the
removal of the larger, more massive
particles that comprise most of the total
mass of particle emissions. There is
growing evidence, however, that the
smaller particles, although represent-
ing a smaller fraction of the total mass,
present the greater threat to public
health and welfare. Because of their
light-scattering properties, the smaller
particles also contribute to a larger
extent to the reduction in visibility. As a
result of the growing concern over the
possible health effects of smaller diam-
eter particulate matter in the atmos-
phere, there has been an increase in the
development and use of aerosol sam-
pling equipment that measures the
concentrations of particles in the in-
halable and fine particle-size fractions.
New regulations, now under review by
the Environmental Protection Agency,
may include standards not only on the
mass of total suspended particles but
also on the levels of inhalable and fine
particles.
-------
The effectiveness of these regula-
tions would be enhanced greatly if the
sources and relative emission rates of
these particles were known. Determin-
ing the sources of airborne paniculate
matter is a very difficult problem be-
cause of the complexity of urban eco-
systems. The problem is aggravated by
the presence of fugitive sources and the
lack of understanding of the modes of
formation and transport of airborne par-
ticles. The advent of receptor meth-
odology and recent developments in the
areas of trace-element analysis and
multivariate statistical techniques now
permit a much more detailed analysis of
ambient aerosol samples. By providing
detailed information on the sources of
fine and inhalable particles, such tech-
niques could become a major part of any
strategy for controlling airborne panic-
ulate matter.
The Study
One approach to the problem of
determining the contribution of each
source to the total level of suspended
particulate matter is the method of
source-emissions inventories and at-
mospheric dispersion models. Using
estimated emission rates for all known
sources along with a model for their
behavior following emission, this method
attempts to determine the contribution
of each source to the total aerosol
concentration. The results of source-
emissions inventories are too often
unsatisfactory because of a lack of
accurate source-emission profiles and
the inability to include fugitive sources.
The development of improved aerosol
sampling equipment permittedthe
development of receptor methods for
the determination of source contribu-
tions. These methods employ statistical
analysis techniques to calculate the
ambient aerosol samples. Receptor
methods have primarily taken the form
of the chemical element balance (CEB)
method. In this method, it is assumed
that the number and composition of the
sources are known. The observed ele-
mental concentrations in a set of am-
bient aerosol samples are fit by means
of a regression analysis. Early applica-
tions of the chemical element balance
were to ambient aerosol samples from
Pasadena, California; Ghent, Belgium;
Heidelberg, Germany; and Chicago,
Illinois.
These early applications of the chem-
ical element balance method suffered
from a lack of reliable data for the source
emission profiles needed for the anal-
ysis. Recent improvements in the quality
of source profile determinations have
produced good agreement between the
calculated and measured elemental
concentrations for airborne particulate
matter in Washington, DC, and for
dichotomous sampler data in St. Louis,
Missouri, and the Great Smoky Moun-
tains. The scope of the chemical element
balance method was further expanded
in an extensive study of the Portland,
Oregon, aerosol as part of the PACS
program. This study included compo-
nents such as vegetative and grass
burning and chemical species such as
volatile and nonvolatile carbon, nitrate,
and sulfate that had not previously been
used. By including the larger number of
source components, very good agree-
ment with the measured concentrations
was achieved for the 23 elements and
chemical species included in the anal-
ysis.
Chemical element balance methods
have a major drawback in that they
require an a priori knowledge of both the
number and composition of the sus-
pected source emissions. The only
quantities calculated are the contribu-
tion of each of the presumed sources to
each of the measured samples. Though
extensive efforts by the groups in Mary-
land and Oregon have produced much
better determinations of the elemental
profiles of a number of sources, these
determinations are time consuming,
expensive, and, in many cases, inappli-
cable to the resolution of aerosol sources
in different locations. In addition, source
determinations of in-stack material
cannot account for chemical and phys-
ical transformations of volatile species
following emission.
Multivariate statistical techniques,
including factor and cluster analyses,
represent a different approach to aero-
sol source resolution. A major advantage
of these methods is that they require no
prior assumption about the nature of the
system under study. In an early applica-
tion of multivariate statistical tech-
niques to elemental source resolution,
factor and cluster analyses were used to
identify sources of airborne particulate
matter in the Boston urban aerosol. In
this study, the data were first trans-
formed into a standardized form that
removed the effects of using different
metrics in describing the data. The
identification of sources was made by
associating the largest calculated factor
loading with the marker element for
each source. In a similar fashion, factor
and cluster analyses were used to
characterize the aerosol of Tucson,
Arizona. Factor analysis was used to
identify sources of paniculate matter in
St. Louis, Missouri. Factor analysis of a
matrix of standardized variables re-
solves only the variance in the system,
not the actual data values. In addition,
the calculated factor scores provide only
a measure of the relative importance of
each source rather than the actual
amount of contributed material.
Recently, a new form of factor anal-
ysis called target transformation factor
analysis (TTFA) has been employed.
This technique permits an analysis
similar in nature to the chemical ele-
ment balance method, but without the
presumptions as to the number or
nature of the sources in the system.
Unlike thenjrior applications of factor
analysis, TTFA allows the investigator to
use his understanding of the system to
determine the best representation of
the actual source concentration profiles.
TTFA also permits the direct calculation
of the contribution of each source to
each sample.
Target transformation factor analysis
has been applied to a subset of the
aerosol composition data acquired during
the Regional Air Pollution Study (RAPS)
for St. Louis, Missouri. In the RAPS
program, automated dichotomous sam-
plers were operated over a two-year
period at 10 sites in the St. Louis
metropolitan area. Ambient aerosol(
samples were collected in fine, 2.4 |/m,'
and coarse, 2.4 to 20 fjm, fractions.
Samples were analyzed at the Lawrence
Berkeley Laboratory for total mass by
beta-gauge measurements and for 27
elements by x-ray fluorescence.
To determine the most appropriate
way to apply target transformation
factor analysis to a large set of aerosol
samples, two subsets of this data were
analyzed—all the samples collected
during July and August at one station
and the samples collected during a
single week at all 10 RAPS stations. The
data from station 112 were selected for
the first subset and the data from the
week beginning July 31, 1976, for the
second. The relative completeness of
the data in these two subsets was the
primary criterion for their selection.
Station 112 was located near Wash-
ington University, west of downtown St.
Louis. During the 62 days of July and
August, filters were changed at 12-hour
intervals, producing a total of 124
-------
samples in each of the fine and coarse
fractions. Data were missing for 48 of
the samples, leaving a total of 200. Of
the 27 elements determined for each
sample, a majority of the determinations
of 10 elements were below the detec-
tion limits. Since a complete and accu-
rate data set is required to perform a
factor analysis, these 10 elements were
eliminated from the analysis. Table 1
lists the full 27 elements and the number
of values below detection limits. Note
that the element arsenic was excluded
because of the large number of deter-
minations below detection limits. Arsenic
determinations by x-ray fluorescence
are often unreliable because of an
interference between the arsenic K x-
ray and the lead L x-ray. A neutron
activation analysis of these samples
should produce better arsenic deter-
minations. Reliable data for arsenic may
be important to the differentiation of
coal fly ash and crustal material, two
materials with very similar source
profiles. The remaining 200 samples
were analyzed in two separate parts.
Two different divisions in the data were
made to determine which division per-
Table 1. RAPS Station 112. July
and August 1976.
Number of Values Below
Element Detection Limits
Al
Si
P*
S
Cl
K
Ca
Ti
V*
Cr*
Mn
Fe
Ni
Cu
Zn
Ga*
As*
Se
Br
Rb*
Sr
Cd*
Sn*
Sb*
Ba
Hg*
Pb
'Elements
15
2
179
0
35
0
0
39
162
151
29
0
95
39
0
194
196
113
1
130
73
161
160
180
131
196
0
excluded from the analyses.
mitted the extraction of the most in-
formation. Separate analyses were then
performed on each group. First, the data
were divided into groups corresponding
approximately to the months of July and
August. In the second analysis, the data
were divided into groups corresponding
to the fine and coarse fractions of both
months.
During the week beginning July 31,
1976, filters were changed at 12-hour
intervals at five of the stations and at 6-
hour intervals at four of the stations. A
total of 182 samples were collected in
each of the fine and coarse fractions.
Data were missing for 43 samples, and
another 50 samples had a majority of
determinations below the detection
limits; these 93 samples were excluded
from the analysis. Of the 27 elements, a
majority of the determinations of 12 had
values below detection limits. These 12
elements were excluded. Table 2 lists
the number of values below detection
limits in the data for one week. Again,
two separate divisions in the data were
made and each individually analyzed.
The data were first divided into two
groups composed of days Saturday,
Table 2. One Week Beginning July
31, 1976.
Number of Values Below
Element Detection Limits
Al
Si
P*
S
Cl
K
Ti
Ca
V*
Cr*
Mn
Fe
Ni*
Cu
Zn
Ga*
As*
Se
Br
Rb*
Sr*
Cd*
Sn*
Sb*
Ba
Hg*
Pb
135
27
265
16
115
0
170
1
318
288
149
1
267
120
18
321
316
201
0
294
224
273
258
309
291
320
3
'Elements excluded from the analyses.
Sunday, and Monday (SSM), and of days
Tuesday, Wednesday, Thursday, and
Friday (TWTF). The first group, with 115
samples, corresponds to days 213, 214,
and 215, and the second group, with
156 samples, to days 216 through 219.
The ana lysis was then repeated with the
data divided into groups corresponding
to the fine and coarse fractions with 139
and 132 samples, respectively.
The 27 elements determined for each
sample account for only a small fraction
of the total sample mass. The remaining
mass consists primarily of oxygen,
nitrogen, and carbon. Determining the
source of particulate carbon is of grow-
ing concern because of the potential
mutagenicity associated with carbo-
naceous particles. Measurements have
shown that both soot and organic carbon
can account for a sizeable fraction of the
total mass of airborne particulate matter.
Though no measurements of carbon are
included in the RAPS data, that portion
of the sample mass must be accounted
for by the other sources. This can often
lead to distortions in the scaling factors
produced by the multiple regression
analysis. In order to produce the best
possible source resolutions, it is vital to
have both accurate TSP measurements
and determinations for as many elements
as possible.
For use in the target transformation
portion of the analysis as test vectors, a
set of 17 potential source profiles was
assembled. These include nine different
crustal factors, two determinations of
coal fly ash, three automobile factors,
and three industrial sources. The 17 test
vectors are listed in Table 3.
Conclusions
Using these data, target transforma-
tion analysis was applied, and several
conclusions were made; they are as
follows:
• Target transformation factor anal-
ysis (TTFA) can be used to resolve
sources of airborne particulate
matter from large sets of ambient
aerosol samples. With this method,
the contribution of each source to
each sample can be calculated. In
addition, TTFA requires no a priori
assumptions as to the number or
nature of the sources in the system.
This method could be an important
tool in the achievement of man-
dated limits on suspended partic-
ulate matter in the urban areas of
this country by identifying those
sources where imposition of control
-------
Table 3. Source Profiles Used as Test Vectors, mg/g
Soils
Element
Al
Si
P
S
Cl
K
Ca
Ti
V
Cr
Mn
Fe
Ni
Cu
Zn
Ga
As
Se
Br
Rb
Sr
Co-
Sr)
Sb
Ba
Hg
Pb
Element
Al
Si
P
S
Cl
K
Ca
Ti
V
Cr
Mn
Fe
Ni
Cu
Zn
Ga
As
Se
Br
Rb
Sr
Cd
Sn
Sb
Ba
H9
Pb
Shale*
80
273
0.7
2.4
0.18
26.6
22.1
4.6
0.13
0.09
0.85
47.2
0.068
0.045
0.095
0.02
0.013
0.0006
0.004
0.14
0.3
0.0003
0.006
0.0015
0.58
0.0004
0.02
Fly Ash
NBS
127
210
0.042
16.1
47
7.4
0.24
0.13
0.5
62
0.16
0.2
0.058
0.01
0.012
0.13
1.7
0.007
2.7
0.075
Sandstone
25
368
0.17
0.24
0.01
10.7
39.1
1.5
0.02
0.035
10
9.8
0.002
0.001
0.016
0.02
0.001
0.00005
0.001
0.06
0.002
0.01
0.007
Carbonate
4.2
24
0.4
1.2
0.15
2.7
302
0.4
0.02
0.011
1.1
3.8
0.02
0.004
0.02
0.004
0.001
0.00008
0.0062
0.003
0.61
0.0002
0.01
0.009
IAEA
82
332
1
19
22
5
0.15
.03
15
45
0.013
0.077
0.4
0.02
0.09
0.001
0.005
0.14
0.33
0.0015
0.56
0.6
0.001
0.13
Soil 1
82
200
15
15
4
0.06
1.1
32
0.04
0.08
0.6
0.2
Soil 2
69
260
25
10
3
0.04
0.6
20
0.02
0.02
1
0.06
Motor Vehicles
Gladney
133
208
13.6
10.6
8.1
0.26
0.19
0.39
123
0.0006
0.23
0.078
0.16
0.025
0.006
0.11
0.81
0.006
1.14
0.056
Auto 1*
0.1
0.054
0.05
0.005
0.01
0.0005
0.38
0.002
0.0008
0.013
1
Auto 2
0.9
6
19
0.6
1
3
37
0.9
148
Auto 3
68
4
1.4
•
79
400
Cement
24
107
5.3
460
1.4
11
Soil3
71
330
0.8
0.85
0.1
13.6
13.7
4.6
0.1
0.2
0.85
38
0.04
0.02
0.05
0.03
0.005
0.00001
0.005
0.1
0.3
0.5
0.01
Industrial
Paint
422
4
204
Rock
82
282
1.1
0.26
0.13
20.9
41.5
5.7
0.14
0.1
0.95
56.3
0.075
0.055
0.07
.015
0.002
0.00005
0.003
0.09
0.4
0.002
0.002
0.0002
0.43
0.0001
0.013
Steel
11
7
1
12
31
570
Soil 4
60
350
0.14
0.06
0.99
37.8
0.08
0.005
0.0006
0.0006
* Normalized to Pb - 1.0.
-------
technology would be the most cost-
effective approach.
• TTFA can resolve strong sources
that are present in only a few of the
samples, and a source-profile vec-
tor for the source can be predicted
from the data. Other methods for
performing quantitative source
resolutions cannot account for
such sources unless their presence
is initially known and a measure-
ment of the source composition is
available.
• The presence of several samples
with one very strong source was
found to conceal the presence of
other, weaker, factors in the full
data set. Removal of the affected
samples permitted the resolution
of additional sources.
• Source profiles of coal-burning
power plant emissions may be
identified by TTFA. Distinguishing
crustal material from fly ash would
be more certain if sources were
available.
• TTFA can resolve sources with
dependence on the same element
such as lead in particles from
motor vehicle exhaust and lead
and/or zinc smelters.
• The analysis of the July and August
data from one RAPS station pro-
duced a reasonable source resolu-
tion. The data from one week at all
RAPS stations seem to be composed
from a number of sources that are
significantly different at each sta-
tion. The large number of sources
contributing to the 10 RAPS stations
precluded the resolution of any but
the most important sources.
• By dividing both data sets intofine-
and coarse-fraction subsets, addi-
tional sources were resolved, in-
dicating that the sources contribute
primarily to either the fine or coarse
fractions, but not both.
Recommendations
With some additional development,
target transformation factor analysis
could be a valuable tool in the efforts of
local authorities to devise cost-effective
strategies for the control of airborne
particulate matter. To ensure the credi-
bility of any sort of statistical analysis, it
is important that there be some indication
Df the reliability of the calculated results.
n the present study, the actual nature of
•he sources was not known, so the
^accuracy of the quantitative results
P:ould not be determined. Testing, under
more controlled conditions, is essential
to ensure that the calculated results are
both accurate and reliable. One way to
evaluate the accuracy of this method
would be to analyze computer-generated
data sets containing randomly generated
errors that approximate typical ana-
lytical uncertainties and source-com-
position variability. Comparison of the
starting data with the calculated results
would give an indication of the reliability
of the analysis. Another way to verify
the quantitative results would be to
analyze samples prepared in the lab
from known amounts of well-charac-
terized materials. Comparison of the
calculated results with the known con-
centration of each sample would deter-
mine the quality of the source resolution.
A third, and more complex, way to verify
the quantitative results would be to
compare the results of the factor anal-
ysis of real samples with results obtained
by x-ray diffraction and/or microscopic
examinations of the same samples.
Further development of target trans-
formation factor analysis would include
error models to give a better indication
of the reliability of the calculated source
profiles and to provide limits on the
uncertainties of the calculated contribu-
tions of each factor. In addition, more
objective and explicit criteria for deter-
mining the number of correct factors to
be retained are needed. The analyses of
computer-generated data sets could
shed additional light on the still nebulous
process of determining the number of
factors.
Performing a source resolution of
ambient aerosol samples requires a
complete and accurate data set. If all the
sources that contribute to a system are
to be resolved, it is important that the
data set include as many elements as
possible. In the present study, the
measured elements accounted for less
than half of the total measured mass of
each sample. The inclusion of other
elements such as carbon and other
gaseous species such as SO2 would
have greatly facilitated the analyses. In
addition, the exclusion of the element
arsenic because of the large number of
determinations below the detection
limits inhibited the resolution of a factor
for coal fly ash. Future air quality studies
should attempt to produce accurate data
for both the total level of suspended
particulate matter and as many elements
and chemical species as practical.
-------
This Project Summary was authored by Stephanie C. Davis of WAPORA, Inc..
Cincinnati. OH 45233.
T. G. Dzubay and C. W. Lewis are the EPA Project Officers (see below).
The complete report, entitled "Modified Factor Analysis of Selected RAPS
Aerosol Data," was authored by Daniel J. Alpert and Philip K. Hopke of the
Ihstitute for Environmental Studies and Nuclear Engineering Program.
University of Illinois. Urbane. IL 61801.
The above report (Order No. PB 81-120 784; Cost: $9.50. subject to change) will
be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield. VA 22161
Telephone: 703-487-4650
The EPA Project Officers can be contacted at:
Environmental Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park. NC 27711
0 US. OOVERNMENTPWNTINOOFFICE 1W1-757-01Z/7086
-------
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Postage and
Fees Paid
Environmental
Protection
Agency
EPA 335
Official Business
Penalty for Private Use $300
PS 0000329
U S tlNVIR PROTECTION AGENCX
REGION 5 LIBKAKSf
230 S DEARBORN STRtET
CHICAGO JL 60604
------- |