*-
vi ra S ra ra 4J
c S u c vi
O O ra f
^ rau- 01 4J
4-( C 1- -0 ro
3 > - O Q 4J
a 4j g vi
-<- -a oi o
t- oi c ro -ax
4J 4-1 14- 01 C
c ra in IN ro
O r- 3 X £
U 3 43 U
U 01 01 C
01 r U -O 4J O
U ra S 01 3
u u ra c ex oi
O 01 ro ro O -O
I/) 43 43 Ol U ra
VI
ra
u
3
in
VI
S~ a.
a
* s
VI CQ
X
I/I -
01 <
ra
4V (/)
(4 in
O CL
VI
-------
a
01
3 IS>
C -1
_ uj
*j a
§ i
u
as
a
i a.
Ol UJ
Ul Ul
_j a:
01
t
5,
4-1
f~
fH>
43
^
U
'o
3
^
2
S
j
It
It
4-1
3
a
'i
a
4>J<
0
4-
£
c
H-*
Ul
OJ
'_
4*
it:
a
Ui
r
o
-^
c
a
u
t
in 01 a
01 U 4-11
c u l/l
C 1 0) U O,f
c 4-1 o; it) i o a it]
- x c in o c
Ul
c
U 01 0
01 u -
U i *J
rO 3 3
1-0 -Q
4.J Ul f-
«- C *" "
o a c
o
x x-i-i u
- 1- Ul 01
4J 01 O U
c a. a. u
a o S 3
3 i- a a
O" O- O l/l
Ul X Ul
01 4J C
O. I- 01 O
X OJ 0.
4-i O. X 4J
O -w rQ
01 u u in
U Q. 01 J-> 01
i- 0 C
3 - s- 01 Q.
O 01 3
01OC>4^-^ lt3U^~C-4-l^"U Cit)
3 einit>B 01 -J coo o r
O" Ul o jC ' O) U Ul <4_ O fQXC O 3
^ 4J U C O 4J V. 4.) O UOJ
c* oi^coi 3a *oa **« u. * 01 1~» o
3XUr OJU QUinCC 01J-> UitlC
uu3uu ui CO) C a.-^- uura
oiusocs oia o.-o oj a ui 3
cojo.£aa (wu^-o>C£uo aoifo
aa.uiu. ««-,«-aa «**
rO
01
C
Ij C
o
01 -
a. ui
.,- 01
i Ol
r
in
01
I
Ul
pK
o
1
("=
X
Ul
c
o
3
A
J3
r
a
u
01
u
u
3
a
i in
U OJ
01 r-
Q. U
O
U 4-1
a u
rQ
itt
U 01
4-t k.
a. 3
a o
^^ Ul
(Q U-
u a
in in
x 01
a. 4^
4J Ul
u OJ in
»- OJ
us 'a
ui u E ,
Q.
a
i r OJ
o -u a. 4-» c
'- ro a 3 oj
a. o < j3 -a
4J
Ul
X
It)
c
(t)
"O
01
c
It)
u
4-1
I/I
OJ
3
o-
01
CE
-26-
-------
o
O)
3 l/»
i- lu
u a
o 2
u
> a:
O
fx H-
i a.
CM LU
<-)
_i ae
ca
r-
.u
.a
re
u
^^
Q
a
c
1
^^
i/i
'o
i
X
i/i in
§ §
3 fQ
J3 U
i. U-
§ln
in
CJ fB
01 U
U
u a>
o «-
V) (/I
i
u
0) S
ra
CO C
§01
01 2
^* ^"* ff
IU-
VI 01
01 U u
i < g u»
«. 21 § £
o m S i/i ra
C I/I I/I 0) C "O
I/I O in > ra 0)
a» ra 01 ex r«
^ ^^ p' i"* S ^ «
ra 3 u o> S 01 u
£ ja e in e«
^ ^ 01 «. ^.
jj u N u) at ra 3
=
O J-t ra O
i/i u in *J i/i u
o> a> in i u 01
aoi -a c 013 u ai
^ O ^" O ^~ B l^ *^" ^3
> S- > r- 01 3
o 3 o .u a^> c o- c
u o ura Q. 3 oj at »
en C o
52°
^ " 0
C U I.
ro 0) U
vi ui Z
VI
at
in
o
i
X
in
|
3
"ll
C
O
U
at
u
3
O
OI
01
U b.
0 °
ex at
e
c
Of-
- in
w ra 01
U J-l
ra in ex
iv^ 1
<** u tfi in
01
o o c
>, -^ 01
U 01 U .g
x 13 ex <
in
01
0- ° **
O inr rl
C rfl in
in o^
01 U 01
4^ ^J 0V ^*
§3 jj e>
° § e
Hu o»
in <->
f U JJ
> i_ in r-
o 3 >< ex
u o u a.
ex, in u <
c
o
2
u
ra
ra i^>
at <<-
xa
§
^,'
ra
u
ra
ex
01
ex
01
&
ra
I/I
O
z
-27-
-------
2.3.1 Chemical Mass Balance Model
The mass balance model (Friedlander, 1973; Watson, et al., 1981; Cooper
id Watson, 1980; Henry, et al., 1983) solves a set of equations to determine
le combination of source contributions that provides a best fit to measured
nbient concentrations of selected particulate matter components (i.e.,
lemical elements, ions, or compounds; size fractions; or other measured
roperties.) The mass balance model, which is also known as the chemical mass
alance (CMB), chemical element balance (CEB) and chemical species balance
Ddel in the literature, requires information concerning the composition of
oth source emissions and ambient concentrations. Refined results can be
btained if the component concentrations are compiled on a particle-size
gecific basis.
In the mass balance equations, the total ambient concentration (C) of a
omponent (i) from all sources (j) is set equal to the linear sum of the
roducts of 1) the component's fractional mass contribution (F) to the
ource's total emissions, and 2) the source's fractional mass contribution (S)
o the total ambient concentration measured at the receptor:
C V& *C 1 J " J
hus, if the total ambient particulate matter concentration (within the
pecified size range) and the (corresponding) ambient and source component
oncentrations "are measured, source contributions to the receptor can be
alculated.
The simultaneous solution of the mass balance equations is an integral
art of the mass balance model. There are a number of available solutions to
he mass balance equations. The tracer solution (Miller, et al., 1972, Kneip,
t al., 1972) assumes certain tracer properties are unique to each source
-28-
-------
type. It is the simplest solution and unlike other solution methods requires
no computer. The intent of the linear programming solution (Mayrsohn and
Crabtree, 1976; Henry, 1977, Hougland, 1983) is to find optimum values for the
source contributions. The ordinary weighted least squares solution
(Friedlander, 1973), which weights variables by the inverse squares of the
precisions of the ambient concentrations, has been supplemented by the more
general effective variance least squares solution (Britt and Luecke, 1973;
Dunker, 1979; Watson, 1979; Watson, et al., 1983). The effective variance
solution includes weighting by the precisions of the source compositions as
well as weighting by the precisions of the ambient concentrations. The intent
is to weight the most precisely known measurements most heavily in arriving at
a least squares solution yielding source contributions to observed ambient
concentrations of particulate matter. The effective variance least squares
solution is more valid than the ordinary least squares solution only if the
precision of the source data are known. Such data are often missing or of
poor quality. A ridge regression solution (U.S. EPA, 1983a) is intended to
minimize instabilities in a least square solution when two or more source
categories being considered have similar chemical compositions. In ridge
regression, the goal is to introduce a small bias into the solution in order
to achieve a large reduction in the random error, so that the total error is
reduced. As Henry has shown (1982), the introduction of such a bias can
result in an underlying source matrix that is not representative of the true
source profiles. Ridge regression should not be used indiscriminately, as if
its application would automatically eliminate all problems caused by
multicollinearity, since it cannot be guaranteed that its use will produce a
better solution than conventional least squares in every application.
The assumptions of the mass balance model are (Watson, 1982):
-29-
-------
Compositions of source emissions are constant over the period of
ambient and source sampling.
Components (e.g. chemical species) do not react with each other,
i.e., they add linearly.
All sources with a potential for significantly contributing to
the receptor have been identified and have had their emissions
characterized.
The number of sources or source categories is less than or equal
to the number of components.
The source compositions are linearly independent of each other.
Measurement errors are random, uncorrelated, and normally
distributed.
Model verification and evaluation studies have been conducted and
published by Watson (1979), Gordon, et al., (1981); Currie, et al., (1983);
and Dzubay, et al., (1983). These studies consist of 1) comparison of
different mass balance solutions operating on the same measured data, 2)
comparison of different solutions applied to simulated data with the known
source contributions from which the data are derived, and 3) analytical
examination of the solutions of the mass balance equations. The results of
those evaluations as they relate to the above assumptions are discussed in
Appendix A.
The selection of source types, source compositions and aerosol
measurements is an integral part of the mass balance model application. This
selection is often dictated by the measurements which are or can be made
available". These measurements are the subject of Section 3.
Volume III of this technical series (U.S. EPA, 1983a) describes a chemical
mass balance model algorithm developed for EPA. Other, similar mass balance
model algorithms have also been developed. These models are usually applied
by allowing the user to add various source types and types of ambient data
into one of the solution methods until most of the measured properties at a
-30-
-------
receptor are accounted for by the source contributions. Unfortunately, in
some cases, several mathematical "solutions" can be found which "account" for
these receptor properties equally well, with no detrimental effects on the
least squares fitting criteria. In intercomparisons of a number of mass
balances solutions applied to a 9-day average, size-segregated particle data
base from Houston, Texas, Dzubay, et al., (1983) found several different
solutions which were considered acceptable by the participants. In similar
intercomparisons using simulated data, a number of different, acceptable
solutions were found to the mass balance equations also (Currie, et al.,
1983), as was the case when multivariate techniques were used. Watson (1979)
observed that "The receptor model tells what could be the contributors, not
necessarily what the contributors are." The results from the Houston data
base analysis reaffirm that it is advisable to have a qualitative knowledge of
the contributing sources before seeking to obtain quantitative source
contribution estimates using the mass balance model.
The major contributors to average concentrations in the simulated data set
were estimated, for the most part, to be within +30 percent of their true
values. Minor contributors were often within a factor of two (Currie, et al.,
1983). Although the true contributions in Houston were unknown, the range of
values for most source types was less than a factor of two (Dzubay, et al.,
1983). A further note of caution on the Houston analyses, as well as for
analyses conducted previously in other cities (U.S. EPA, 1981b), is that
sulfate, nitrate, carbon and crustal material are defined by chemical
characteristics and not their sources. Therefore, continued research is
required to obtain parameters that will account for the generation,
transformation and removal of particles during transport from source to
receptor.
-31-
-------
The uncertainty of the mass balance results is probably comparable on
average to that expected of source models intended to accomplish the same
purpose. Some aspects of this problem are discussed for the simulated data
set (Currie, et al., 1983) by Gerlach, et al., (1982), and more generally by
Henry (1982).
2.3.2 Factor Analysis
Factor analysis solutions can be obtained by different methods. Classical
factor analysis or principal component analysis can be applied to ambient
sample composition measurements at a receptor to produce a set of uncorrelated
factors which are associated with various sources. Under certain
circumstances, these factors can be interpreted as being representative of
specific sources and, using additional steps, can provide estimates of source
emission compositions. Factor analysis for receptor modeling requires
elemental and chemical or other constituent concentration measurements at
receptors, but does not require quantitative characterization of source
emissions. However, some qualitative information on sources and the
characteristics of the local meteorology are necessary to interpret the
factors (Cooper and Watson, 1980). The derived source emission compositions
can then be used in a mass balance to determine the relative contribution of
different sources to each observed concentration.
Factors are derived from the correlations of observed particulate matter
or other ambient concentration measurements that represent different time
periods of sampling at a receptor location. Factor analysis assumes that the
composition of particulate matter from a given source remains constant over
all of the ambient samples and that inter-species correlations are due to
changes in source strength. Advantages of factor analysis include the ability
to distinguish among sources which contain the same constituents (in different
-32-
-------
proportions), the ability to identify previously unsuspected sources and no
requirement of quantitative source composition measurements. Disadvantages
include added complexity, the need to impose arbitrary constraints on source
emission composition in order to obtain a unique solution, and the need for
detailed ambient composition data for a relatively large set of ambient
measurements. However, the latter can provide information on sources over the
course of a season (Kneip, et al., 1983; Morandi, et al., 1983) or year
(Kleinman, et al., 1980; Thurston, 1983) if sufficient samples are taken to
satisfy statistical constraints. For mass balance analyses, a limited (one or
more) number of samples are required in each season, but more source testing
may be necessary.
The factor analysis model (Henry, 1977; Kleinman, et al., 1980; Hopke,
1981; 1983; Henry, et al., 1983) consists of 1) forming a correlation matrix
by summing over aerosol samples (known as "Q mode") or over aerosol properties
(known as "R mode"), 2) finding the eigenvalues and eigenvectors of the
correlation matrix, 3) discarding a number of eigenvectors and eigenvalues
which are deemed insignificant; the number of values remaining are examined
and interpreted for identification of source types, 4) forming linear
combinations of eigenvectors which represent source compositions or choosing
representative tracers, and 5) using these source compositions in the mass
balance equations. Each of these steps includes a variety of options, the
selection of which differentiates one factor analysis model from another.
Correlation or covariance matrices around the mean or the origin can be
used (Rozett and Peterson, 1975; Duewer, et al., 1976) in R-mode, or in Q-mode
(Alpert and Hopke, 1980; 1981). These matrices can contain as many aerosol
properties and as many samples as desired (available), though Henry, et al.,
(1983) estimate that the minimum number of samples needed is:
-33-
-------
n + 3
m > 30 + 2
where m = number of samples
n = number of aerosol properties
Henry, et al., (1983) conclude that if 20 aerosol properties are used, an
adequate correlation matrix can be formed only if at least 42 samples are
available.
The determination o£ eigenvalues and eigenvectors is performed by a
variety of computer subroutines which are common to most computer libraries,
and though individual algorithms vary, they yield essentially the same results.
The selection of the number of eigenvalues to retain is open to question.
Alpert (1980) proposed five criteria: 1) eigenvalue equal to 1, 2) chi square,
3) Exner function, 4) root mean square (RMS) error, and 5) indicator
function. Alpert (1980) found most of these tests indicated the same number
of significant aerosol sources in practical applications, but that for certain
cases they were inconsistent. Hopke (1982) warns that "Many statistical
packages, including BMDP and SPSS, set the eigenvalue of 1 criterion as
default and do not examine additional factors unless the default is
specifically counter-manded. This procedure can lead to exclusion of
significant factors." Thus an appropriate test for selecting the number of
contributing sources from the available eigenvectors needs to be developed.
For principal component analysis, Roscoe, et al., (1982) have suggested that
for eigenvalues of less than 1 to be included, the variance of a rotated
factor should be greater than 1.
Selected eigenvectors can be more easily related to source compositions
after a vector rotation has been done by a 1) varimax, 2) quadrimax, or 3)
target transformation method. The target transformation method has been shown
-34-
-------
to be useful for aerosol receptor models by Hopke, et al., (1983). However it
is constrained by the selection of the profile used as a target. Henry (1977)
chose targets which were actual source compositions similar to those known to
be present in the study area by minimizing the sums of the squares between
target and rotated vector coordinates. The new vector coordinates were taken
to be a more refined estimate of the true source composition.
Recently Thurston (1983) used principal component analysis for source
identification in Boston, Massachusetts. In this instance, the factor scores
were adjusted to absolute zero and the resultant adjusted factor scores were
used to conduct a mass balance analysis.
2.3.3 Multiple Linear Regression
Regression techniques provide a straightforward method for characterizing
the composition of a set of concentration measurements. Multiple linear
regression provides a least-squares solution which apportions a total or
size-segregated particulate matter concentration among a set of chemical
elements and/or other constituents. The general form of the solution is:
? = Ao 4- A! Xi + A2 X2 + . . . An Xn,
where Y is the observed concentration, the X4's represent the various
constituents, and the coefficients At are calculated to provide a least
squares fit to the data.
Unlike the mass balance method, multiple linear regression, taken alone,
does not require data regarding the composition of the emissions from sources,
and does not directly identify source contributions to ambient concentrations,
unless proper selection techniques (e.g., factor analyses) are used to select
the constituents that are analyzed. The analyst must separately determine
-35-
-------
which sources are responsible for emitting a specific element or chemical
species and ensure that tracers are available for use in the analyses.
The assumptions of the multiple linear regression model taken alone
include the following:
All of the important constituents of ambient particulate matter
have been identified and measured.
The contributions of different constituents are linearly
independent of one another.
For any given set of constituent values (Xt's), the observed
concentration values are normally distributed.
The multiple linear regression results will rank the independent variables
(constituents) in order of importance, for explaining variations in the total
concentration, and will also identify those constituents which play no
significant role in explaining variations in the total concentration.
The above is a very rudimentary approach to the analysis of source
apportionment data. However, if multiple regression techniques are coupled
with procedures that independently identify the Xi's, the regression
analyses can be applied to mass balance calculations. Kleinman, et al.,
(1980) published work for New York City that used cluster and factor analysis
to identify source profiles and subsequently selected fairly unique tracers
for use in a stepwise regression model to apportion the ambient particulate
matter mass. Subsequently, Kneip, et al., (1983) and Morandi, et al., (1983)
used the Factor Analysis/stepwise Multiple Regression Receptor Model (FA/MRRM)
to apportion size-selected samples for New York City and the Airborne Toxic
Elements and Organic Substances (ATEOS) site in Newark, New Jersey,
respectively. Hopke, et al., (1983) applied scaling factors to his source
profiles obtained from target transformation and used a regression model to
attribute the ambient mass measurements. As mentioned in the previous
-36-
-------
section, Thurston (1983) has developed adjusted (absolute) principal
components for source types that were subsequently used in a regression model
to apportion the ambient mass.
2.3.4 Optical Microscopy
Examination of particles collected on a filter medium using an optical
microscope provides insights into the origins of particulate matter (Draftz,
et al., 1980; Crutcher, 1982; U.S. EPA, 1981a; 1983b). In the hands of an
experienced microscopist, optical microscopy (OM), also known as light
microscopy (LM) and polarized light microscopy (PLM), can often yield a
semi-quantitative determination of the relative contributions of source types
if the categories of sources are few in number and produce completely
different types of particles, e.g., distinguishing between particles of
mineral origin such as road dust and particles produced by combustion
processes (Throgmorton and Axtell, 1978).
Particles may be examined in situ on the filter medium with or without
immersion oil or may be removed with adhesives, probes, or ultrasonically for
examination under more controlled conditions. However, any method in which
the particles are removed from the original filter media may compromise the
representativeness of the analysis and produce biased results.
Particle sizes below about 1 or 2 urn cannot be readily identified by
optical microscopy since these particle sizes begin to approach the
wave-length of visible light. Since ambient particle size distributions are
typically bi-modal, consisting of two families of particles, with the mode of
one greater than about 10 urn and the mode of the other less than about 2
pm, most fine particles cannot be identified using optical microscopy.
Particles can be examined by both reflected and transmitted light.
Transmitted light may be polarized by any of a number of orientations allowing
-37-
-------
the skilled microscopist to identify a variety of mineral species. Although
color photographs of known particle types are helpful to the microscopist in
making particle and, hence, contributing source identifications, (McCrone,
1973), many analysts prefer to have source emission samples collected within
the study area.
The various material characteristics that are examined include structure,
transparency, color, and optical properties. The identification system is
used with a series of tables such as those found in Kerr (1959) which
associate the observed properties or parameters with a list of materials
having those properties. Some of the properties observed are: color, habit
(shape and form), cleavage, index of refraction, isotropic versus anisotropic,
birefringence, arid uniaxial versus biaxial.
Particles that have been identified are grouped within categories such as
combustion products, minerals, biological material (e.g., pollen), rubber, and
eventually a miscellaneous or unknown category.
In order to assign a quantitative value to the relative abundance of a
given species it is necessary to carry out a particle size distribution for
each of the identified categories. Then, if a particle density can be
assigned to each category, the relative percentage by weight for each category
can be estimated.
Several attempts have been made to evaluate the reproducibility of
microscopic analysis or particle collections by such techniques as
inter-laboratory analysis of blind replicate samples. The results show wide
disagreement in results among analysts. Factors contributing to differences
in the results of analyses by skilled microscopists include: 1) lack of
common agreement concerning particle removal or in situ examination; 2)
disagreement over source classification groupings, 3) different particle
sizing and counting methods, and 4) statistical counting errors (e.g. if 103
-38-
-------
particles of Sum diameter and 2.5 g/cm3 density are examined, only 0.3
Vig out of a typical deposit of 10s ug, or less than 10"3 percent of
the sample, will be examined).
Further details on the use of optical microscopy in source apportionment
studies can be found in Volume IV of this series (U.S. EPA, 1983b).
2.3.5 Scanning Electron Microscopy
Another method used to provide particle-specific information is the
scanning electron microscope (SEM). An automated scanning electron microscope
(ASEM), which produces digitized particle images, can be used to overcome many
of the problems associated with individual particle analyses. A scanning
electron microscope combined with automated image analysis and X-ray
spectroscopy (SAX), also known as computer controlled scanning electron
microscopy (CCSEM), can be used to provide simultaneous measurement of
individual particle size, sha'pe and elemental composition.
CCSEM combines three analytical tools under computer control: 1) the
scanning electron microscope, 2) an energy dispersive spectrometry (EDS) x-ray
analyzer and 3) a digital scan generator for image processing (Casuccio, at
al., 1983).
In the CCSEM, a finely focused electron beam impinges upon the sample
surface. The interaction of the electron beam with the sample produces
secondary and/or backscattered electrons that are used to create a viewing
image, while the x-ray emission is monitored to determine the elemental
chemistry of the particles.
The automated image analysis of CCSEM is normally conducted in the
backscattered mode. The backscattered signal is sensitive to differences in
atomic number and is thus suited to the analysis of particles. Solid-state
-39-
-------
backscattered electron detectors are positioned above the sample; their
individual signals are amplified and then summed to produce the final
backscattered electron signal.
The computer compares the image intensity at each point with a threshold
level. This comparison is used to determine whether the electron beam is "on"
a particle (above threshold) or "off" a particle (below threshold). If the '
signal is below the threshold level, the computer selects a new (x,y)
coordinate and directs the beam in a preset pattern of rotated diagonals to
determine the dimensions and shape of the detected feature. The pattern is
repeated twice, once to locate the feature centroid and once to determine the
lengths of the diagonals. For each feature, the maximum, minimum, and average
diagonals are stored, along with the centroid location. The centroid is used
to prevent double counting.
To classify and distinguish different particle types, it is essential to
determine the elements contained in each particle, along with the relative
intensities for each particle. Once the electron beam is positioned on a
particle, the x-ray spectrum is collected. All elements heavier than sodium
are simultaneously detected. The computer's classification routine identifies
the most significant peak(s) and assigns each particle to the group of
particle types having the same major elemental constituent(s). The relative
intensities of the major and minor elements in the spectrum are then used to
assign each particle to a specific particle type within the group. Particle
types (e.g., spherical iron and cenospheres) may also be classified by the
aspect ratio or shape factor. The absence of elemental peaks or a low
peak-to-background signal causes the particle to be classified as carbon.
Particles found on ambient filters are compared to those on source emission
filters (preferably collected within the study area) to identify contributing
sources and determine source contributions.
-40-
-------
One advantage of this technique is the ability to rapidly analyze large
numbers of particles, thus overcoming the statistical limitations inherent in
optical microscopic methods. Because each particle analysis requires two to
five seconds, 1000-2500 particles may be examined in a two-hour period.
(Although this is a larger number of particles than is usually analyzed
optically, it is on the order of 0.001 percent of the deposit on a filter
sample and may not be statistically representative of the entire sample.)
This method can analyze particles from any filter medium if particles are
removed prior to insertion into the microscope. Care must be taken to ensure
that the removed particles are representative of the original sample (Kelly,
et al., 1980). Nuclepore or Teflon filter media are usually required for the
in situ examination of particles. However, in situ analysis requires lightly
loaded filters, which may not be practical in some instances.
Further discussion of this technique can be found in Volumes I and IV of
this series (U.S. EPA, 1981a; 1983b) and in Johnson, et al., (1983).
2.3.6 X-Ray Diffraction
The x-ray diffraction (XRD) technique is also a method used to provide
particle information. The method depends upon the wave character of x-rays
and the regular spacing of planes in a crystal. X-rays impinging on the
crystal are diffracted in a manner that is unique to that crystalline
structure. The diffraction "fingerprint" obtained can be compared to a
library of data (e.g., American Society of Testing Materials, 1955) and the
structure identified. Various automated cameras and devices are now in use to
accelerate this process.
This technique is specifically utilized to identify the crystalline phases
present in a sample: e.g., quartz, calcite, dolomite, halite, lead ammonium
-41-
-------
sulfate. It can provide a direct identification of such sources as pavement
minerals, soil minerals, ores, pigment, cement, metal oxides, asbestos,
abrasives, sulfates and nitrates, and indirect identification of road and sea
salt. Sometimes it is the only method available for determining which of the
possible polymorphic forms of a substance are present: for example, carbon in
graphite or in diamond. Differentiation among various oxides such as FeO
Fez03, and FesO*, or between materials present in such mixtures as KBr
+ NaCl, KC1 + NaBr, or all four, is easily accomplished with x-ray
diffraction, whereas chemical analysis would show only the ions present and
not the actual state of combination. XRD can also identify the presence of
various hydrates.
XRD is basically a qualitative analysis method, although it is also
adaptable to quantitative applications since the intensities of the
diffraction peaks of a. given compound in a mixture are proportional to the
fraction of the material in the mixture. However, direct comparison of the
intensity of a diffraction peak in the pattern obtained from a mixture is
difficult. Corrections are frequently necessary for the differences in
absorption coefficients between the compound being determined and the matrix.
The method is applicable to most filter media and no sample preparation is
necessary. A loading of at least 200 ]jg/m3 on the filter is optimum, but
not necessary.
"Further information on this technique may be found in Willard, et al.,
(1974); Kuwana, (1980); U.S. EPA, (1981a, 1983b); and Johnson, et al., (1983).
2.3.7 Preliminary or Qualitative Receptor Models
The foregoing subsections describe some sophisticated techniques that can
provide quantitative estimates of source attribution. Application of those
-42-
-------
techniques requires that the user have either specialized equipment and
training or a significant understanding of mathematical and statistical
manipulation of data. The methods described in this subsection require
spatial and/or temporal correlation of aerometric data plus a moderate
understanding of statistics and may be used to supplement the results of other
receptor modeling or may be used to draw inferences from available data on
categories of sources contributing to measured particulate matter levels.
These methods use historical data for particulate matter, collected using both
standard high volume (hi-vol) and size selective air samplers, together with
local meteorological data to establish background levels, trends,
interrelationships between station values, geographical and wind
direction-stratified patterns, and relative influence of non-traditional
sources. It can be assumed that many of these analyses will have been
completed prior to conducting more complex receptor modeling simply because
they represent a means of developing common understanding of the receptor data
and their interrelationships.
These qualitative receptor modeling techniques include the following:
Background Concentration Determinations
Historical Trends
Frequency Distributions
Tests for Lognormality
Monthly Variations
Weekday/Weekend Analysis
Wet Day/Dry Day Analysis
Episode Day Analysis
Spatial Mapping
Correlation Coefficients
Time Series Analysis
Wind Trajectories and Pollution Roses
Each of these are described in the following paragraphs. Further details on
these methods and their interrelationships as source apportionment methods can
-43-
-------
be found in Brookman and Yocom (1980), Throgmorton and Axetell (1978), and
U.S. EPA (1983d).
2.3.7.1 Background Component Determinations
For purposes of developing a control strategy for particulate matter,
background concentrations must be taken into consideration. The method that
federal regulations (Section 51.13c) specify for estimating background is to
use a concentration measured at a non-urban site that is in or near the study
area and is unaffected by nearby emission sources. This concentration is
assumed to be composed of material that is transported into the study area
from external sources over which the area (i.e., the local control agency) has
no control, plus material generated within the study area from natural and
agricultural sources. Unfortunately, an ideal, unaffected, non-urban site in
the area of study does not always exist. As an alternative, the background
particulate matter level can be established by developing a composite
particulate matter rose where data gathered at several monitoring sites are
averaged based on the frequency of winds coming from selected directional
sectors, as described below.
To remove the influence of the study area sources on particulate matter
levels within the study area, only data collected at selected monitoring sites
located at the far edges of the study area, during periods when the sites are
upwind of the study area sources, are chosen for analysis. The particulate
matter concentrations at these sites are then screened for all sampling days
with a wind persistence factor (Heidorn, 1978) above a selected value (e.g..
-44-
-------
>0.71*). These daily concentrations are then classified by wind direction
sector and the sector average of each data set is calculated for each site. A
composite particulate matter rose is constructed utilizing the wind direction
sector-average concentration values from the selected sites. To determine a
weighted average background level, the percent of time the wind blows from
each directional sector is used. The average particulate matter level for
each directional sector is multiplied by the frequency of wind from that
sector and the results are summed in order to give the weighted average.
The weighted values obtained are usually composed of particulate matter
from natural and agricultural sources, transported secondary particulate
matter and locally-generated particulate matter from such sources as
residential heating, vehicle exhaust, reentrained road dust, tire rubber, and
vegetative burning. In most instances, the influence of the latter sources
will be relatively minor. This background particulate matter composition can
be further defined by analysis of the material collected on samples selected
to represent background conditions.
Particle size-fractionation analyses can be used to help differentiate
between transported particles (usually in the fine fractions) and locally-
generated particles (usually in the coarse fractions). Chemical composition
analyses can be used to differentiate between particle types and thus help
define their origins. In this manner, a' background particulate matter level
for the study area can be established. In subsequent analyses, this
background level can be applied uniformly to data at all sampling sites within
the study area.
*Wind persistence factor is defined as the ratio of the vector averaged wind
to the average wind speed over the 24 hour sampling period. A factor near 1.0
indicates a wind that blows consistently from one direction during the entire
sampling period. A persistence factor >0.71 is equivalent to an hourly wind
direction deviation of 45°.
-45-
-------
2.3.7.2 Historical Trends
Using the historical particulate matter data, the yearly arithmetic and
geometric means are calculated for each monitoring site. These means are then
plotted as a continuous function of year for each station in order to identify
the occurence of sudden changes at that particular station. It is usually
helpful to subtract the study area background level from the annual mean at
each station before plotting or else to plot the background level for each
year alongside the monitoring site means. This will serve to either factor
out or allow for an evaluation of the effect of year-to-year variations in
meteorological conditions (e.g., precipitation) on the sampling days. This
analysis technique can provide the following types of information.
A large change in the yearly particulate matter levels at only one
site can indicate a local source, such as a construction project,
starting up or shutting down. It could also be indicative of a
siting change.
A large change in the yearly particulate matter levels at several
"sites in a large geographical region can indicate a major
particulate matter source starting up or shutting down or
undergoing a major change in operations.
Gradual changes in the yearly particulate matter levels at a
particular site or several sites can indicate the effectiveness of
implemented control measures or the gradual deterioration of a
control program.
2.3.7.3 Frequency Distributions
"Analysis of the frequency distribution of the ambient data will often
provide insight into the mix of contributing sources. A common methodology
involves grouping the data into concentration bins which are then displayed
graphically as histograms. If the distribution is bi-modal or shows extreme
values (outliers) of much larger concentration than the mass of data, then a
directionally dependent effect on that monitor, or intermittent, infrequent
source or other similar causes for the anomolous values may be suspected.
-46-
-------
2.3.7.4 Tests for Loqnormality
The presence of predominant local sources can often be confirmed by
testing the ambient concentration data for lognormality. This can be done by
applying statistical tests (e.g., Kolmogorov-Smirnov) to the data or by
plotting the data (Larsen, 1971). In the latter case, the percentage of
measurements greater than a selected particulate- matter concentration is
plotted as a function of the logarithm of that particulate matter
concentration. If the resulting curve is nearly a straight line, the
observations are lognormally distributed and the site is likely subject to
large-scale or general influences (i.e., many sources). If there is a major
local influence, such as a nearby stack or a strong area source in a specific
direction with respect to the sampling site, then the data will most likely
either not exhibit lognormality or will deviate from it at the plot extremes
(Larsen, 1971).
2.3.7.5 Monthly Variations
Several years of particulate matter concentration data are averaged by
month and plotted to show the monthly and seasonal variations that exist at
each of the sites. Corresponding plots of background concentrations can be
used to show the effects of changing meteorological conditions. An abnormally
high winter average can be an indication of heavy traffic influence due to the
combination of longer morning inversion periods, cars idling while cold, road
t
sanding/salting operations and increased space heating, especially by wood
stoves. A high summer level can be an indication of dry surface soil in
conjunction with agricultural activity or the increased use of school
playgrounds. High summer and fall concentrations may also be caused by
secondary aerosols, construction activities, pollen, leaf burning, etc. This
can be verified by chemical analysis.
-47-
-------
A variation of this analysis can be performed where the monthly average
plots are grouped for stations that are similar in character or in close
proximity. This helps to determine whether these groups of stations are
influenced by regional-scale, urban-scale or neighborhood-scale sources. Two
stations in an urban area might exhibit similar seasonal particulate matter
patterns, indicating that an urban- or regional-scale source is the major
contributor. On the other hand, one station might have a higher overall
particulate matter concentration, indicating a local- or neighborhood-scale
source influence.
2.3.7.6 Weekday/Weekend Analysis
During weekends, driving is typically done in the middle of the day when
dispersion conditions are at their best. During weekdays, peak driving is
typically done in the early morning (0630-0900 hours) and late afternoon
(1500-1800 hours) when dispersion conditions may be poor, thus keeping
traffic-suspended particulate matter in the vicinity of the point of
generation in a relatively undiluted condition. Weekday and weekend traffic
volumes also differ considerably. By computing the arithmetic averages for
weekday and weekend periods as well as Saturday and Sunday individually for
each site, a measure of the influence of traffic on ambient particulate matter
levels can be obtained. Most industries operate on a 7 day week, but if a
source does operate on a 5 day week or has a lengthy shutdown, then its
contribution may also be estimated from this type of time series averaging
analysis. Other sources with a weekend/weekday dependence could confound this
analysis. The analysis may be enhanced if weekdays/weekends with similar
meteorological conditions are compared, but additional work is required to
stratify the particulate matter data by meteorology.
-48-
-------
2.3.7.7 Wet Day/Dry Day Analysis
It may be assumed that on days with sufficient snow cover, with greater
than 0.5 centimeters of precipitation, or following days with greater than 0.5
centimeters of precipitation, the principal contributions to particulate
matter levels are traditional sources, home heating, vehicle exhaust, and
material transported from outside the study area. The rain and snow suppress
the majority of the local fugitive and reentrained dust. These days are
defined as "wet" days and all other days as "dry" days. The meteorological
data for the study period are examined and the particulate matter levels at
each site on wet days are then averaged and compared to the average
particulate matter levels on the dry days. The differences give a first
approximation of the reentrained dust and local fugitive emissions
contributions at each of the sampling stations. In an extension of this
analysis, the particulate matter levels on days with snow cover can be
averaged versus days without snow cover. Such an analysis is often helpful in
approximating the effect of total control of windblown dust from sources such
as inactive storage piles. When conducting these analyses, care must be taken
to consider the effects of winter sanding and salting in those areas where
this is practiced. Results may be misleading in such instances.
2.3.7.8 Episode Day Analysis
Days are selected on which higher than typical particulate matter
concentrations are recorded throughout an area. The meteorological data are
examined to determine the possible influence of inversions, stagnation
conditions, days since last significant precipitation event, etc. Large-scale
sources, most likely outside the study area (i.e., long-range transport), are
possible reasons for such regional-scale emission events. Examples might be
dust storms, forest fires, volcanos, etc.
-49-
-------
2.3.7.9 Spatial Mapping
In this method, the particulate matter data obtained on all sampling days,
or only those days when the wind persistence factor (the ratio of the vector
averaged wind to the average wind speed) is above a selected value (e.g.,
>0.71), are sorted by wind direction category (compass sector). The
particulate matter concentrations for each sector are averaged for each site
based on the number of observations and the averages are then used to form
maps representing particulate matter concentrations associated with winds from
each of the compass directions. If sampling is conducted at a number of sites
in the study area, these maps may help identify the location of contributing
sources, based on the sites and wind directions associated with the highest
average concentrations. Such maps may be particularly useful for PMio
studies since the spatial variability of PMio is less than that of TSP.
However, in most urban situations the maps may be misleading because
predominant, very localized influences can overwhelm the contributions from
sources with more widespread impacts.
2.3.7.10 Correlation Coefficients
A correlation coefficient between two variables indicates a co-variation
of the individual measurements of those variables. Many times a common cause
of this co-variation, owing to meteorological or emissions variability can be
inferred. If concentrations between sites display high correlation
coefficients (i.e., they track nearly identically), then the monitoring
stations could be influenced by the same types of sources or changes in
pollution dispersion. This would be the case where there are regional-scale
influences.
-50-
-------
Conversely, low correlation coefficients between sites could be caused by
local influences. An example would be the case where one monitor is located
adjacent to a steel mill while another monitor is near a highway interchange.
2.3.7.11 Time Series Analysis
The time series analysis is another valuable technique for obtaining
insight into source/receptor relationships. In one such analysis, the 24-hour
concentration measurements for all the sites in the study area can be plotted
versus time on the same graph. The concentrations will usually vary in a
similar manner. Sites and periods which do not vary similarly are indicative
of events that occur on a neighborhood scale. Time series plots can also be
prepared to compare changes in particulate matter concentrations to changes in
meteorology (e.g. wind direction, wind speed) and other parameters.
2.3.7.12 Wind Trajectories and Pollution Roses
Pollution roses, which depict the average particulate matter concentration
for various wind directions, are constructed for each site for each of the
study years. They can be based on concentration averages for eight or sixteen
compass points for all sampling days or only those sampling days possessing a
pre-selected wind persistence factor.
These roses are useful in associating average pollutant concentrations
with" a particular wind direction or directions. They are capable of showing
the following types of influence from sources:
Lack of any specific directional effect of sources on background
stations.
Diffuse influence of distant industrial complexes.
-51-
-------
Combination of diffuse influence from distant sources and nearby
sources.
Influence of nearby large sources.
As an aid in performing the analysis, the percent of time the wind blows
from a particular direction (e.g., persistence >0.71) is determined from the
available wind data. This wind frequency can then be combined with the
pollution rose information to determine the particulate matter level
contribution from a particular compass sector. The value of such analyses
depends on the representativeness of the available meteorological data, as
discussed in Section 3.3. A good example of the use of pollution roses and
wind trajectories is the work performed by Gordon (1980).
-52-
-------
3.0 AM EVALUATION OF EXISTING MODEL INPUT DATA
3.1 Overview
The application of either a source or receptor model requires the
acquisition of input data. Generally these input data include source data
relating to emission strengths and characteristics, meteorological data, and
ambient measurements for the pollutant(s) of concern. Quite often, acquiring
input data is the most costly part of any model application. However, there
is a large amount of existing data available for use in model applications,
and often the objectives of a study can be met without the need for further
measurement programs. Frequently, a source or receptor modeling application
can be designed around an existing data base. It is useful to examine
existing data to determine if further measurements are needed to implement a
given application. Review of existing data to determine their quality and
validity is necessary before use of the data in any modeling application.
The general types of data used by source and receptor models are shown in
Table 3-1. Further discussion of these data are provided in the following
subsections.
3.2 Source Data
Source data have generally been compiled in emission inventories, which
contain a listing of all recorded sources for a given area. The data
contained in emission inventories can vary widely depending upon who has
compiled the inventory and the purpose for which it was compiled. Thus it is
advisable to review the initial purpose of an inventory before using it in a
given application.
-53-
-------
C/3
u
w
Q
X X
XXX
(I)
> 01
H CD
4J 01
H (0
X
X X
XXX
X
I
co
a
03
ss
Oil
CD
T3
£
01
T3 CU
CD Oil
C >1
* rtl
H «4-l (TJ
C] CD C
Ol OS <
-I
01
U
CD
CU
01
H
Q
CD
O
U
3
O
co
?.
-* CD
C 01
CD >
CD r-l
u fl
0 C
CO <
XX X X X X
X X X X X
XX X
XXX
o
fll
4-1
a
U-l
o
co
0)1
§1
oj
01!
Oil
HI
01
01 4->
CD £
4J C7>
m -H
U 0)
O CD
H 01
CQ in
01 CD
»M II
S cu
W Oi
01
01 O
H CU 4-1
CO
§
01
O
01
O -u O
* 3 Oi
CO O
s o
i-(
U
(Q CD
Oi U
U
O
CO
S
o
o
d>
u
£
>1
]y
Q
~4
O
O
CD
4-1
/ll
01
0
4-1
0
CD
^4
4
o
T3
C
3
01
T3
0)
CD
Oi
t3
c
^
01
CO
H
4J
«^
« t
H
j3
(Q
4J 01
CO 4J
JJJ
O O>
H -^
U CD
CD &
A
Qj n\
01 C
O *
£ X
4J -rt
< £
01
CD
^4
3
4J
IT)
CD
0,
a
CD
H
O
>fH
4J
(T)
4-1
- 1
01
c
0)
4J
4J
(0
CU
«
01
u
4.
O44-I
rt
O
CD
^4
o.
04
o
c
>l
CO
01
c
o
4-1
01
c.
o
4J
n)
u
4J
c
CU
o
c
o
CJ
4J
c
CU
'1
-------
3.2.1 Emissions Inventories
A basic source of emissions inventory information is generally the State
(or local) air pollution agencies. These agencies are usually responsible for
gathering information on emission sources within their jurisdiction and
compiling the information in an appropriate format. Often these agencies will
gather and update source information on a more or less continuous basis but
will generate updated inventories at fixed intervals, perhaps annually or
biannually. Thus it may be necessary to determine whether the information in
a compiled inventory is the most recent available. Often areas with special
problems, such as areas with a nonattainment designation, have had special
attention made to the completeness and accuracy of their inventories.
State and local air pollution control agency emissions inventories may be
limited to point sources in some cases. However, information on non-point
source emissions is sometimes available from these agencies. Often during the
formulation of SIPs, information has been developed concerning area source or
aggregated small source emissions. These are sometimes available on a county,
town, or gridded square basis. Information that was used in developing these
area source values, such as traffic counts, vehicle miles travelled (VMT),
miles of paved and unpaved roads, and population densities are sometimes
available. Aggregate fuel consumption statistics are often developed for
various types of fuels during SIP development, and these may be broken down to
county or grid square values.
There are three reasons for exercising caution when using area source or
fugitive emissions inventories: 1) the data used in SIP or nonattainment plan
development may have been valid only for the time period when the SIP was
generated and may need significant updating, 2} the emission estimates are
-55-
-------
generally based on emission factors that are imprecise, and 3) the inventories
usually do not contain all of the area sources that may be affecting the
ambient air quality. The state-of-the-art in emission factors for fugitive
dust sources is still far behind that for point sources and estimated
emissions can be incorrect. Area source inventories are useful for obtaining
an overall picture of the study area, but they may not provide exact modeling
results. This is particularly true since most area source inventories only
compile the major sources (such as roads, home heating, parking lots) and tend
to omit many others (such as industrial fugitive sources, dirt race tracks,
drive-in movies) which are potentially important.
The U.S. SPA National Emissions Data System (NEDS) is the standard federal
emissions inventory archive. NEDS contains information on both point source
and area source emissions. The point source data in NEDS are generally
obtained from State and local air pollution agency files. Area source
emissions are generally developed on a county-by-county basis from aggregate
statistics obtained from State sources. NEDS does not contain the source
emissions composition data used by receptor models.
In situations where only a single large source is under study it may be
necessary to develop a refined emissions inventory for the given source. Some
plants have had comprehensive inventories generated for use in permitting
applications or litigation. These are especially useful if reentrained dust
and fugitive emission sources were considered or if emissions were based on
actual operations rather than rated capacities as is usually done with State
inventories. Such in-plant inventories may or may not be available for use in
a given study.
There are also several "specialized" inventories which may be of use in a
source or receptor modeling study. One type of specialized inventory is the
-56-
-------
compilation of surface loadings and silt contents of road dust. This is
essentially a subset of a general area source or fugitive emissions source
inventory and is typically generated for a particular industrial facility.
The information on loadings and silt can be used with published emission
factors to provide emission estimates for various roadway segments.
Another type of specialized inventory is a compilation of particle size or
composition information. This information is obtained using a variety of
analytical techniques including optical and scanning electron microscopy,
x-ray diffraction, x-ray fluorescence and instrumented neutron activation
analysis (INAA). Such information is useful for "fingerprinting" a source so
that analyses of the material collected on hi-vol filters can be related to
their point of generation.
A third type of specialized inventory is the microinventory which
typically presents detailed information on point and area sources in the near
proximity of a monitoring station. The distances suggested for including
sources in a microinventory are a five mile radius for point sources and a one
mile radius for area sources. It is further suggested that all sources within
a one mile radius be included, but from one to three miles distance, only
those sources greater than or equal to 100 tons per year actual emissions need
to be included and from three to five miles distance, only those sources
greater than or equal to 250 tons per year need to be included.
Microinventories are useful for locating a local source, such as a playground,
which may be affecting the recorded levels at a monitoring station but which
would not show up in an area-wide inventory. Further information on
microinventories can be found in Pace (1978).
-57-
-------
3.2.2 Data Quality
The validity of any emissions inventory must be reviewed before its use in
any source or receptor modeling application. Validity relates not only to the
accuracy of the actual emission values or source parameters but also their
representativeness to the application in question. Emission values are often
based on annual factors such as fuel consumption, total production, etc. The
actual day-to-day operation of a source may vary considerably from these
annual average values. In some cases, a short-term inventory may be
warranted. In many instances, ambient data are collected for a study period
of only weeks or months rather than a full year. The actual operation of
important sources during the study period should be reviewed to determine if
the inventory data accurately reflect those operations during that time
frame. This review should extend to fugitive sources, as well as point
sources, to determine if fugitive generating activities, such as unloading or
conveying, traffic, and storage pile maintenance occurred with the same
frequency as estimated in the inventory.
As discussed earlier, inventories are generally considered valid for a
particular point in time and may require updating for a given receptor
modeling application. This may involve review of air pollution agency
permitting files for inclusion of new sources, reductions due to emission
controls, plant shutdowns, and operating limitations. Economic conditions may
also' affect emissions inventory validity through effects on total plant
production, hours of operation, etc. An attempt should be made to include
these effects in inventoried data.
3.2.3 Source Smission Compositions
Although few such data are currently available, size-segregated and
chemically resolved data are obtained for certain specific sources by
-58-
-------
analyzing "grab" samples collected from emission control devices, piles of
fugitive materials or in-stack sampling. For other sources, these composition
data are obtained by analyzing samples collected using sophisticated
ground-based plume sampling, dilution stack sampling, vehicle emission
sampling or airborne plume sampling. Technical considerations affecting the
use of these data include the compatibility between ambient and source
sampling periods, the existence of unique (i.e., tracer) emissions, the
temporal variability in emission compositions and the similarities in
composition of different source emissions. Further discussion of source
sampling and emission composition analysis techniques is contained in Gordon,
et al., (1983).
One of the biggest impediments to receptor modeling has been the lack of a
systemized library of source compositions. EPA is developing such a library
(Gordon, et al., 1983; U.S. EPA, 1984). The prerequisites of source
compositions for use in mass balance calculations are the following:
Chemical species and particle size fractions should be the same
as those commonly measured at the receptor.
Sampling conditions should be such that the composition at the
source is similar to that received at the receptor. This may
require dilution stack sampling or airborne sampling. For
certain substances which do not change state or composition with
temperature, less expensive grab sampling is adequate.
Operating conditions of the source must be thoroughly defined
since these may affect the emissions compositions. Source
configuration, throughput, fuel composition, and temperatures
should all be documented.
Very few of the source tests to date meet these criteria. The most
commonly used compilations of source compositions are Watson (1979) and
Taback, et al., (1979). Appendix B contains a list of sources for which
-59-
-------
chemical characterizations have been made and some of the references
containing the data from those characterizations. As mentioned above, a more
detailed library of the chemical characteristics of source emissions is being
prepared (U.S. EPA, 1984).
3.3 Meteorological Data
*
Meteorological data are an important part of any source or receptor
modeling study. Meteorology defines the transport conditions under which
pollutants move from the source to the receptor, and provides the link between
the two. Meteorology can have important effects other than simple transport.
Dust suppression due to rain, pollutant buildup due to stagnation, and wind
generation of fugitive emissions can have important effects on ambient
pollutant concentrations.
3.3.1 Data Quality
A large set of previously compiled meteorological data is available for
use in source or receptor modeling applications. However, since
meteorological data can be highly variable in space and time, the available
data must be evaluated for their representativeness to the given application.
The physical scale of the problem under study is especially important.
Microscale effects from fugitive sources often reguire on-site data, whereas
multi-state transport studies can use routinely gathered NWS data. If
source-receptor relations are being examined, the extent to which measured
data reflect conditions at both the source and the sampler should be
evaluated, particularly in regard to the sampling height and the mean height
of transport. In larger urban areas, multiple meteorological data measurement
locations are usually available and spatial averaging may be more appropriate
than the values from any single station.
-60-
-------
The accuracy and precision of meteorological instrumentation should be
evaluated before use of the data in a modeling study. EPA has published
specifications for meteorological instrumentation used for on-site prevention
of significant deterioration (PSD) monitoring purposes (U.S. EPA, 1980b).
Precision is usually a function of the instrument itself and can be obtained
from manufacturers' specifications. However, the recording and reduction of
measurements may affect the precision of the actual data produced.
Meteorological instrument accuracy is generally dependent upon the quality
control/assurance (QA/QC) procedures used by the operator. The NWS, which is
responsible for gathering most airport meteorological data, as well as the
EPA, have developed comprehensive QA/QC procedures for collecting field
meteorological data (U.S. EPA, 1982b). Data collected by other agencies
should have been subject to similar or otherwise acceptable procedures. Data
QA/QC procedures should include an effective procedure for removal or
correction of any erroneous values.
3.3.2 Data Sources
The largest source of meteorological data available for source and
receptor modeling applications is that data recorded at airports by NWS and
Federal Aviation Administration (FAA) observers. These data have been
archived at the National Climatic Data Center (NCDC) in both hard copy and
computer compatible forms. NWS data are produced by observations made at a
once per hour interval. Generally the observer attempts to record a value
which is an average over the several minutes previous to the observation
time. Thus, since NWS values are not hourly averages and may not represent
conditions over the full hour, care should be taken in using these data in
combination with hourly average ambient concentrations. Because NWS data are
-61-
-------
collected, reduced and reported according to standardized QA/QC procedures a
generally high level of confidence can be placed in their accuracy. The MCDC
archiving facilities also allow generation of large volumes of data in a
variety of formats not generally available from other sources.
The U.S. EPA, as well as State and local air pollution control agencies,
have often maintained meteorological monitoring networks in combination with
their pollutant monitoring networks. Generally the meteorological monitoring
is performed at the same location as a pollutant sampler. The collocation of
meteorological and pollutant sampling instruments can provide more accurate
estimates of near field transport conditions compared to airport data.
Pollutant monitoring-based networks generally have several sites in a given
region so that overall wind. patterns can be deduced. Like the NWS data,
monitoring network data are collected and reduced under standardized
procedures so a high level of confidence can be placed in their validity.
The EPA and State agencies have from time to time established special
monitoring networks during various air pollution investigations. The
meteorological data from these studies are sometimes archived at the EPA's
National Aerometric Data Bank (NADB). Many of these programs were developed
to examine the effects of a single large source and thus give finer spatial
scales. Private companies, especially electric utilities, have often
supported similar programs, and these data are sometimes archived at State
agencies. Data validity from these types of programs should be evaluated by
examining calibration procedures and results if they are available.
3.3.3 Meteorological Variables
3.3.3.1 Wind Speed and Direction.
Wind speed and direction are probably the most critical meteorological
parameters for source and receptor modeling applications since they define
-62-
-------
pollutant transport conditions. The most critical factor to be evaluated when
using available wind data is their spatial scale of representativeness. Wind
speed and direction can change markedly from place to place, especially in the
vertical or in areas with varied topography. Airport data are generally taken
in large open areas that are often elevated above the surrounding terrain.
These data may not represent conditions at industrial sites in river valleys
where the topography induces channeling and wind speeds are generally lower.
However, airport data may be more representative of the general wind flow over
an urban area than is a monitoring site tower located near an individual large
source. Many other airports located near coastlines are influenced by
seabreezes which do not affect inland industrial sources or urban areas.
Generally the spatial scale of representativeness increases with the
measurement height. Most airport data are measured at 20 feet above ground
and monitoring network data are often measured at 10 meters (33 feet).
However, airport data may be more indicative of mean wind conditions due to
the open siting requirements of airports. A few monitoring locations have
taller towers with instruments at multiple levels. These measurements can
sometimes be used to determine wind flow affecting tall stack sources as well
as wind direction in the immediate vicinity of the sampler. Rawinsonde
balloon measurements can be used for determining long-range transport or
regional conditions. Wind measurements made in areas of pronounced topography
should be evaluated for the effects of phenomena such as downslope flows,
obstacle steering of winds, and increased calms due to flow shielding and
stable stratification. These effects may mask the true transport pattern from
a source, especially if it is located in a topographic regime separate from
the measurement site.
As mentioned earlier, airport observations are actual short-term averages
made on a once per hour basis, rather than hourly averages. This is a
-63-
-------
critical factor for wind speed and direction measurements since the conditions
at the time of the observation can easily be anomalous when compared to hourly
average conditions. Thus pollutant monitoring network data, which are usually
generated as hourly averages, would give a more faithful picture of transport
conditions. Airport wind direction observations are recorded to the nearest
ten degree value while aerometric monitors can record to single degree
values. Wind directions recorded to the nearest single degree may be
advantageous when assessing near field transport from closely spaced sources
in an industrial complex. Aerometric wind measurements are also generally
made with lighter and more sensitive instrumentation than airport measurements.
3.3.3.2 Temperature, Precipitation, and Humidity
Although not as important as wind speed and direction, these variables can
affect the generation of fugitive emissions, ambient particulate matter
characteristics, and deposition patterns. Temperature, precipitation, and
humidity generally do not display the same spatial variability as wind speed
and direction, thus evaluation of spatial representativeness is not as
critical. Often existing airport data are sufficient for use in a source or
receptor modeling analysis. These variables are often not measured at
aerometric networks. Temperature and humidity measurements made on towers
should be reviewed if ground based fugitive sources are being modeled to
determine if the values reflect actual surface conditions. In some cases
where a single precipitation measurement is used for a large area, some
indication of the rainfall "spottiness" may be necessary to accurately assess
dust suppression abilities.
-54-
-------
3.3.3.3 Mixing Height
The concept of mixing height is widely used in source models and can be
utilized in some receptor modeling analyses as well. Generally, the mixing
height is considered to be the height at which the atmospheric lapse rate (the
rate of change in temperature with height) suppresses further upward mixing of
pollutants. The upward mixing of pollutants may be suppressed when the rate
of temperature decrease with height is less than that at lower heights or when
the temperature increases with height (producing an inversion). The
height/presence of such a lapse rate can only be deduced through vertical
atmospheric soundings. The only widely available source of such data is NWS
twice per day rawinsonde measurements, which are archived at NCDC, however,
some data may be available from utilities, industries, and regulatory or smoke
management agencies (e.g., the U.S. Forest Service). In non-mountainous areas
use of data from the nearest NWS sounding station is generally adequate
because of the broad spatial scale of the data. Capping inversions sometimes
occur in valley locations which may not appear in the airport based sounding
measurements.
The ideal method of developing mixing heights from balloon data is through
individual inspection of each sounding. This may be impractical for a
long-term application, however, and the modeler must rely on mixing heights
developed from analysis of sounding data by standard algorithms, usually the
Holzworth procedure (1972). Some review should be made of the algorithm being
used to determine if it is accurately assessing the mixing heights at the
location under study.
-65-
-------
3.3.3.4 Stability Classes and Dispersion Coefficients
Stability classes and dispersion coefficients are widely used to quantify
atmospheric dispersion processes and are central to the formulation of the
Gaussian source models discussed in Section 2,2. Stability classes and
dispersion coefficients are generated from other meteorological variables.
Thus, their degree of representativeness and validity is dependent upon those
measurements. A variety of stability classification schemes can be generated
from different types of meteorological measurements. The lack of a certain
measurement will restrict the user to schemes based on those measurements that
are available. The Turner method (1970) develops stability classifications
from widely available airport data. Other classification schemes requiring
measurement of temperature differences (AT/AZ) and the standard deviation
of horizontal or vertical wind directions (a
-------
Particulate matter nonattainment problems are identified using data
obtained from networks of samplers, such as hi-vols operated for the purpose
of determining compliance with NAAQS. Thus, a basic requirement is to have a
data base of several years for a number of fixed stations representing not
only the nonattainment area itself but adjoining areas including one or more
"background" stations. Data from dichotomous samplers or other types of
size-selective samplers, when they are available, can also be especially
useful since sampling results can be used directly to infer source category
contributions based on size ranges (e.g., large particles tend to be from
nearby sources of fugitive dust and small particles tend to be from combustion
sources, sulfates or distant sources).
Since trend analysis is an important technique to use on hi-vol data, it
is recommended that at least three years of data with a sampling frequency of
at least once every 6 days be available for each sampling station to be
considered in a trend analysis. It will facilitate the analysis if these
data, together with appropriate meteorological data, are archived on a
computerized data base and made accessible to manipulation by suitable
programs.
Chemical analyses of the particles collected in ambient samples are
extremely useful in source apportionment studies. As pointed out elsewhere in
this document, chemical and microscopic (optical and scanning electron)
examinations of selected samples provide further insight into sources
contributing to the measured levels, especially when correlated with
meteorological variables.
3.4.1 Data Quality
No ambient sampling methodology is perfect for depicting the actual nature
and distribution of particulata matter in the atmosphere. Important factors
-67-
-------
affecting the validity of particulate matter data include sampling location,
instrumentation, sampling schedule, filter material and the methods used to
prepare the sample for analysis (U.S. EPA, 1983b; Gordon, et al., 1983).
The siting of samplers is especially important in source apportionment
work. Samplers shielded by overhanging trees or nearby buildings can produce
non-representative results. Samplers located near busy road intersections,
while typical of locations affected by a strong nearby source of fugitive
emissions, will be insensitive to the contributions to that location from
other types of sources. Guidelines developed by EPA for National Air
Monitoring Stations (MAMS) and State and Local Air Monitoring Stations (SLAMS)
can be used to evaluate existing sites to determine if they meet acceptable
standards for representative ambient sampling (U.S. EPA, 1979b).
Particulate matter sampling is usually performed using hi-vols, sometimes
with cascade impactors or size-selective inlets attached, or with dichotomous
samplers. Problems with the standard hi-vol include an inability to separate
particles by size and imprecise collection efficiencies, as affected by
particle size and wind speed and direction. The cascade impactor and size-
selective inlet collect material by particle size but retain the other
problems of the hi-vol except that sampling with the size-selective inlet is
relatively unaffected by wind speed and direction. Dichotomous samplers
divide ambient particles into a fine and a coarse fraction but dichotomous
sampler -collection efficiencies are affected by wind speed and humidity (John,
et al., 1983).
Most hi-vol sampling is conducted on an every sixth day sampling schedule
and produces samples collected over a 24-hour period. This sampling schedule
provides representative samples for a year but obviously omits sampling on 5
out of every 6 days. The 24-hour sampling period precludes the measurement of
-68-
-------
diurnal or other similar variations in particulata matter concentrations. The
amount of particulate matter collected during a 24-hour sampling period may be
too much or too little for certain types of constituent analyses. Another
consideration is the compatibility between periods of ambient and source
sampling.
Most hi-vol particulate matter samples are collected on glass fiber
filters. On the whole, such filters are useful for total particulate matter
sampling but are less suited for particulate matter constituent analyses.
Difficulties with glass fiber filters include particle penetration into the
filter and trace element contamination. In addition, the particles on glass
fiber filters tend to undergo chemical reactions with each other, with the
filter and with gases in the air. These chemical reactions usually involve
acid aerosols and acid gases. This problem, called artifact formation,
results in erroneous reported concentrations, primarily for the sulfate and
nitrate particulate matter fractions. Various types of membrane filters are
also used to sample particulate matter. Membrane filters overcome many of the
problems associated with glass fiber filters and are the best filters for
microscopic analyses, but samples have to be treated carefully because
membrane filters are brittle and particles readily fall off them.
Once collected, the particles may be analyzed to determine their
composition. In some cases this can be done on the filter, but in others,
particles must be removed and inserted into an instrument for examination. Of
concern are particles embedded in the filter and particles on top interfering
with the analysis of particles underneath.
-69-
-------
3.4.2 Data Sources
In the United States most ambient particulate matter data are gathered by
federal, State, and local air quality regulatory agencies. Monitoring sites
are distributed throughout the nation to monitor compliance with NAAQS. Most
of these data are archived using the Storage and Retrieval of Aerometric Data
(SAJROAD) system in Research Triangle Park, North Carolina (U.S. EPA, 1971).
Large amounts of additional data are collected by utilities and industries at
and near their facilities. The bulk of these data have been collected using
hi-vol samplers, glass fiber filters and an every sixth day sampling
schedule. For many years EPA has been compiling particulate matter chemical
composition data from designated National Air Surveillance Network (NASN)
stations. Such stations are generally a subset of the SAROAD sites operated
by State and local agencies. The samples are collected on glass fiber filters
using hi-vols operated on an every twelfth day sampling schedule. Chemical
composition data are obtained using atomic absorption and wet chemistry
methods. EPA has also been compiling particle size and chemical composition
data from an Inhalable Particulate Network (IPN) of sampling sites operated in
major urban areas (Watson, et al., 1981). Additional comprehensive air
quality data are available for St. Louis, Missouri from the Regional Air
Pollution Study (RAPS) conducted during the mid-1970s (Strothmann, 1979).
Such potential sources of ambient data should be considered before any new
sampling is conducted.
-70-
-------
4.0 COMPOSITE SOURCE/RECEPTOR MODEL APPLICATION PROTOCOLS
There are many source and receptor models available for use in qualitative
and quantitative determinations of the sources responsible for measured
ambient particulate matter concentrations. As discussed earlier, the
limitations and underlying assumptions of each model are such that all fall
short of a perfect representation of reality. Therefore, when seeking
solutions to air quality problems, it is highly advisable to use more than one
model. A higher degree of confidence can be obtained if several models,
operating independently, produce the same results.
4.1 Considerations in Method Selection
This document cannot describe all of the possible combinations of source
and/or receptor models that can be used to apportion the contributions of
sources to ambient air quality. There are five factors that must be
considered before beginning the source apportionment process. These are:
the time frame of the problem;
the existing data base;
the nature of the problem;
the applicability of complementary methods; and
the resource availability
Each of these factors is discussed below as a preface to the discussion of
a three level approach to source apportionment. Many of the particulars of
each factor are discussed in more detail in other sections of this document.
4.1.1 Time Frame of the Problem
The selection of methods must begin by a recognition of the time frame of
concern, annual versus 24-hour, which depends on the existing air quality
versus the corresponding NAAQS. An annual time frame offers the most
-------
flexibility for method selection. Some methods such as the multivariate
models (e.g., factor analysis, TTFA and MLR) and the dispersion models work
best using multiple samples. If only selected days require consideration in
order to develop a control strategy, such methods as chemical mass balance or
microscopic analyses may be preferred. However, as described in Section 4.4,
an annual time frame study may be performed by merging the analysis results
from carefully chosen individual samples.
4.1.2 Data Base
Both ambient and source data bases must be considered. The compatibility
of filter material with the needed chemical analyses is important. For
instance, the concentrations of certain materials, such as silicon and organic
carbon, cannot generally be determined on. glass fiber filters because of
interferences from the filter material and binder (Pace, 1983). The nature
and variability of source emissions may severely limit the use of dispersion
models for short-term problems because of the difficulty of reproducing a
reliable 24-hour inventory. Generally, receptor models are preferable to
dispersion models for short-term retrospective analyses for this reason. This
is reflected in Tables 4-1 and 4-2 which suggest preferences for selecting
source apportionment methods based on the time frame of the problem, the
ambient data base and the filter media (Pace, 1983). The availability of
source mass emissions or chemistry data may be limited by whether it is
possible to test the source or to predict the timing or consistency of its
operation.
4.1.3 Nature of the Problem
The complexity of the problem, (that is, whether it involves many
suspected sources or a single likely source), influences the selection of
methods. In a simple situation, some of the less quantitative or screening
-72-
-------
TABLE 4-1
PREFERRED APPROACHES FOR SOURCE APPORTIONMENT (Pace, 1983)
Ambient Data Base Available
TSP*
PM
i o
Time
Scale
of the
Nonattainment
Problem
Annual
24-hour
(a) applicable dispersion
model corroborated by
SEM or optical
microscopy
(b) applicable dispersion
model
(a) applicable dispersion
model corroborated by
SEM or optical
microscopy
(b) applicable dispersion
model
(a) applicable disper-
sion and receptor
model
(b) applicable disper-
sion model
(c) receptor methods
(at least two)
(a) receptor and
applicable dis-
person model
(b) receptor methods
(at least two)
(c) applicable disper-
sion model
Assumptions
* 1. TSP is collected on glass fiber filters, other filter media may be more
suitable for receptor modeling.
2. Short-term emission inventories are difficult to obtain because of the
nature of the source.
-73-
-------
TABLE 4-2
APPROPRIATE FILTER MEDIA FOR USE WITH RECEPTOR
MODEL STUDIES (Pace, 1983)
Laboratory
Technique
Elemental Analysis
(Mass Balance)
(Factor Analysis)
Filter Type
Glass
Fiber
Quartz
Fiber
Polycarbonate
1
Fluorocarbon
Optical Microscopy
Automated SEM
Carbon
2
2
2
2
2
1
2
1
3
2
2
3
1. Recommended
2. Useful under some circumstances or to some analysts
3. Not recommended
-74-
-------
methods may be adequate. In such cases, combinations of some of the simpler
dispersion models, such as the PT series, or PAL, with receptor models such as
optical microscopy, tracer methods, or spatial or temporal correlations, may
suffice. Multiple sources may require the more refined dispersion models such
as MPTER, RAM or ISC along with chemical mass balance or multivariate receptor
methods.
The spatial extent of the problem affects the size and type of emissions
inventory needed. If the problem is largely transport the applicability of
the available models may be limited.
The nature and chemistry of the emissions of the suspected sources is
important for several reasons. Some sources are precursors of secondary
particulate matter such as sulfates.. Determining the sources of these
sulfates generally requires a dispersion model that accounts for chemical
transformations. In may cases, particularly in the eastern United States,
regional scale models would be needed. The chemistry of the emissions of each
source or source category must be sufficiently different to enable the
chemical mass balance and the multivariate methods to distinguish among them.
Sources that are chemically very similar cannot generally be distinguished
using these methods.
The size range of the emissions of tha suspected sources affects method
selection. Sources that emit primarily large particles include the "fugitive
dust" sources such as windblown or mechanically generated or resuspended
dusts. Large particles are amenable to analysis by optical or scanning
electron microscopy or x-ray diffraction but, if they come from a variety of
soil or crustal related sources, their specific sources are difficult to
distinguish by chemical methods. Pine particles, generally defined to include
those smaller than 2.5 jam, are usually associated with combustion sources.
-75-
-------
They are too small for optical microscopy and, in many cases, they are
primarily carbon. In that case, they can only be distinguished by the
chemical methods if they have a unique tracer such as lead in gasoline or
vanadium in fuel oil. Fossil, (as in fuel oil or diesel fuel) and contemporary
(as in wood smoke or vegetative burning) sources of carbon can be
distinguished to some extent by carbon dating chemistry (Stevens and Pace,
1984).
4.1.4 Complementary Methods
The complementary use of models can involve the use of two or more
receptor models or the use of a combination of receptor and source/dispersion
models. Table 4-3 shows the models best suited to complementary uses. The
models most suited to complementary uses are those that operate on different
data bases, algorithms and assumptions such that source contribution estimates
are obtained independently.
In viewing Table 4-3, receptor methods may be thought of as three distinct
groups of methods that represent, conceptually, three distinct perspectives on
source contributions. The first of these groups, the qualitative methods,
generally rely on ambient mass concentrations, often in conjunction with
meteorological data. The second and third groups are listed as quantitative
methods in the table. Group 2 includes mass balance (MB), factor analysis
(FA), Target Transformation Factor Analysis (TTFA), multiple linear regression
(MLR), and use of tracers. These Group 2 methods all rely on chemical
features of the sources and the bulk ambient samples. In contrast. Group 3,
consisting of optical microscopy (CM), scanning electron microscopy (SEM) and
X-ray diffraction (XRD) relies on interpretation of chemical and physical
-76-
-------
1
4
i
]
"3
h
a
u
i
3
S
!
I
«
W
a
M
u
i
a
5
n a
^ I 2
I 8 |
3 K *
" S 2
s §
§ 1
S
<
1
X
8
fet
O
M
i
3
-
"3
j
*
a
ae
z
G
<
(J
«
>
i
*
\
a
:
0
J
I
^
J
I i 2
s u .1
ol u u
21
fig' j
\\ii
1
ft
X J U U U U U
f <
ft. a.
M n 91
5 a fcj a e < < a a
E sy
j|
u] a a A a a a a
H
1 -
11
1
- a a < « < < < <<<<
h.1 <<4VV a a a a a a a
d
2|
"I «o*««oo <<« <<<
H
U H *
n y| ^
j e U
3 ol * -^
** c S
« d 1 5
a Sj o # «
« t|l
'9 i* ° '
a 3 u a 4 «<44<<>««<< * * < <«< ^ a 2
* i i !?l
S a 57 : - d _. ^j*
2§.i|X aSl 1 i . ;*£
lg-3ftTll? -Isl^a-c .7-1 « 111
4 a sl=El2!*2 i s.'SI'a sis Ic : « a 2 ls|
«*« ^*OfliM«»^o^cc«^a^-*o "Kjf"* *" o « 5
^ o e«|-^j'^_»'- ^ o^ffoQ'^** a * cj »w ^*JQ
a a a^Ciw a OWM^^KM'^M* u to 2 * au &uou scnui
V « * & 444&UI3M u t*t*V * O) 3 Sfl -
S a3n a xt-nffaxx* g E £ > w «*- d'N>
9E fl> Ml < a U
77
-------
features of discrete particles found on the samples. Dispersion modeling
provides yet a fourth group, or perspective, by virtue of its reliance on
source emission rate estimates, meteorological data and dispersion parameters.
In general, the most complementary methods are those that are from
differing groups, bringing differing perspectives to the analysis. Thus,
methods within a group are usually most effectively used in conjunction with
methods from a second group and not with other methods from the same group.
Perhaps the most noteworthy is the combined use of dispersion models with any
of the other three groups.
The following is a further discussion of the complementary aspects of
selected features found in Table 4-3:
4.1.4.1 Background
A background determination complements and is even essential to almost any
source apportionment study. It is useful in identifying particles of natural
origin that are not subject to control; but also, it helps estimate the
relative contributions of local influences as compared to the long range
transport component.
4.1.4.2 Data Distributions
Analyses such as investigations of lognormality and frequency
distributions are not particularly useful for source apportionment by receptor
modeling, although they are needed to determine the representativeness of the
days being modeled. This analysis may be useful for dispersion modeling where
hypothetical meteorological conditions are used and the representativeness of
these conditions vis-a-vis measured values needs to be determined.
-78-
-------
4.1.4.3 Time Correlations
Time correlations give some preliminary insights into a problem. They are
essential if the suspected source operates intermittently and changing
meteorology can be easily accounted for, but in most cases the meteorology
confounds a time series analysis such that temporal differences in
concentration cannot be attributed to sources by time series alone. More
elaborate statistical analyses such as multiple linear regression are usually
needed.
4.1.4.4 Spatial Correlations
In this discussion, "spatial correlations" include correlations with wind
direction and trajectories, as well as observed differences or simulations
concerning sites measuring particulate matter. Spatial correlations are
usually more useful than temporal analyses for particulate matter because the
localized nature of many particulate matter sources lends itself to spatial
comparisons. Such analyses may be particularly useful in conjunction with
24-hour analyses such as mass balance, macroscopic methods and tracer analyses
when their results are classified or grouped by wind direction. Spatial
correlations are a reasonably effective supplement to dispersion modeling.
For example, pollution roses can be used to check the reasonableness of source
contributions determined by dispersion models and trajectory analyses can
provide valuable insights into background contributions which are unaccounted
for by dispersion models.
4.1.4.6 Mass Balance
Mass balance is generally appropriate and useful with any other method,
but is especially useful with dispersion models. Methods such as factor
analysis, multiple linear regression and tracers are often done as a
-79-
-------
preliminary step before doing a mass balance. These techniques may be useful
in selecting which sources and fitting elements to consider in the mass
balance approach.
4.1.4.7 Factor Analysis
Factor analysis provides information on which groups of aerosol features
(elements, etc.} vary together; these groups can be interpreted as source
categories. Thus, FA is moderately useful in determining if dispersion models
have "missed" a major source category. Factor analysis is useful with other
receptor techniques because it provides insight into source fingerprints.
This information is vital to MB and MLR and helpful to microscopic studies.
Target Transformation Factor Analysis is a form of factor analysis, thus it is
redundant with that method.
4.1.4.8 Target Transformation Factor Analysis
TTFA is a combination of MB and FA and thus is redundant with these
methods. It is useful as a complement to the microscopic methods, just as FA,
MB, or MLR alone would be.
4.1.4.9 Optical Microscopy/Scanning Electron Microscopy
Polarizing light or optical microscopy
-------
4.1.4.10 X-Ray Diffraction
This method is particularly useful for identifying specific crystalline
structures in soils and thus, is most useful when soil has been identified as
a major source component but the methods used, (e.g., mass balance or the
multivariate methods) cannot provide more specific information on sources.
The x-ray diffraction analysis can only help when the suspected soil sources
have differing minerology or crystalline structure.
4.1.4.11 Multiple Linear Regression
MLR is particularly useful with optical and scanning electron microscopy
because it relies on different information or features of the ambient sample.
It is also used to help quantify the contributions of source groups identified
by factor analysis. It is usually redundant with mass balance, TTFA and
tracer methods. It is most useful when chemical composition data for a large
number of observations are available.
4.1.4.12 Dispersion Models
The dispersion models generally require an analysis of background
concentrations because as much as half or more of ambient particulate matter
in an urban area may originate from beyond the local airshed and thus would
not be included in the model. Dispersion models in general are complemented
by receptor methods, especially microscopy and mass balance analysis. This is
because receptor methods bring an entirely different perspective to the
analysis and rely on different information and assumptions. They are
especially useful to identify uninventoried or misinventoried sources or
source categories. The microscopic methods are a useful complement to
-81-
-------
screening or refined models and are also helpful to corroborate the refined
dispersion models. The screening models such as the PT series are useful
preliminaries to a refined model analysis.
4.1.5 Resources
No discussion of method selection would be complete without a discussion
of resource considerations. The selection of methods is affected in some part
by the expertise and experience of the staff in the various methods, as well
as their availability. Equally important is the availability of funds to
purchase consulting or contract assistance. The time available to come up
with an answer may also preclude certain types of analyses, particularly those
that require data gathering.
One way of conceptually structuring a study design so that time, personnel
and available funds are considered is the three level approach (Stevens and
Pace, 1984). The three levels are not "cookbook" recipes since all of the
ingredients (methods) and their amounts (resource requirements), etc., are not
firmly fixed. The three levels are intended as rough guides, or benchmarks,
and are flexible to accommodate the wide range of real world situations and
constraints that are likely to be encountered. Within each level, the time
frame and nature of the problem, the existing data base, the applicability of
complementary methods and the available resources must be considered.
4.2 A Three Level Approach
Level I uses existing data or data that can be readily obtained from
analyses of existing samples. The models used are those that apply to the
data set and that do not require extensive resources for their computation.
In Level I, basic analyses are used to identify contributing sources and, if
-82-
-------
possible, obviate the need for further analyses. Otherwise, the Level I
analyses will serve to narrow down the areas to be studied in more detail in
Level II. Some source-types can be eliminated from further consideration
after Level I is finished, thereby focusing data gathering efforts in
Level II. Combinations of source and receptor models can be used in this
process. If the sources contributing to the high concentrations of
particulate matter are apparent and sufficiently certain, no further work will
be needed beyond Level I.
Level II encompasses more refined analyses than the first, including
performing additional analyses on existing samples, acquiring new samples from
existing monitoring networks and a limited amount of new sampling. Such new
sampling could be used to obtain component "fingerprints" for source emissions
(e.g., road surface loadings and silt content) or selected ambient samples.
Level II uses models that require more detailed inputs and larger
computational facilities. However, models and analytical methods are of a
standard nature. Combinations of methods can be used.
Level III involves the acquisition of new data from new sampling
activities and programs. These programs can be quite extensive, e.g., the
deployment of a special monitoring network in order to obtain a year's worth
of particle size-segregated samples suitable for detailed chemical
characterization. Analysis procedures can incorporate a variety of models
with very detailed inputs. Models and analytical methods may be of a
nonstandard nature, i.e., developed exclusively for the project.
Suggested approaches to these three levels are described in the following
subsections. The suggestions and recommendations made are not rigid
prescriptions but are simply meant to convey the levels of analysis that may
be required, based on the collective experience of the authors.
-83-
-------
4.3 Level I: Gathering, Evaluating, and Using Existing Data with Simple Models
The objective of a Level I study is to determine the likely causes of a
given pollution problem with only a modest investment of time and resources.
In a Level I study, existing data and simple analyses are used to obtain
answers to questions pertaining to source/receptor (source/monitor)
relationships. A Level I study does not involve collecting any new data
through new monitoring programs. The first step in such a study is to gather
the existing data, the second is to evaluate the data, and the third is to
determine if additional data are needed and are feasible to obtain. Needed
for these analyses are source, meteorological, and ambient data. The
collection, compilation, and review of these data are discussed first below,
followed by suggested analysis procedures. While a Level I study may, in some
cases, (e.g., where the culpable source is isolated and easily defined,} be
adequate for control strategy development, in most cases, a Level I study will
serve as the basis for defining the scope and nature of a Level II or Level
III study.
4.3.1 Source Data
Various types of source data can be used in a Level I analysis. Such data
include emission rates by total mass or particle size, composition, and
variability with time, and source configuration parameters (location, height,
flow rate; temperature, etc.). These data are discussed in Section 3.2.
The available data should be organized and reviewed to identify likely
important contributors to the ambient pollutant levels. An overview of the
major source-to-monitor alignments and initial insight into the likely
important contributors to the measured pollutant concentrations can be
obtained if the major point sources are plotted on a map of the study area.
Additional information on likely contributing sources can be gained by
-34-
-------
gathering and reviewing the SLAMS and NAMS site surveys (U.S. EPA, 1979b) and
data from any previously developed microinventories for the area(s)
surrounding (within a radius of 2 km) the monitor(s). If SLAMS and NAMS site
surveys do not exist, such surveys can be performed with minimal effort for
sites of critical concern in a Level I study, or they can be delayed until it
is determined whether a Level II analysis is needed.
The source emissions data should next be organized and compiled.
Compilations by source-type categories (e.g., power plant, steel plant,
municipal incinerator, traffic-related), with the emission rates in each
category summed and then sorted by the category emission totals are
recommended. These source groupings can then be maintained through subsequent
dispersion modeling analyses (e.g., ISC) and also used for comparison to
measured concentrations (by particle composition, if available). If particle
composition data for the source emissions (chemical species, particle sizes,
etc.) are available, they could also be tabulated for each source, starting
with categories of highest emission rates. From these tables, chemical
species unique to specific sources (i.e., tracers), and the dominant chemical
species and particle sizes in the study area can be identified.
4.3.2 Meteorological Data
All readily available meteorological data should be collected and
evaluated for validity and representativeness to the study area. Possible
sources and evaluation procedures for meteorological data are described in
Sections 2.3.7 and 3.3. Once a valid, representative meteorological data set
has been assembled, it is recommended that the locations of all the wind
measurement stations be noted on the same map with the major point sources and
that 24-hour vector average wind speeds and directions be generated for the
-85-
-------
days of interest. Vector averages at each station for a given day can then be
plotted at each of the meteorological station locations. Once the wind
vectors have been plotted, the daily wind fields can be analyzed for dominant
directional patterns and any anomalous wind directions due to channeling
effects. Other suggested evaluations of meteorological data include comparing
tall tower data (if available) with surface level data (to estimate any change
of direction with height) and comparing local flow patterns with the overall
synoptic pattern apparent from NWS surface weather maps (to assess long-range
transport).
4.3.3 Ambient Data
Measured ambient pollutant data should be gathered from all existing
monitoring networks. Historically, most particulate matter monitoring has
involved hi-vol samplers which collect TSP data. More recently, some sampling
has been conducted using dichotomous, size-selective hi-vol or other samplers
hich collect PMio data. Most Level I studies will involve the analysis of
TSP data. Depending on the analyses performed, such studies may also provide
information regarding source contributions to PMio concentrations.
Potentially helpful data bases are described in Section 3.4. In order to
assess the source-to-monitor relationships, the locations of the monitors can
be plotted on the same map as the source and meteorological data.
4.3.4 Procedures (Level I)
A Level I modeling study involves performing a series of relatively
simple, straightforward analyses designed to enable the analyst to deduce
cause and effect relationships between source emissions and ambient
concentrations. Each analysis can provide information to make it possible to
-86-
-------
eliminate insignificant sources and focus in on important contributors. Work
performed by Brookman (U.S. EPA, 1983d), and summarized in Section 5.2,
provides a good example of a basic Level I study approach. Additional
protocol information is contained in the study report.
4.3.4.1 Qualitative Receptor Modeling Analyses
A good place to start conducting a source apportionment study is with the
qualitative receptor modeling analyses described in Section 2.3.7. These
analyses can be performed readily using total particulate matter
concentrations, and, if data are available, they can also be performed using
constituent and size-segregated concentrations. The results of such analyses
must be interpreted with due consideration of the sampling instruments, filter
media, and other factors involved in the collection of the particulate matter
samples, as discussed in Section 3.4. An appropriate initial analysis for a
Level I receptor modeling study is the determination of background
concentrations. By conventional definition, background concentrations are
those caused primarily by sources located outside the study area. Background
concentrations can be developed for both long-term (annual, seasonal, and
monthly) and short-term (< 24 hour) periods for the various particulate
matter concentrations. Several years of data can be combined to obtain
sufficient samples for short-term and monthly averages.
-The next analysis of interest is the trend analysis in which annual mean
ambient concentrations, less the corresponding annual background
concentrations, are plotted as a function of time (for as many years of data
as are available). The results can then be compared to yearly emissions data
(if available) to see if there is any agreement between changes in ambient
levels and events such as source shutdowns, source startups or implemented
-87-
-------
control strategies. A correspondence (i.e., 1:1) between percentage changes
in total emissions versus percentage changes in total ambient levels (less
background) is supportive of the accuracy of the emissions inventory. Poor
agreement may be indicative of deficiencies in the inventory. If no trend
emerges, it may be possible to obtain some trend information if the individual
daily concentrations are first grouped according to recurring patterns of
synoptic (i.e., air mass) scale meteorological conditions. When these
averages are plotted, trends may become evident for certain wind directions
relevant to specific source-to-monitor alignment directions.
Next, more detailed temporal and spatial correlation analyses can be
performed. These include seasonal and monthly trend analyses, weekday/weekend
analyses, wet/dry day analyses, episode day analyses and spatial mapping.
Seasonal and monthly trend analyses can be used to identify traffic,
agricultural, and other fugitive dust influences. Weekday/weekend analyses
can also help identify traffic-related impacts. Wet/dry day analyses can
provide an initial estimate of the magnitude o£ the total non-traditional
fugitive particulate matter impacts. Episode day analyses are useful for
separating local from distant source contributions. Spatial mapping can be
used to show source contribution patterns for various time periods (annual,
wet/dry days, episode days, persistent wind days, etc.), with comparisons from
one time period to another revealing the relative magnitude of various source
category influences.
Correlation coefficients and time series analyses can also be used to
relate receptor concentrations to sources. Analyses performed on particulate
matter data for multiple sites can be used to identify sites and periods
influenced by local sources. Analyses performed on particulate matter data
-88-
-------
and other parameters can be used to help reveal the location and relative
influence of the nearby sources. Also suitable for further study are those
cases where concentrations at all sites rise simultaneously (i.e., episode
days).
If the wind direction was persistent on the sampling days of interest,
pollution roses can be of value in determining contributing sources,
especially if chemical composition data are available. If the bulges in the
pollution rose concentrations at multiple sites point toward the same
location, triangulation may help pinpoint specific sources.
4.3.4.2 Dispersion Modeling Screening Analyses
For the sampling sites of concern, screening source/dispersion modeling
analyses can provide information concerning source contributions based on the
inventories of source data obtained for each site. Various models can be used
for this purpose depending on the objective of the analysis, the type of
sources in the area and the availability of computational resources. The
primary objective of a screening dispersion modeling analysis is to eliminate
insignificant sources from further consideration. Dispersion modeling to meet
this objective is discussed below. This screening dispersion modeling will
also provide information on the relative magnitude of source contributions and
thus help focus the direction of subsequent analyses. Dispersion modeling
techniques designed to better quantify source contributions are discussed in
Section 4.4 (Level II).
If possible, the model selected for a screening analysis should have the
capability to 1) predict individual source contributions at user-defined
discrete receptor points (i.e., the monitoring sites of concern), 2) determine
contributions from multiple sources, 3) accept user-defined meteorological
-89-
-------
data, 4) account for particle settling (pollutant deposition), and 5) predict
contributions from point, area, volume (i.e., certain elevated area), and line
sources. Ideally, the model should also be able to directly determine
contributions for the time period(s) of concern and be able to reliably
calculate impacts in locations with complex terrain. Since receptor modeling
analyses are generally concerned with particulate matter collected on filters
during short-term (1-24 hour) time periods, discussion of screening dispersion
modeling analyses in this subsection is limited to these short-term time
periods. A comprehensive evaluation of the capabilities of some dispersion
models, for use in receptor modeling applications, is shown in Table 4-4. The
model capabilities ratings in Table 4-4 provide a useful measure of model
features, but when models are selected for use in specific situations EPA
guidance (U.S. EPA, 1978a) should also be considered. See Section 2.2.2 and
the respective model user's guides for additional discussion of the
capabilities of these dispersion models.
If computer resources are not available, rough estimates of the potential
magnitude of the source contributions can be obtained using published
nomographs or by performing hand calculations following procedures described
by Turner (1970) or the Guidelines for Air Quality Maintenance Planning and
Analysis Volume 10R (U.S. EPA 1977a). If limited computer resources are
available, such estimates can be obtained using the EPA UNAMAP models PTMAX,
PTDIS, "or PTPLU, but all three of these models are designed for point sources
only. PTMAX and PTPLU will not predict source contributions at user-defined
receptor locations. The above methodologies can be used to eliminate
insignificant sources from further consideration, but are generally inadequate
for estimating the combined contributions of multiple sources. Furthermore,
none of the methodologies have the capability to account for particle
deposition.
-90-
-------
i
co
M
t*
5
|
<
i y
^ j
« 5
J o
< *
H _
§
2
5
S
£j
X
B
1
B
i
B
j
i
B
£
K
S
S
g
g
J
BM
Cb
W
1
3
g
M
i
1
I
1
i
MfiCt MMM^ W * ^*r-»W<*«
-
** W
rtKnMM otMMM Mmm mmMwiiAtn
>i
a
x* * A
« M * d
-*o* *°^ i 2
»* » w e u x d 91
J'iv - 1- * - 1 *
S-i- S*S.Jl a
: .: "! s* 11 3 .-1 s 5
* * S"il > * « o ^ a « o o«« ** o « m
s .».--. s^is-a 2. 1 s ua= : s?-ss-:i
^ sfi.:? §ss:s« 55' - §25 s iijsu-s.
BM O««q^» JJ g uv*w^*«^9VQ,una
i jj £ - j ^ ^ j. - j< y J ^ S 9 «>* S o S.-
3 41 J O * 0. * M 0. Oft 88 « « 2 ^ U S 0 J H H . J * < H * S
91
-------
r
S
i
Ei
0> "
i <
5 a-
jj <
c u
I i.
7 1
i 2
5 |
a
on
u
1
to
f
u
09
1-4
U
i
«
3
CO
z
i
j
8
M
a
i
o
u
a
z
00
i
84
s
z
8
J
<
04
Ol
&
01
H4
a
S
Oi
£
X
X
a
s
o
3
1
JJ
i
4)
O
O
X
in
Ioa
rt -4 <-jj -4 in
9t
f^
e, -, 0. S
9
2
X
0
o a
o si a
M S 01
" * e
B «w 44
a c! a u jj
cOa>t3-»Bg
o-4jjeo»-4.S
" c a. 5 rf «4
u 4 e t >4 f
«O44o-ii.>e<
i444u«oixoa
-4'«2"1*'ocg2
a a 41 a 44 5 1-1
C«O»JJ«3-io-4 3 jj a a
5 jj -, u o o <
ji -. o jj . x . i
3--*43> «>,
4 <0a W 144)5 341-J
(0 jj44 a K o Q*S<
u u) a 2 au 3) e<
3 JJ O
K 0 S
U
'S
O
u
c
o
u
V
*
IU
a
S
a
'JJ Is 5
3 Q, C
a e a o
-* 0
? « ? S
< u o>
a S S j
44 u a a
jj 3 >, j< a 4*
v^ O -4 U JJ) o O
§£ u n jz u u
13 JJ 0> 33
-i « o a 44 o o
o w rn a a a a
> rtoa OJ
05-= 81 jj o jj
n u j3 jj oo N »
C si ut ^4 MM & c o Q
44 _- 95 -> M -fc». - .5 * - * Z
U i*^ >^ >H ^1
9 9t 4) 41 O
(B a-c kc( .2 MS
O
t-4
JJ
fl
W
9 at C o c «
a S o a o c
N *H -M. ^ -m
^4(9 a 4j « « tu
s §. i«.i
a « o o a 1
444 <^ >4 IVOIUU
o 4) a -4 >UQ«
441 C*ji4 C *»4>a
jj O a o o >> -3 as a a
LIU -4 5 _ s «; a
ao e~2^
O*H>^ 0
C J 0 * * 4C -<*-HOMCB
^O>W ^atDOR -4Q CTi--. U fl, »
.W -C O O O W JJ <^*Af'QBSO
Jc .u « w w w jjio (duaBu
O m u 3 3 3 WO 50 -« H Q*
^ -^ C U O Q O «S 0 U 301V
u jj 11 3 ca co >. to >. u - oo as*-*
JJ U ^H O JJ^ .U>tUOj|
o c ,3 « « -H « --« u « e jj
-M 3 » ^^^^J-H O *J -.4 ca -4 » »
J= ..H -- ^, J *-. 2 bU jjJoC-i^H^r
CO«O* CT -^ ^ fl *J « « U ^f-4-^jQw.wA
HP»I(^ irt »^«i -^P,I ^*m
i a a >* >, x
CO 4) «
* X -«.-<-
* * *
-92-
-------
A simple computerized screening model for multiple point sources is the
EPA UNAMAP model PTMTP. With PTMTP, the user can define the receptor points,
input the meteorological data of interest, and specify the time periods to be
modeled. Alternatively, if area and line sources are expected to be
important, the EPA point, area and line (PAL) source model can be used for
screening modeling analyses. Unfortunately, neither PTMTP nor PAL can account
for particle deposition, a serious deficiency for analyses of monitors
significantly impacted by fugitive dust sources.
If sufficient computer resources (core space and data storage) are
available, the EPA MPTER and RAM models offer some additional valuable
capabilities for use in screening modeling analyses. MPTER is designed for
rural point sources only. RAM should be used for urban point and area
sources. These models are designed for use in more comprehensive dispersion
modeling analyses, but, with appropriate input data, MPTER and RAM can be used
in a screening analysis mode. MPTER and RAM can be used to calculate particle
deposition, but only through the use of an exponential decay term.
If the computer resources and required input data are available, the EPA
ISCST model is probably the best available for use in screening modeling
analyses in relation to receptor modeling applications. In addition to the
capabilities of the models discussed above, ISCST can predict contributions
from point, area, and volume sources and directly account for particle
settling by particle size. Like MPTER and RAM, ISCST is designed for use in
refined dispersion modeling analyses, but can be operated in a screening
analysis mode when appropriate data are used as input.
All of the aforementioned models are short-term models designed to use
hourly meteorological data to calculate hourly impacts which are then averaged
or otherwise applied to 24-hour periods. As an alternative, for screening
dispersion modeling analysis purposes, 24-hour impacts can be determined using
-93-
-------
models designed to calculate long-term (i.e., monthly, seasonal, or annual)
average impacts. For such analyses, artificial or hypothetical meteorological
input data must be constructed to simulate conditions that occur over 24-hour
periods. Of the models of this type shown in Table 4-4, only AQDM, CDMQC, and
ISCLT can be used to predict source contributions at discrete user-defined
receptor points. Only ISCLT accounts for particle deposition.
None of these models is specifically designed for screening analyses for
the 24-hour impacts of multiple sources. Screening dispersion modeling
analyses can proceed at different levels of detail. The choice of the level
of detail employed and the model selected for a given situation will depend on
a number of factors including 1) the objective of the analysis, 2) model and
input data availability and compatibility and 3) the desired precision of
results. Simplifying assumptions must be made in screening modeling
analyses. The initial simplifying assumptions are those contained within the
computational algorithms of the various models. Additional simplifying
assumptions are made in 1} the choice of a model, 2} the selection of model
options and 3) the treatment of model input data. Thus, screening modeling
analyses inherently involve making trade-offs between the reliability and the
efficiency of the analysis.
The following screening dispersion modeling approaches are suggested to
meet the objective of identifying insignificant sources. These approaches are
intended only as starting points for other ideas since many workable
approaches are possible. For a given analysis, the specific procedures used
must be tailored to the individual situation encountered.
Insignificant point sources can be identified with reasonable reliability
and efficiency using the PTDIS model. PTDIS calculates 1-hour impacts based
-94-
-------
on a user-specified array of wind speed and atmospheric stability
combinations, regardless of wind direction. An array of applicable wind
speed/stability combinations is provided in Table 4-5. With PTDIS, each
source must be modeled separately with the source-to-receptor (i.e.,
source-to-monitoring site) distance specified. To ensure that no potential
impacts are overlooked, a few closer and more distant receptors should also be
modeled. Each source should be modeled using simplifying assumptions that
maximize the modeled impacts. Thus, point sources are modeled at maximum
emission rates, assuming no particle deposition. Sources with insignificant
24-hour impacts are identified as those that PTDIS shows have insignificant
1-hour impacts. Since the modeled contributions of all the modeled sources
will tend to be overpredicted, no potentially significant sources will be
overlooked.
TABLE 4-5
APPROPRIATE COMBINATIONS OF WIND SPEED AND STABILITY
FOR USE IN PTDIS SCREENING MODELING ANALYSES
Pasquill-Gifford
Atmospheric
Stability
Classifications Wind Speeds (m/s)
A 1 to 3
B 1 to 5
C 1 to 5, 7, 9, 12, 15
D 1 to 5, 7, 9, 12, 15, 20
E 1 to 5
F 1 to 3
-95-
-------
Insignificant area, volume, (i.e., elevated area) and line sources can be
identified using the following two equations. In order to use these
equations, all area, volume and line sources must be treated as square area
sources. The first thing to be determined is whether the monitor in question
is located within or "near" the area source. For the purposes of this
screening analysis, "near" is defined as within a factor of 0.6 Ax + 100
meters of the center of the area source, where Ax is the length of the side
of the area source. Equation 1 applies to monitors located within 0.6 Ax +
100 meters of the area source center and equation 2 applies to more distant
monitors
Qj = Ci (Ax)l'7s » u Equation 1
18 10"
Qi = Ci (Ax)1 '* u ._ Equation 2
18 10s
where:
Qi = an emission rate in g/3
Ci = an impact in ug/m3
Ax = the length of the area source side in m
u = a wind speed in m/s
Equation 1 was obtained from the U.S. EPA Guidelines for Air Quality
Maintenance Planning and Analysis Volume 10R (1977a) and equation 2 was
derived empirically using the ISCST model.
"If a 1-hour concentration of 10 ug/m3, (roughly equivalent to 24-hour
concentrations of 5 u/m3, see Volume 10 above ) is assumed to be
insignificant sources as those with a Q/(Ax)l'7s ratio of less than 5.56 x
10"7 and equation 2 identifies insignificant sources as those with a
Q/(Ax)l's ratio of less than 5.56 x 10~s. Since the above procedures
are designed to overpredict contributions, sources whose impacts are deemed
insignificant can be eliminated from further consideration.
-96-
-------
4.3.4.3 Qualitative and Quantitative Receptor Modeling Analyses
Very often additional measurements taken on filter samples may be
available. Several State and local agencies perform routine chemical analysis
of selected filters for certain chemical species. Others submit a certain
number to microscopic analyses. As mentioned in Section 3.4, EPA has been
collecting size and chemically resolved samples at IPN sites in many major
U.S. cities (Watson, et al., 1981). These data can be used in additional
Level I analyses to verify the conclusions drawn from previous analyses. The
data quality issues discussed in Section 3.4 should be taken into
consideration when these data are analyzed.
Time series plots, pollution roses, and spatial mapping of chemical
species such as lead and bromine (which are almost always indicative of auto
exhaust contributions) and aluminum and silicon (which are usually contributed
by soil dust) can reveal source contributions which might not be evident when
only mass measurements are used.
Rudimentary mass balances can be performed to determine the relative
contributions of various source categories. For example, assume that
approximately 8 percent of soil is aluminum and that 8 percent of the
aggregate auto exhaust emissions are lead. By dividing aluminum and lead
concentrations by 0.08, one achieves a rough estimate of the soil and auto
exhaust contributions to a mass concentration. One could choose a lower limit
of these compositions, e.g. 5 percent, and could achieve a very conservative
approximation of these sources' contributions. If they are still minor
contributors, they could be removed from further consideration for control. A
review of the source compositions in Appendix B for sources suspected as a
result of analyses performed on total particulate matter data would allow this
procedure to be applied to a number of sources, if their key chemical species
were measured.
-97-
-------
Microscopic analyses are most informative when three conditions are
satisfied: spatial and time series plots should show individual sites or "hot
spots" with high concentrations; pollution roses should show a directional
tendency; and alignment of the monitor(s) and the major sources in the
emissions inventory should show correspondence with the prevailing wind
direction. Though it requires a skilled microscopist to perform
semi-quantitative analyses of filter samples, a qualitative examination of the
majority of the particles observed on an ambient filter, in comparison to
those observed on sieved samples taken from suspected sources, can be
revealing. Samples can be taken, dried, sieved and mounted on slides in 1.513
index of refraction immersion oil for examination. Several portions of a
hi-vol filter can also be immersed in an oil of 1.518 index of refraction
(this makes the filter disappear under transmitted light). If the filter
deposit is dominated by the source material, it is probably a major
contributor. If absolutely none is found, then other sources must be sought.
If the deposit shows a mixture of materials, more complete microscopic
analysis by an expert will be required.
Probably the most important step of a Level I assessment is the evaluation
of the extent to which the existing data support a control action. In many
cases, particularly those of an exceedance of standards at a solitary site,
the conclusions drawn from the varied uses of existing data described here
will be sufficient to identify major contributing sources and devise a.
strategy to control them.
Even if this is not possible, the Level I examination will result in
conclusions about which existing sources are not major contributors, thereby
narrowing the list of those which must be dealt with in greater detail in a.
Level II strategy.
-98-
-------
4.3.5 Resources
Assuming that existing data have been acquired and validated (a task which
can take from one man-week to one man-year, depending on the specific
situation) the analyses described here should not result in more than several
man-months of effort. Computational capability requirements are minimal,
though a hand calculator or personal computer with spreadsheet and graphics
software would greatly facilitate the work. The analyses discussed here can
be performed with little or no need for additional monitoring and with a
minimal number of laboratory analyses. The cost of studies of this magnitude
should range from $5,000 to $25,000.
4.4 Level II: Acquire More Data without Extensive Sampling and Use More of
the Capabilities in Refined Models
The cause and effect relationships between source emissions and ambient
concentrations can often be deduced using analyses performed at the level of
sophistication described in the preceding section (Level I). However, while
the answers might be obvious as to which sources are primarily responsible for
a given air pollution problem, a quantitative apportionment of that
responsibility may be impossible or may lack a reasonable degree of certainty
and thus be insufficient to justify the formulation and implementation of
expensive emission control strategies. If this proves to be the case, a more
rigorous study, of the type discussed in this section (Level II), can be
performed.
4.4.1 Source Data
The emission inventories for the sources that are considered to be the
most likely contributors to the problem (on the basis of the Level I analyses)
should be checked and verified or corrected. This process could start with
-99-
-------
the 5 to 10 most likely contributors and, if need be, proceed to other sources
based on. a prioritization scheme starting with the largest and nearest
sources. Parameters of concern include material throughputs, control
efficiencies, stack heights, stack diameters, exhaust gas flow rates, exhaust
gas temperatures, etc. Emission rates can be checked by comparing inventoried
values to new values calculated using the best available current emission
factors. If emission, rates vary greatly from day to day, hour to hour or over
other relevant time periods, it would be valuable to calculate such short-term
emission rates for inclusion in the inventory.
Sources such as road dust, storage piles, open burning, playground dust,
etc. are often important contributors to particulate matter problems. If
these types of emissions were not evaluated in the Level I analyses, or if
they were included on a broad spatial scale only, then it would now be
advisable to compile these emissions using microinventory techniques around
each sampling site. Comprehensive microinventories may not be required in all
cases, but they are usually of value, especially at receptor sites suspected
of being influenced by nearby sources of fugitive dust. The procedures for
developing microinventories are discussed in Section 3.2. Daily emission
inventories may be needed if emission rates vary greatly from day to day, but
they can be quite difficult to obtain.
Most existing inventories do not contain data on particle size or chemical
constituents. Therefore, for a Level II analysis, it may be necessary to
obtain source emission samples for physical and chemical analyses using
inexpensive grab sampling techniques. Chow, at al., <1981) describe the
following methodology for taking these samples:
"Several common source types are amenable to "grab sampling",
resuspension sampling and subsequent chemical analysis. These source
types include: road dust, road salt, soil, storage pile contents,
loading and unloading materials, and baghouse and cyclone residue.
-100-
-------
A representative sample of the material can be collected in a
ziplock bag. In the laboratory it can be dried, sieved through a
Tyler 400 mesh sieve (38 urn screen size) and resuspended for
sampling with devices similar to those used to take ambient samples.
The chemical analyses performed on these resuspensions should be the
same as those of the ambient samples."
4.4.2 Meteorological Data
The desire to obtain better quantification of source contributions to
ambient concentrations may result in the need for additional, or more
representative, meteorological data than were used in the Level I analyses.
In Level I, dispersion modeling analyses can be performed using representative
hypothetical meteorological data (e.g. Table 4-5), and receptor modeling
analyses can be performed using data obtained from published or other readily
available historical records. In a Level II study, it may be necessary, or at
least advisable, to construct a computerized data base of meteorological
information for multiple years, parameters, and monitoring sites. Such a
computerized data base could be used to calculate surface and upper air wind
trajectories based on data obtained from multiple NWS or other monitoring
sites.
The more refined dispersion models can use either hypothetical or actual
meteorological data. The refined short-term models can calculate hourly,
daily, or other short-term average impacts using up to a full year of hourly
meteorological data. Such a large data set can be used efficiently only if it
is computerized. If dispersion modeling is going to be performed on a large
number of sampling days of data, then it is advisable to obtain computerized
data sets of the hourly meteorological data. For NWS stations, such data sets
can be obtained from the NCDC in the format required for input to the CRSTER
preprocessor.
-101-
-------
The representativeness of the meteorological data is of considerable
importance to the Level II study. Proximity and similarity of location are
the primary concerns as discussed in Section 3.3. The meteorological
station(s) should be located reasonably close to the sources and sampling
sites and in an area of similar topography. If no NWS stations meet these
criteria, it may be possible to obtain wind data from a nearby non-NWS
location and then merge these data with data for parameters other than wind
obtained at the NWS site. Uncertainties should be assigned to each of the
parameters measured. If the available meteorological data are deemed
unrepresentative, a Level III study will be required to gather appropriate
data.
4.4.3 Ambient Data
Level II analyses become necessary when the results of Level I analyses
are inconclusive or incomplete. This will often occur because of a lack of
existing data on the composition of the ambient particulate matter. It is
often possible to obtain such data by analyzing the material on existing
filters or by collecting a limited number of new filters. A suggested
procedure for selecting filters for subsequent analysis (for particle size,
chemical constituents, and other properties) follows.
For an "ideal" receptor modeling study, a large number of the available
filters from a monitoring site would be subject to analysis and subsequent
receptor model evaluation. Such an approach is not feasible when resources
are limited and only a small subset of the available filters can be analyzed.
It becomes necessary, therefore, to select filters which are representative of
a wide range of conditions. When short-term particulate matter concentrations
are of concern, the selection of filters for analysis can be limited to some
-102-
-------
of the days of high concentrations plus a few clean day filters for
comparison. When long-term particulate matter concentrations are the problem
a broader selection of filters is warranted. If both the long- and short-term
time frames are of concern, a suggested approach is to select filters that,
taken together, are reasonably representative of annual average conditions but
that also include conditions where maximum short-term concentrations and
source impacts occur.
The following criteria are suggested for use in the selection of the
ambient sampling day filters to be analyzed:
Select filters from day(s) with widespread high particulate
matter concentrations (episode days). By selecting the date(s)
on which most of the monitors indicated high concentrations, the
primary causes of the particulate matter problem can be
identified and studied.
Select filters from day(s) with light and variable winds.
Filters from such days would represent a homogeneous mix of
contributions from various sources.
Select filters from day(s) with "average" meteorological
conditions. Filters from day(s) which have wind speeds and
stabilities close to the average for the year are selected to be
representative of days with typical contributions from each
source.
Select filters from day(s) with high wind speeds. Filters
analyzed for such days will indicate the extent to which certain
resuspended material and fugitive emissions affect the monitors.
(Useful if windblown dust is suspected as a major source.)
Select filters from day(s) with low particulate matter
concentrations. Filters from those days having low particulate
matter concentrations (non-precipitation days) are selected to be
representative of days with minimal impacts from each source.
(Useful if annual average concentrations are of interest.)
Select filters from day(s) with persistent winds coming from the
direction of sources previously identified as suspected
contributors. Filters from such days can be used in direct
comparisons to dispersion model results to evaluate the accuracy
and representativeness of source emission rates and other source
configuration parameters.
-103-
-------
It is also advisable to analyze several typical "background" filters from
monitors outside of the study area in order to obtain data on the composition
of material coming into the study area.
If additional ambient sampling is necessary, important technical
considerations include sampling locations and periods, sampler type, particle
size, filter media and the sample preparation requirements of the particle
analysis instruments (Gordon, et al., 1983). For a Level II analysis only a
few samplers should be needed at locations where existing particle size and
composition data are inadequate. Such sampling would probably best be
performed with dichotomous samplers to obtain fine and coarse fractions on
Teflon or Millipore filters to facilitate chemical and coarse microscopic
analyses. Quartz filters could be used if sampling for carbon and ions is
needed. The results from such sampling could then be used to judge the
accuracy of existing particle composition data obtained from hi-vol sampling.
4.4.4 Procedures (Level II)
4.4.4.1 Receptor Modeling Analyses
Some idea of the most likely contributors to each receptor should have
resulted from the Level I efforts and this can be used as guidance in choosing
the further analyses of receptor samples. For these sources, the source
characterization literature in Appendix B can be surveyed to 1) determine
which chemical and physical properties will identify and help to quantify
these sources, and 2) decide which analysis methods are appropriate for the
existing samples.
Not all analyses can be performed on all filter media. For example, x-ray
fluorescence (XRF) is inappropriate for samples taken on fiber filters because
of the x-ray absorption in this thick filter media. XRF is very good for
Teflon and other thin pure membrane filters. The blank levels of the filters
-104-
-------
must be much less than the concentrations of the species being measured.
These levels can vary from batch to batch of filters, and if blank filters
from the sampling batches have not been saved, erroneous results may be
obtained. Many glass fiber filters contain organic binders which will hinder
any analysis for carbonaceous species.
Watson, et al., (1981) discuss the effects of sulfate and nitrate artifact
formation on filters and show how a change of filter media during a sampling
program can yield an artificial increase in both total particulate matter and
sulfate levels which might be interpreted as a sudden change in source
contributions. The data interpretation efforts in Level II cannot be
separated from the measurement process, and each additional analysis of
existing samples must be thoroughly validated.
An important aspect of receptor modeling is the use and propagation of
measurement precision to the model results. These precisions must be acquired
with any additional analyses through routine replicate and blank analyses.
Watson, et al., (1983) describe methods of using these analyses to obtain the
precision of analysis results.
To quantify source contributions, a chemical mass balance calculation may
be the most straightforward approach. This requires a source composition
matrix in addition to ambient component concentrations.
If more than 50 receptor samples with chemical characterization
measurements are available, then the multiple linear regression or factor
analysis models can be used to narrow down the number of contributing sources
and possibly to calculate source compositions.
A standard factor analysis can be performed using many standard
statistical packages such as Statistical Package for the Social Sciences
(SPSS), Biomedical Data Processing Programs (BMDP), and Statistical
-105-
-------
Applications Systems (SAS). A standardized correlation matrix of all
variables and a varimax rotation is the most commonly used configuration. The
number of dominant factors should correspond to the number of dominant
sources, and the chemical species with which these factors are most highly
correlated can be used with existing source composition information drawn from
Appendix B to associate each factor with a source type.
If the source mix is such that fairly unique tracers can be identified
with different sources, factor analysis will confirm this. Such a tracer will
be highly correlated with one factor but not correlated with any other
significant factor (Kleinman, 1977). The multiple linear regression (MLR)
model can be applied using tracer species. The coefficients represent the
inverse of the tracer concentration in the source emissions, and the product
of each coefficient and the receptor concentration of the corresponding tracer
element yields the contribution of that source to the total mass concentration
measured at the receptor. If the "effective variance" weighting (Watson, et
al., 1983) is used to perform the MLR least squares fit, then each measurement
will be weighted in inverse proportion to its uncertainty and a realistic
error bound will appear in the result.
If it appears that several sources are contributing substantial quantities
of the same chemical species, then a Target Transformation Factor Analysis
(TTPA) can be used to derive a source composition matrix. Hopke, et al.,
(1983) have -created the FANTASIA computer program which will perform this
analysis and the subsequent mass balances to calculate source contributions.
It relies on correlations about the origin and a target rotation.
If fewer than 50 samples are available, the statistical significance of
the multivariate models is in doubt. A mass balance calculation can be used,
but this requires a source composition matrix assembled from specific source
-106-
-------
tests or literature values. If reentrained dust or fugitive emissions are
suspected, then grab samples may be taken and analyzed as discussed
previously. Since the variability of source compositions contribute to the
overall uncertainty of the mass balance results, replicate analyses should be
performed on samples of materials collected at several emission points within
the same source so that average source compositions and their .standard
deviations can be obtained. Compositions for sources which cannot be sampled
with Level II resources (e.g. smokestacks) must be obtained from literature
such as is cited in Appendix B.
The mass balances are iterative, and several combinations of sources can
provide equally valid fits to the receptor chemical concentration data
(Watson, 1979). Judgement must be used to obtain a physically significant
solution when using existing computer models.
The optical and scanning electron microscopy models combine the analysis
and modeling together. Since neither the analysis nor the classification
schemes in use are accepted by all researchers, analysis discrepancies
described in Section 3 are often compounded by interpretation discrepancies,
making the results semi-quantitative. Nonetheless, microscopy can provide
conclusive source contribution results in certain situations. The cost of
analysis per sample is also five to ten times the cost of the bulk chemical
analysis used for input to the factor analysis, MLR and mass balance.
Microscopic analyses and models may be used in Level II to 1) confirm the
conclusions drawn on a sub-set of samples by one of the other models and 2) to
resolve anomalies found by these models. Mote, however, that not all fine
particles can be readily identified with microscopic analyses.
-107-
-------
4.4.4.2 Dispersion Modeling Analyses
The dispersion modeling analyses in a Level II study are more detailed and
generally require more input data than the Level I analyses. Since it is
important to be able to determine source/receptor relationships on both a
short-term (e.g., daily) and long-term (e.g., annual) basis, techniques for
modeling both time periods are discussed in the following paragraphs.
Procedures for modeling daily impacts are discussed first.
The primary objective of dispersion modeling is the reliable
quantification of source contributions. The screening dispersion modeling
approaches discussed in Level I provide a reasonably reliable and efficient
means to eliminate insignificant sources, but the simplifying assumptions
employed can completely obscure the relative magnitude of the contributing
source impacts. More reliable source contribution estimates can be obtained
if more detailed dispersion modeling analyses are performed.
In the following paragraphs, two different approaches are suggested for
the more detailed short-term modeling analyses. The first approach can be
used to determine the relative magnitude of source impacts for 24-hour periods
in general. The second can be used to do the same thing for specific 24-hour
periods on which particulate matter sampling has already been conducted. The
first approach uses hypothetical meteorological data that are representative
of conditions that produce high concentrations of particulate matter. The
second approach uses meteorological data actually measured on the specific
days when the particulate matter sampling was conducted. As a first step,
both approaches make use of the results from the Level I screening dispersion
modeling analyses to eliminate insignificant sources. Because of its superior
capabilities, the best model for both approaches is ISCST, although PTMTP,
PAL, MFTER, and RAM could also be used. All of these models provide
individual source contribution listings for each receptor.
-108-
-------
In the first approach, the combined 24-hour impacts of multiple sources
are obtained using a methodology patterned after the treatment of
meteorological data in the PTDIS model. With PTDIS, 1-hour impacts (of a
single source) are calculated using an array of combinations of wind speed and
atmospheric stability. Likewise, the combined 24-hour impacts of multiple
sources can be determined using a representative array of combinations of wind
directions, wind speeds, and stabilities.
Each combination of wind direction, wind speed, and stability is input to
the model to obtain a 1-hour average prediction. The impacts calculated for
1-hour periods are assumed to be representative of 24-hour periods. This
simplifying assumption is reasonable because the monitoring site will receive
the largest 24-hour impact from a given source if the wind direction is
persistent. However, daily impacts will be overpredicted, primarily because
wind direction variability is greater than is assumed here.
The large number of wind speed/stability combinations in PTDIS must be
reduced to those combinations likely to persist for extended portions of a
24-hour period. Representative combinations are shown in Table 4-6. When
some of these wind speed/stability combinations are used in the modeling it is
necessary to consider the variability in atmospheric stability during 24-hour
periods. Neutral ("D") stability conditions occur most often and can last
24-hours. Unstable ("A", "B" and "C"> conditions occur only in the daytime
and "Stable ("E" and "F") conditions occur only at night. Very unstable "A"
conditions are likely to persist for only a few hours during the day.
Table 4-6 and the Level I screening modeling results may be used to select
the wind speed/stability combinations for the more detailed modeling
analyses. The PTDIS point source modeling results and the area, volume and
line source modeling results from Level I are used to identify the 5 to 10
-109-
-------
TABLE 4-6
COMBINATIONS OF WIND SPEED AND STABILITY THAT ARE LIKELY
TO PERSIST FOR EXTENDED PORTIONS OF 24-HOUR PERIODS
tall
stack
sources
Wind
Speed
(m/s)
2
4
7
10
Pasquill Atmospheric Stability Classifications
Daytime Only Nighttime Only
B C D E F
/ V V / V
/ / /
Rural areas
only
highly dependent upon the source-receptor alignment relative to the input wind
direction. A small difference (10°) in wind direction can produce an
entirely different picture of the magnitude of the source impact at a given
receptor.
There are two ways to ensure that the effects of wind direction are
adequately considered. Either all compass wind directions (in 10°
increments) can be modeled or only critical wind directions can be identified
and modeled. The critical wind directions will be those associated with the
sources that produce large impacts at the monitoring site. As mentioned
previously, the Level I screening modeling results may be used to identify the
5 to 10 sources that have the largest impacts. In many cases, important
sources will be found to be grouped together in one or two geographic
-110-
-------
sources that have the largest impacts at a monitoring site. The wind
speed/stability combinations selected for further modeling are taken from
among those that produced the highest impacts from those 5 to 10 sources.
Table 4-6 also highlights a few factors that can be used to guide the
selection of the most important wind speed/stability combinations.
Of all the meteorological parameters, modeled impacts at a given location
are most sensitive to wind direction. The relative impact of each source is
sectors. In that event, the wind directions can be input to the model in
10° increments from those sectors. Alternatively, the exact source/receptor
alignment wind directions for selected sources can be input to the model.
Maximum impacts can be estimated using just a few wind direction/wind
speed/stability combinations in the modeling. For the critical wind
directions identified above, four combinations of wind speed and stability are
suggested:
low wind speed/unstable (B or C)
low wind speed/neutral
-------
The second approach to the more detailed Level II dispersion, modeling
analysis depends on meteorological data recorded on specific days when
particulate matter sampling has also been conducted. This approach is
preferable if comparisons between measured and dispersion-modeled pollutant
concentrations are anticipated. If the meteorological data are available,
this approach has the advantage that the selection of meteorological input is
simple. The meteorological data used in the model must adequately represent
the actual meteorological conditions in the study area. As discussed earlier,
modeled impacts are very sensitive to wind direction. For this reason alone,
when modeling only involves a single sampling day and a single receptor, there
is little likelihood of agreement between measured and modeled
concentrations. The use of additional receptors in close proximity to the
monitoring site will illustrate the sensitivity of predicted values to wind
direction.
As mentioned earlier, it is often important to be able to determine
source/receptor relationships on an annual average basis. A number of
dispersion models are available for this purpose as shown in Table 4-4. The
available models use either an annual stability wind rose or a full year of
hour-by-hour meteorological data to calculate annual average source impacts.
Procedures for using these models with a full year of data are contained in
the respective model user's guides.
4.4.4.3 An Approach for Comparing Dispersion and Receptor Model Results
Comparisons between receptor and dispersion modeling results can provide
important insights into source-receptor relationships. The two analysis
approaches use different input information and incorporate different
assumptions. The uncertainties associated with each approach can be
-112-
-------
substantial, but the similarities and differences between results from the two
methods can often be used to identify erroneous or inadequate data and to
achieve more reliable dispersion model predictions.
At the outset, the key to making comparisons between source (dispersion)
and receptor modeling is to obtain source model results for the same
monitoring periods when ambient measurements were collected. The sampling
days selected for the analysis should represent a variety of dispersion
conditions. The uncertainties associated with results from a single day are
quite large; comparisons of results for a larger number of individual days (10
or 15 is recommended as a minimum) are more meaningful. The dispersion model
used in these analyses must provide impact predictions for individual
sources. The ISC models (ISCST and ISCLT) are best suited for this purpose
because, in addition to the features described earlier, the ISC models will
report impacts for selected groups of sources as well as for individual
sources. The dispersion and receptor modeling results can then be compared on
a source-by-source (or source group-by-source group) basis. Source groups are
defined by the receptor modeling results, since receptor modeling does not
distinguish among sources with similar emission compositions.
In the event that the receptor model results agree with the dispersion
model results on a source-by-source basis, the receptor model analyses would
constitute convincing evidence for the validity of the dispersion modeling
analyses and vice-versa, since each will have been conducted independently.
(As discussed previously, receptor model source contribution results should be
established using two or more receptor models.) It is more likely that the
source culpability information provided by the receptor model will differ
considerably from that provided by the dispersion model, particularly for
-113-
-------
fugitive dust sources. In the latter case, validated receptor model results
can be used to direct a re-examination of the dispersion model inputs to
determine where inadequacies may exist.
The process of modifying the dispersion model or model inputs, based on
comparisons between dispersion model predictions, observed ambient
concentrations, and receptor modeling results, is primarily intended to
identify and correct errors or shortcomings. The purpose of this exercise is
to obtain more reliable, technically supportable dispersion model predictions,
not to create the illusion of agreement between two problem-solving
approaches. If reasonable results can be obtained through dispersion
modeling, then a powerful analysis tool is available for estimating air
quality at locations other than monitors and for assessing control strategy
options.
The challenge is to accomplish this end in a systematic fashion, and to
minimize the use of subjective, trial-and-error techniques. One effective
device for controlling this process is the use of a "comparison protocol",
which defines the stepwise analysis procedures to be followed once dispersion
model and receptor model results are obtained. The protocol, prepared before
modeling is performed, anticipates areas of possible disagreement between the
two modeling approaches, identifies the model inputs and model features which
could produce such disagreement, and outlines the decision criteria for
revising the inputs or the model. Examples of the procedures such a protocol
might contain are discussed below. The concept of a model comparison protocol
is defined in considerable detail in the EPA document Interim Procedures for
Evaluating Air Quality Models (U.S. EPA, 1981c). This document describes
procedures for comparing the performance of two dispersion models and is not
entirely applicable to the present topic.
-114-
-------
There are a number of factors which cause dispersion model predictions to
disagree with measured particulate matter concentrations. Some of these
problems are source-specific, primarily affecting individual source-to-monitor
impacts, while others are more likely to produce systematic or widespread bias
in the model predictions. Examples of the former include:
Erroneous emission rates caused by such items as the omission of
unknown sources from the modeling or the use of inaccurate
throughput information.
The use of inappropriate emission rates, such as the use of total
particulate matter emission rates which include particle sizes
that settle out prior to impacting the monitoring site.
Incorrect information concerning daily source operating
parameters. For example, while a source may operate at 45 percent
capacity on an annual basis, its actual mode of operation may be
at 90 percent capacity for 50 percent of the days in a year.
* Neglect or incorrect consideration of downwash from tall stack
sources.
Neglect or incorrect consideration of resuspended particles.
Building interference causing source-to-receptor (i.e., source-to-
monitor) geometry to be incompatible with Gaussian dispersion
assumptions.
Local meteorology differing from that modeled. A typical problem
is wind direction shift or channeling caused by buildings or
topographic features.
Examples of factors that are not addressed by the dispersion model and
that may cause systematic biases in model results include:
Heat island effects which cause the actual near-source dispersion
of elevated emissions to be greater than that modeled.
Sea/land breeze effects.
Effects caused by the fumigation of tall stack emissions.
Effects caused by the development of thermal internal boundary
layers (TIBL) over areas with varying surface heating
characteristics.
-115-
-------
There are also potential systematic and random biases in receptor models.
Causes of such biases include:
The selection of ambient sampler locations and filter samples for
analysis such that the data studied are not spatially or
temporally representative of the real air quality.
Incomplete sampling schedules such that anomalous events are
disproportionately represented in the data base. For example,
samples may have been collected during periods of emission control
equipment failure, or, in contrast, during periods when important
sources were not operating. Conversely, a lack of such samples
would also bias the data base.
Incomplete or inaccurate information on source emission
compositions, or the existence of sources with similar emission
compositions such that the contribution of one source cannot be
distinguished from that of another.
Emission transformations where the amount or composition of
emissions changes between the source and the receptor. Examples
include particle deposition rates that differ among the chemical
components, chemical reactions in the atmosphere, and the
reentrainment of emissions that have already been deposited.
Ambient sampling or filter analysis deficiencies where the data
collected do not accurately reflect what is in the air. Examples
include disparate collection efficiencies of the various filter
media, artifact formation on filters, and analysis techniques
that only see large particles, particles on certain parts of the
filters or certain types of particles.
Unrepresentative meteorological data. Examples include
misoriented pollution roses and upper air trajectory analyses with
directional and distance errors.
A three phase process is suggested to determine the need for making any
changes in the dispersion model or input parameters. The first phase involves
taking steps to eliminate obvious dispersion model input errors; the second
involves modifying the dispersion model based on receptor model results for
those sources whose input parameters are most certain; and the third involves
the systematic comparison of predicted with measured concentrations to
document model performance.
-116-
-------
4.4.4.3.1 Phase 1. Refine Emissions Inventory. When large discrepancies
are found between observed ambient concentrations and dispersion model
predictions, emissions data are the first item to investigate. Receptor
modeling results will indicate the contributions from different sources to
observed concentrations. Comparison of source contributions predicted by
receptor and dispersion model analyses will serve to focus attention on
individual sources or source groups. Additional test data or more detailed
engineering information can then provide a basis for inventory revisions.
Phase 1 can be accomplished through the following stepwise process:
Identify "significant" sources or source groups from receptor
model results. Sources which account for only a small fraction of
observed ambient concentrations should not be investigated further.
Assign a "level of confidence" to emission parameter estimates for
each significant source or source group, based on engineering
estimates and on receptor model results. Table 4-7 illustrates
the confidence level information needed from this analysis.
Obtain dispersion model results. Compare source contributions
predicted by dispersion and receptor models. Identify any large,
systematic discrepancies between source contributions from the two
models.
Re-examine the emissions data for sources with large
discrepancies. Revise the inventory through additional source
testing or improved engineering estimates, where confidence in
existing data is low.
Recognizing that substantial uncertainty in particulate emission
parameters and dispersion model predictions is unavoidable for most urban or
industrialized regions, this Phase 1 effort is designed to focus attention on
significant sources and on obvious errors. Analyses should be based on
multi-day modeling results, rather than an individual day, since dispersion
model predictions for short-term averaging periods can be very sensitive to
small errors in plume transport direction.
-117-
-------
TABLE 4-7
ASSESSMENT OF CONFIDENCE IN DISPERSION MODEL INPUT PARAMETERS
Source Number/Type/Name
Source-Specific Errors
Emission rate 3 1 etc.
Particle size
characterization 2 1
Source locations/dimensions 3 2
Source operation parameters 1 3
Downwash 3 1
Building interference 1
Wind direction 2
Other meteorology 3
Other
Key:
0 = Least confidence (in parameter or effect as modeled)
3 = Greatest confidence (in parameter or effect as modeled)
-113-
-------
Confidence levels for emission parameters should be determined prior to
dispersion modeling, to minimize the bias introduced by knowing in advance how
well the dispersion model performs for any given source. Some receptor
modeling techniques also provide quantitative estimates of the uncertainty
associated with calculated source contributions. Model performance evaluation
studies for dispersion models have shown that uncertainties of 20 to 30
percent are common for long-term average concentrations predicted at a given
monitor location (Londergan, et al., 1983). Efforts to identify serious
errors in the emission inventory should focus on sources with discrepancies of
at least a factor of 2 between receptor and dispersion model predictions.
The pattern of discrepancies at different receptors may also aid in
identifying potential inventory errors. For example, large overproduction by
the dispersion model at a receptor very close to the source would suggest
possible errors in source geometry; large overprediction at receptors far from
a source would suggest an erroneous particle size distribution. Experience
with dispersion models and emission inventories will play an important role in
diagnosing potential inventory problems.
The protocol will serve as a check on the subjective process described
above. The knowledge that a discrepancy can be "fixed" by increasing the
emissions of a specific source is not sufficient grounds for revising the
inventory. Technical justification will be required before changes are made.
Through this process, however, efforts can be focused on the relatively few
sources which are critical to dispersion model performance.
4.4.4.3.2. Phase 2. Further Dispersion Model Modifications. After the
inventory revisions have been incorporated, revised dispersion model
predictions will be obtained. The revised source contributions predicted by
-119-
-------
the dispersion model are again compared with receptor model results, to
determine whether further modifications to the dispersion model or
meteorological inputs are warranted. For this comparison, attention is
focused on sources with the highest confidence rating for emission parameters
and receptor modeling results. Any large, systematic discrepancies between
dispersion and receptor model predictions for these sources would indicate
shortcomings in either the receptor or the dispersion model.
The protocol will identify, in advance, the limitations inherent in the
dispersion modeling approach. Such limitations may include meteorological
inputs, assumptions concerning dispersion rates, particle deposition, effects
of surface roughness or terrain, effects of land/water interface, etc.
Remedial measures could include acquisition of additional meteorological data
(if available>, choice of different model options, or modification of
dispersion algorithms to suit local conditions.
The pattern of model performance at different receptor locations and for
different source types can provide useful information for diagnosing
dispersion model shortcomings. Once again, however, emphasis should be placed
upon large discrepancies between dispersion and receptor model results,
recognizing the uncertainties inherent in both modeling approaches. Clusters
of receptors, placed around each monitor, can be used to assess the
sensitivity of dispersion model predictions to small changes in plume
transport direction (and indirectly to indicate model sensitivity to one cause
of prediction uncertainty).
Model modifications should be undertaken in an effort to improve
performance by incorporating features which take into account local dispersion
conditions and source characteristics. Problem-specific model modifications
will often prove a cost-effective alternative to additional sampling and
analysis.
-120-
-------
4.4.4.3.3 Phase 3. Document Dispersion Model Performance. After
revisions to the emissions inventory and/or the dispersion model have been
implemented, it is important to document model performance using available
monitoring data. Before the dispersion model is used to determine control
requirements or to assess alternative strategies, the uncertainty associated
with model predictions should be understood. For this purpose, observed and
predicted ambient particulate matter concentrations should be compared, with
primary emphasis on the concentration values associated with air quality
standards (e.g., annual average and peak 24-hour concentrations). The
comparisons should not be restricted to those days selected for receptor model
analysis, but should include at least one year of ambient monitoring data.
The meteorological data used should correspond to the particulate matter
sampling schedule.
Detailed recommendations concerning model performance evaluation are
contained in the reports Judging Model Performance, from the 1980 American
Meteorological Society (AMS) Workshop (Pox, 1981), and Interim Procedures for
Evaluating Air Quality Models (U.S. EPA, 1981c). In the present context, a
limited analysis of model performance is envisioned, rather than the
comprehensive statistical evaluation described in these reports, but the basic
analysis approach is quite similar.
Results from the performance evaluation will provide a basis for deciding
whether to rely upon the dispersion model to formulate control strategies, or
whether to use monitoring data and receptor modeling results. If the
dispersion model is demonstrably reliable at the monitoring sites then model
predictions can be made for the entire study area and control strategies
should be developed based on the model results. If the dispersion model is
not proven reliable, and measured concentrations show nonattainment, then
-121-
-------
control strategies would be developed based on the measured data and receptor
modeling results. However, in this case, the issue of the air quality at
non-monitored locations must be addressed. If the monitoring sites are not
representative of the entire study area, based on dispersion model
predictions, then additional monitors may be needed at locations of high
predicted impact so that additional receptor modeling analyses can be used to
determine the source impacts at such sites.
4.4.5 Resources
Resources required for Level II analyses are significantly higher than
those required for Level I. At least a mini-computer is required to run
programs, not all of which are in the public domain at this time. Data bases
are large and require sufficient storage. The analyses described here should
not require substantial additional monitoring, but it will usually be
necessary to perform a variety of laboratory analyses on existing samples.
Analysis costs will range from $25,000 to $100,000. Four months to one
man-year of effort will be required.
4.5 Level III: New Sampling, Analysis, and Model Development
The frequent need to solve problems associated with atmospheric
pollutants, other than those materials collected routinely by monitoring
networks, requires the development of sampling and analysis strategies
specifically directed to the individual problem. In many cases, these
problems can be explored using receptor models or dispersion models and can
use readily available sampling and analysis methodologies. However, sometimes
it may be necessary to incorporate vastly different approaches into the study
design. A few examples would be programs to define the sources of 1) sulfate.
-122-
-------
2) organic particulate matter, and 3) the mass of material in particular size
fractions. For each of the above, a similar general approach can be used in
the design of the experiment; however, the details for each case require
considerable discussion.
The following is structured to establish general criteria for the
development of Level III studies.
4.5.1 Source Data
As has been described in other sections of this volume, and other volumes
of this series, a major concern is the availability and subsequent cataloging
of source emissions data for both receptor and dispersion models. In Level I
and II, the methodology for acquiring useful information from existing source
emissions data bases was described. However, these will probably be
inadequate to address many specific air pollution problems because all the
sources, or at least the major sources, in a particular study area may have
only been tested for a limited number of parameters. Furthermore, some data
bases were not designed to mimic the emission patterns that would more closely
represent the physical and chemical fractionations and transformations that
occur in the environment and are observed in receptor samples (for example,
sulfate and organic species). Such a deficiency is important when using a
receptor model to allocate the percent of mass associated with a source or
when" trying to validate the model with some contribution estimates from
appropriately parameterized dispersion models.
The first stage of Level III is to draw upon the previously acquired data
in Levels I and II and outline a strategy based upon the strengths and
weaknesses of available emissions inventories. At present, these data could
be augmented by conducting actual source tests similar to those normally used
-123-
-------
in the federal New Source Performance Standards (NSPS) or National Emission
Standards for Hazardous Air Pollutants (NESHAP) programs, but adding trace
element and other chemical analyses would be required in the development of
source compositions for input to receptor models. In most cases, the
particulate matter size fractionation would not have been attempted, and the
results would be useful only if the chemistry of the emissions does not vary
appreciably between size fractions.
In place of the above, the critical question of compatibility of the
source compositions with those perceived at the receptor could be addressed
using dilution sampling or platform sampling downwind of plumes. These
techniques could also estimate the mass emission rates under a variety of
operating conditions. A selection criteria for the sources to be tested
should be established based upon 1) the adequacy of cataloged profiles (with
concern for regional differences in sources), 2) the emission rates of
sources, 3) the proximity of the source to the receptor site, and 4) the
potential hazard associated with the source. Additional source testing
criteria are discussed in Section 2.2 of Volume II of this series (U.S. SPA,
1981b).
Dilution source testing has been identified as a method to help increase
the possibility of having the source composition more closely reflect that
sampled at a receptor. It involves cooling of the sampled effluent to a
temperature which is nearly equivalent to that found in the ambient
environment to simulate the conditions under which condensation, growth and
other reactions of the emitted species occur. A further distinct advantage is
the possibility for sampling the diluted effluent with techniques equivalent
to those used at the receptor. In the case of particulate matter, the net
result could be the development of source composition catalogs which would be
-124-
-------
size classified and/or be able to capture more volatile species. As mentioned
in previous volumes of this series, a present drawback is that this sampling
is still in the development stage. Therefore, a standardized sampling
methodology is not available as of this writing.
Another approach to the collection of emissions at conditions which are
more nearly equivalent to those anticipated at a receptor would be platform
plume sampling. In the past, this methodology has been used in dynamic
experiments (Lagrangian coordinate system) where the production and transport
of ozone and secondary particulate species were studied in both urban and
power plant plumes. However, plume sampling can be accomplished in an
Eulerian experiment (fixed coordinates) by measuring the plume at a fixed
point with a suspended platform (skyhook, etc.) near the receptor. Obviously,
because of the expense, major sources along the prevailing wind direction
should be of prime consideration and the sampling platform, fixed or
otherwise, must have collection devices which are nearly equivalent to those
used at the receptor. Preliminary or informational studies on the anticipated
emissions should be completed so that the plume sampling is well focused. As
stated previously, the sampling should be size selective (one or two stage)
for particulate matter in order to accommodate future regulatory requirements
and to differentiate between the fine and coarse fraction compositions.
In some types of studies, ground based plume studies may also be
warranted. This technique will be quite useful for receptor sites located
near major automobile and truck traffic arteries, farmlands spraying
pesticides, fugitive dust emissions, and field burning activities. Properly
outfitted vans or temporary trailer monitoring stations can be used to house
equipment.
-125-
-------
The source characterization used in Level III studies will be directed
toward achieving the most representative compositions of the sources that can
affect a receptor. It will combine information available from Level I and
Level II activities with new source tests. These latter tests will be
designed to be conducted at conditions more closely associated with the
ambient environment. To reduce costs, each should be conducted at times when
ambient samples are collected at the receptor and the receptor is downwind of
the source.
4.5.2 Meteorological Data
The Level I and II approaches to source or receptor modeling involve the
accumulation of readily available meteorological data from NWS, State and
local government or privately operated stations. In addition, synoptic or
mesoscale meteorological data are processed via computer programs to give
either upper air or surface wind trajectories. In many cases, this is all the
resolution that may be required, since these data are available near most
major urban centers. For multivariate receptor modeling techniques the same
resolution is sufficient such that with large numbers of samples (200-400), it
will be possible to categorize the apportionment of a given species according
to wind direction. The trajectory analyses will also prove useful in
attempting to determine the contributions of chemical or other species from
local versus regional sources.
In a Level III study it is possible that the meteorological detail will
need to be more extensive. This would be especially important in situations
where the influence of complex terrain, sea breezes or valley channeling will
distort the local wind and atmospheric conditions to such an extent that the
Level I and II meteorological data will be unreliable and comparisons of
-126-
-------
dispersion and receptor models will be difficult. In addition, there will be
situations where the available meteorological stations are too distant or, for
various reasons, are not sufficiently representative of the study site. It is
important to note that although the mass balance for particulate matter
requires as little as one sample to estimate the percent contribution from
individual sources, the support data requirements are significant. Depending
upon the location of the site, more samples will have to be taken for various
meteorological conditions to assess the frequency and magnitude of sources
influencing an area.
The Level III approach to meteorological data acquisition will require
location of a meteorological station either at or in close proximity to the
sampling location. For example, in the New Jersey ATEOS study (Lioy and
Daisey, 1983) the monitoring sites were within 10 miles of an airport NWS site
which eliminated the necessity for a local meteorological station. It should
provide, at a minimum, the types of meteorological data available on the local
meteorological data summary provided by the NWS from airport sites located in
the United States. However it would be advantageous to also have data on the
mixing height and the wind profiles at altitudes above the surface. While
there are various means of obtaining such data, it would be preferable to
install an acoustic sounder at a suitable location. This device is versatile
and will provide computerized output of meteorological information on a real
time basis (continuous) throughout the day. Indirect measurements made by
doppler type acoustic sounders include the wind speed, wind direction,
vertical motion, standard deviation of vertical motion, horizontal components
of turbulence, and echo strength (proportional to the temperature structure of
the atmosphere). In addition, with the use of appropriate algorithms, the
echo strength can be converted to mixing height.
-127-
-------
These devices can be operated for long durations, as long as there is
proper maintenance and periodic review of the data. The mixing height and
wind parameter data from the acoustic sounder will in fact provide a useful
way to estimate the burden of atmospheric pollutants of local origin versus
the amount associated with the regional background pollutant concentrations
(Gaynor, 1977).
4.5.3 Ambient Data
4.5.3.1 Size Selective Sampling for Particulate Matter
As stated previously (Levels I and II) the majority of available data for
particulate matter has involved the total mass collected on a hi-vol or
dichotomous sampler. The hi-vol has a nominal 50 percent cut size of 30-40
]om which indicates that large quantities of coarse materials (perhaps from
fugitive dust sources) will be collected.
Over the period extending from 1973-1983, an increasing number of studies
has been conducted by various organizations in which size selective sampling
of the particulate mass has been completed and composition data reported. In
many cases, a dichotomous sampler with a 2.5 \x& and 15 pm 50 percent cut
size was used for fine (respirable) and coarse particles. This appeared to
help differentiate sources in the specific cases used for source apportionment
studies (Dzubay, 1980; Kneip, et al., 1983; and Dzubay, et al., 1983.)
However, size selective inlet samples (15 \aa cut size), multi-stage
impactors, two-stage cyclone filter samplers and various other techniques have
been used to collect particulate mass samples.
These samples are collected on filter media which are then processed by a
variety of techniques for elements, inorganic compounds and organic
compounds. An up-to-date review of filter media and the concerns for sampling
and analysis has been published by Lippman (1983).
-128-
-------
Unfortunately, an all purpose sampler has not been developed; however,
several observations can be made. If the intent of the studies to be
conducted is on sources of sulfate, total carbon or size fractions of the
particulate mass, the dichotomous sampler would be the common sampling device
used for air pollution measurements. However, if the concern is for species
which need large quantities of samples in order to be above the detection
limit, such as polycyclic aromatic hydrocarbons or other organic species, a
different sampler with the appropriate size selective inlet would be required
for the study.
4.5.3.2 Time Resolution
This aspect of sampling is always difficult to judge precisely. The
purist would want to have a continuous record or at least frequent samples in
a given day. The pragmatist must make a cost effective compromise. What is
actually necessary is good planning, based upon the type of the pollution
problem suspected as well as the modeling techniques to be used for the source
apportionment studies. Traditionally, 24-hour sampling periods have been
found to be adequate for measuring the particulate matter levels and long-term
variations and trends in an area.
In the case of 24-hour samples, a large number of samples (>50) are
required to conduct multivariate analyses. In addition, these would have to
be either spaced over the seasons of the year or conducted in the same season
over a number of years. For mass balance techniques, much fewer samples are
required because the mass balance analysis can be performed on a single
sample. The number of samples required is determined by the extent of the
need to assess the problem on a 24-hour or annual basis.
-129-
-------
Alternatively, to focus on one particular source or to examine diurnal
differences, a series of shorter duration samples could be taken (from one to
six hours in duration). Stricter requirements will then be necessary for the
meteorological support data, and appreciation of seasonal differences must be
noted in the sampling strategy. In addition, these short-term samples must
contain enough material (mass of particulate matter) to make is possible to
perform gravimetric and chemical analyses.
Finally, in all receptor modeling studies that have used catalogued source
profiles, the adequacy of those profiles for representing emission conditions
during the sampling period is always a serious consideration. Therefore, any
new source test should be done as close to the time of sampling as possible.
4.5.3.3 Composition Determination
The previous volumes in this series and other articles, including the
results of the Quail Roost Workshops I and II have discussed, for possible use
in receptor models, a number of species and their limits of detectability. In
Volume II of this series, much time was spent on species which should be
determined for the mass balance technique. However, the use of other
approaches including multivariate modeling and electromicroscopic identi-
fication would have different requirements. To compound this problem, the
Level III approach may deal with some very unusual problems. Therefore, the
selection of -the correct list of species for analysis may require substantial
information gathering studies prior to the development of an analytical
protocol.
-130-
-------
4.5.3.4 Prioritization of Chemical Analysis
After selection of the variable (dependent) to be apportioned, a variety
of chemical analyses will be completed on the collected samples. The optimum
situation would be analysis for all species associated with the sources that
potentially will affect the site. However, some caution must be exercised
since this approach may, in fact, ignore a source that was not identified in
Level I or Level II studies.
Prioritization of the analyses to be completed is a necessity for cost
controls, and those species selected must, in many ways, be tied to the
objectives of the Level III study.
The case of particulate mass sampling (total or size fractions) is a good
example from which to develop some general analytical guidelines.
Farticulate Mass: Since this is the dependent variable in
receptor models, and the objective of source/dispersion models it
must be measured. In traditional studies, it was TSP; however,
this measurement could be of a particular size selected fraction.
(In other cases, the material to be apportioned could be
S04 "2, carbon, organic compounds, etc.).
Trace Elements: A majority of the information that has been
obtained from previous receptor modeling has come from trace
element source signatures. These will be of great necessity in
any Level III study. In fact, these data will be required for
both source and ambient samples since size selected studies will
probably become more frequent in the future and relatively few
particle size fractionated source test results are available.
Methods for elemental analysis described in previous volumes
indicate that the multi-element approach (as opposed to unique
tracers) seems to be the most effective. For each study, the
method of choice must be critically evaluated to prevent the
exclusion of a major element from the analysis. Morphological
studies may also be necessary to further assist in source
identification (see Volume IV, U.S. EPA, 1983b, for details).
Inorganic Ions: In most instances, except those where one source
dominates the collected particulate mass, various anions (and
their cations) will comprise significant portions of the
particulate mass. Therefore, the immediate thought is to
-131-
-------
incorporate them into a receptor modeling study. Before this is
done, however, the partitioning between primary emissions and
secondary products, and the fractionation according to particle
size must be considered. Differentiation will assist in local
versus distant source allocations. To accomplish the task, a
Level III study may require a large amount of ambient samples and
meteorological data for the development of multivariate models and
before such information can be used in a mass balance study.
Without partitioning of the local and regional contributions, the
primary value of the ion data will be for composition rather than
apportionment of sources. In the case of inorganic ions such as
sulfate, the combined use of dispersion and receptor models will
be advantageous in helping differentiate local contributions.
Organic Matter: There are a number of different levels from which
to approach the use of total organic matter or individual organic
compounds in a receptor model.
- Extractable Organic Matter (EOM): Depending upon the fraction
of EOM examined, the extractable mass data can give information
on primary organic material and potentially on secondary
organic material. These types of data are useful in further
defining the sources of the material affecting a receptor.
However, these data may in fact be more effectively used as the
dependent variable in an organic particulate matter source
apportionment model.
- Organic Carbon/Elemental Carbon/14C: These data are
available from previously described analytical techniques, and
can be used for assisting in the apportionment of particulate
mass in the atmospheric samples. Emission tests and ambient
sampling can be completed for carbonaceous material.
Therefore, if these data can be obtained in the same samples as
the trace elements, the ability to differentiate sources may be
extended. A particularly promising tracer is that of
14C/12C, which may be useful in identifying the
contribution of "contemporary" carbon sources, such as
woodburning in fireplaces and stoves.
As in the case of EOM, these materials can also be used as the
dependent variable in a receptor model if the appropriate
emission inventories are available for use in a mass balance
model. However, a thorough catalog of sources would not be a
requirement for use in a multivariate model apportionment study.
- Polycyclic Aromatic Hydrocarbons (PAH): Because PAHs found in
airborne particulate matter are the most carcinogenic component
in animal bioassays, they may present major concern to the
public health. However, for apportionment studies, they can be
used in three ways: 1} specific PAHs can be used as tracer
variables for combustion processes; 2) as the dependent
variable in a receptor model or 3) as an emissions term in a
dispersion model. Depending upon the type of data (source and
-132-
-------
ambient) either mass balance, factor analysis or multivariate
target transformation or multiple regression techniques can be
used to analyze these data in receptor models.
One caution which must be mentioned on this topic - the
partitioning of each PAH between the gaseous and particulate
phase must be considered before it is used in a receptor model
or dispersion model. In some cases, such as Benzo-a-pyrene
(BaP), this may be insignificant. However, for other compounds
a temperature dependence may be apparent and will affect the
validity of any apportionment conducted.
Other Materials: Most of these may be available from routine
monitoring data, and can be useful either as the dependent
variable or as further qualitative information to define sources.
The primary species include: CO, SQz, N0«, Os and HC.
However, in the future, volatile organic compounds will probably
gain in importance from the point of view of toxic substances,
photochemical smog precursors, and as source signatures.
4.5.4 Procedures (Level III)
It is assumed that Level I and Level II and what has been discussed
previously in Level III is in place, or is available for use. In designing a
source/dispersion modeling study, one would select the monitoring network
which most closely reflects the majority of source-meteorological factors
(such as prevailing winds, sea breezes, etc.) that may be encountered in a
region. Thus, the maximum number of sites would be dependent on a coverage
factor for the area source, frequency of meteorological conditions and
strength of source.
Unfortunately, the case is not as simple for a receptor site which would
be an impact site for individual sources that are examined by the dispersion
model. Such a site can be influenced by many different sources depending upon
the season or daily meteorological conditions. A major concern would be
coverage by secondary sites surrounding the primary receptor site. Sometimes,
if a large urban area is to be studied, more than one receptor site could be
chosen since each may be representative of different activities of the area.
For instance, the sites could be distributed in downtown commercial.
-133-
-------
residential, or industrial areas. Obviously, in each case, a different aspect
of an air pollution problem will be examined and the final selection of the
number of sites would be dependent upon the overall goals of the
investigation. For instance, the sites selected for examining TSP or some
size fraction in an urban nonattainment area might not necessarily be
equivalent to a site where there is a potential impact of hazardous materials
in a residential section. For a TSP or PM10 study, two or three sites at
different locations may be warranted. In the latter case, one or two sites
may be used as the receptors in the same neighborhood.
Once the primary site or sites are selected, of utmost importance is the
selection of a rural site located on level terrain away from significant local
sources and meteorological perturbations. These sites must provide coverage
for the dominant wind directions; however, in many cases, at least two
"upwind" sites, i.e., along the dominant wind paths, should be required
although anywhere from 1 to 4 sites are conceivable. Under no circumstances
should the study be done without at least one background site.
A primary consideration at the rural sites is the travel time and
technician time needed to obtain the samples used for determining regional or
background levels of pollutants. Obviously, the major species should be
measured as should the major tracers. Beyond that, it depends upon the
availability of resources. The optimum situation would include completion of
the same set of analyses on the background and urban samples (receptor).
The sampling duration and sampling frequency will have been determined
prior to site selection; however, there should be some flexibility in the
design. It may be necessary to adjust the sampling schedule if a major event
occurs, unexpected results begin to appear, or unanticipated problems arise.
Crucial in all these studies is the development of an adequate sampling and
-134-
-------
analysis quality assurance program. Without this type of control on the
study, the results may be less than adequate for use in model development. In
essence, error terms must be minimized. EPA guidance on the preparation of
quality assurance plans is contained in a document entitled Interim Guidelines
and Specifications for Preparing Quality Assurance Project Plans, (U.S. EPA,
1980c).
The final aspect of this study design is the model development and a
number of a priori issues must be considered. At a minimum for dispersion
model validations, the cases selected require short duration samples and the
periods chosen should adequately reflect the emissions inventory or operating
parameters that could be in effect at that time. Thus, for example, the
analytical protocols could involve the following steps:
Dispersion models are to be run using the source emission
inventories developed in Level III.
Selection of meteorological parameters corresponding to those
observed during the sampling study.
Selection of emission conditions which would closely
approximate the emission strengths that occurred during the
sampling period.
Complete a mass balance or microscopic analysis on a particular
sampling day (or on days selected to represent an annual average).
Select a sampling day during which the source is more nearly
upwind of the receptor site.
- Complete mass balance or microscopic analysis on that day and
estimate mass contributions.
Compare the dispersion model source contributions with those of
the receptor model. Those contributions estimated by each model
which do not differ by more than the sum of their uncertainties
are consistent. This consistency provides a good basis for
justifying a control strategy applied to these sources. The
dispersion model may be re-run for individual sources within one
of these source categories to specify which one might result in
the greatest pollution reduction per dollar spent.
-135-
-------
If source and receptor model results differ by more than the sum
of their uncertainties, the data on which both operate need to be
refined. The nature of this refinement is often indicated by the
model results.
- If the receptor model identifies a source contribution which
wasn't included in the emission inventory for the source
model, an inventory for that source must be carried out.
Wood burning (Core, et al., 1982) was discovered as a major
source by this procedure.
- If the receptor model finds a "downwind" source to be a major
contributor, meteorological inputs to the source model should
be examined and improved.
If all source type contributions calculated from the source
model are high or low with respect to the receptor model, the
source model should be re-run with revised or corrected
stability and mixing height data.
If model calculations differ by a factor or two or more, the
emission factors and positions of the gridded emissions
inventory for the source model should be double checked.
Alternatively, additional source characterization tests for
the receptor model may need to be made. By confining these
activities to major contributors with major discrepancies,
the work can be focused with great savings of effort over
verifying all source emission rates and compositions.
When the data have been refined, both models should be re-run and
the two preceding bulletad steps should be repeated.
If comparisons are still inadequate for decision-making purposes,
and if the cost of a wrong decision merits it, artificial tracers
may be introduced into the questionable emissions, simultaneous
measurements may be made at both the source and receptors, and
source and receptor models may be re-run to identify where the
discrepancy lies. The basic assumptions and structure of the
models may have to be modified as a result.
Concurrently, the mass balance results should be intercompared with
studies that use any one of a number of microscopic techniques. Comparisons
with other mathematical techniques such as factor analysis or multiple
regression, or combinations o£ techniques such as target transformation factor
analysis, or factor analysis with multiple regression, may be useful.
Conversely, if microscopic methods are the primary receptor models used, they
should be intercompared with mathematical methods. In these cases, the
-136-
-------
sources' contributions to a percentage of the mass will be determined. The
results must be compared for consistency and errors.
4.5.5 Resources
Resource requirements for level III are large. A mini- or main frame
computer will probably be needed to store data files and to operate
comprehensive programs. Since new sampling programs will be conducted, the
requirements for monitoring equipment, personnel and laboratory analyses will
be sizeable. Manpower requirements may encompass man-years and study
durations can extend to a year or more. Costs will range from $100,000 to
over $1,000,000.
4.6 Summary of the Three Level Approach
Studies conducted to determine contributions to measured particulate
matter concentrations can require different levels of effort. The level of
effort required in each case will depend on the nature and extent of the
problem and on the amount of information already available. This volume
organizes the analyses used to solve source apportionment problems into a
three level format. Table 4-8 summarizes the various aspects of the three
levels of study. The three level format is flexible and the information in
Table 4-8 should not be rigidly confined to each study level. In real world
applications, there may be intermixing of components of the three levels due
to data and resource availability.
Two items are applicable to all three study levels. First, an analysis of
background concentrations is an essential preliminary step to the other
analyses in each level, and second, two or more source apportionment methods
should be employed, if possible, at each level in order to provide greater
confidence in the study results.
-137-
-------
I"H
t-H
>"
-
^>
UJ
^j
£
^ 5
i .,
0. Ul
a. >
< Ul
_j
-j
Ul
Ul
00 _l
T Ul
Ul
Ul Of
^J X
CO 1^
^
h- Ul
^
U,
a
(/i
*<
u
7
Ui
>
1-
>-
3
>-
H
<_
4
at
o u ui
at vi u a> <4-
VI It) 3 VI O
3 O X
.c i/i i- e
o o i us o
C 3 C
n) ui 4J its -J
I ITS
O Ul 3 01 0
a» S S e
crt OL 01 ^* *^
3 O * ^ *
A5 i -O C 41 4J
i- a» o -o c
ai > i- its g at
o a> a u s -o
w-o u
a o at u
V» 4J O 01
c vi x a vi
X U 01 4) 4J C
» i 4> * U ITS O
t4^ ^ * 4V
U 4-13
ui at o a ts c -a
o 4^ a c m
XOI - V) Ul Ul U -UT30I
i 41 43 I/I 41 41 U
o £ s- its a u u
> a ~a its at i/i ITS 3
co~aa)u aso
>-t < vi a vi vi
4J '« 4J C 0
VI its O u-
*» 01 4V -f 4V
X Ul ..«> Ul 4V 3
oi 3 4V 4= c -a
e at at
«- o its a-o u
O OI 01 3 VI 4V
c u w e
vi c "oxo
01 i. its 01 i U
VI O > U TS 01
x i -o g oi > at
its 3 S 3 3
« s « at at i-» o
ui i. o vi
XXX 3 C
at ot ui *o o vi
c c ^3 at x c o at
§3 i re f
a-. x«-
It) VI > >
4V O C -- 1 1 4V
3^
its
C at at at g vi 3
ui a.**c> S3 cr
o
u v.
IT3 0
X 1 VI <*
i. at
o S vi vi
X <*- 1. ui 3 C
S4V O r i U O
£ -J 01 41 O
a at g 3
c IT) o S vi - a
U 01 V. VI
VI 4J W C O U i.
01 VI a 4V O 4V
u oi 3 j.) e
- > Ul J3 3 O
3 at i- ja u
a* i 4j at s. -
ai a it) a 4V v. at
u s 4j ui e 4V >
o c
VI i -O U O VI
It) (T) V 4J
o at o o c ui
X Ul C^ Cu- O C
at at at «- u o
at 3i."~>- at 4V
4V 3I.C O Ul 41 4V "O cO
ui c ui x vi 4t j* C 3
« >« ^" I/I r « « Ot > 4J
x w * at r *<^ * 3 o fm
at O a ui ns ui cr u ui
4j = xui -Juioia
VI Oi 01 01 C 01 Ul 4j
s |< So IS e 35 I
e
o
jj
a
4
4 1.
j U
4 o3
J O
1
01
CD
OI
C I
-o
vl C
3 3
O
vi ui u
41 S 01
ITS (/I *^" *
u a. 4J o>
< u S
c x ID 3
o ui u
* ui i* a
VI 2
Ul 41 S TT
. M W
S ul «- O «
01 0. VI u. OI
vi 4j a vi <<* 4u
<«_ *. c a oi vi
VI O C 01
at at oi - "
c 3 e « 4Ji
o- <- c a t-
4^ *" 4V VI 01 S 01
Ul C Ul O 3 S U
01 ^ at ar*"> vi its
01 O *fc«> Oj -^
at 4«> at u oi s
U (J 3 Ul
U -O i- i C r- C
3 u 3 it) o a o
o m o u
I/I t3 VI 4J tj VI
c a 3 oi ui
x Q
a- u. 5 S
1401 u its VI JS
Ul 41 U VI O it)
at u e a, i-
4j 3 a -a g oi
its o a c o
w u m ui" 10 u i
u VI
c e *- u -a>r
§& 3 4^ o 4^m
10 o a c> at u vi
^ m i at ^ ot 4^
§'_ CJ U ^ U > c Ul
j: o ID x e o at at
u 4^ a a_i_a z o ui -u
VI
c
o u
C 41
O « Ul U
Ul It)
4J N T U
ITJ Ul S 4J
3 2 ""In c
0) g«- CO 41
o a a
U- U 4J X
c a c vi M
O O M ui
U VI O 41
at 44 s a. u
at 4J 3s w
) VI "O
10 cg3s^ a
ui o it) ol at 3
01 -^- 01 a o
W -J N UI4- X U
i« its it) o -u or
U VI U Ul >
u a vi at vin
O4j'"f 13 W O 4J
4J4IOU1.3 C
vigo o- _i .a ui at
a
It)
a
0)
u
u
3
% -138-
01
u
It)
<*. Ul
3
m - o»
01 T
r \- 01
4V 4-1 01
c ITS a u
o u c e
TH at 01
a x i
1 I'M
3
U U Ul 4-)
O O 01
c
OI OKU. O
c c o
U 4J
i- u a o
005
u 4-1 -a
e u
g g x Its
I £
at - as its its
4W^ 4^ (J ^J
vs
ui -a ui 4j 4_>
e c at at
a x a > >
.
VI
at at i i
u u a> ITS
It) 3 U VI U - i-
<4- 4j a. x ai c at '
uiTsoiitsao a 01
3 u u -a = a a
1/1 oi 01 4j 3 a
§£ -O 4_> us 3
o at 4j -o
^S ^J *^Bt^ (J -^
W -W *4 U W £X U) (Q C
3 at -o -a us
O -Ulr CUC41
£ vi 4J at at * (j at
41 o» a '4- re
. 01 U 4J U U It-
IT) a 41 O C - 3 U
CJU1.SU-IT3VIITSU1 3
4-1 -o o) ui ^ jr u x
4ICeC3OtOI> X
oSio ovi as-
4-1 X 0» 41 a 3
O VI 4W U J= 3 On ITST
acs»
3
B A
VI VI
o . at
at vi u
41 01 3
a 4J
Ul 4J It) -O
o£ "§ e at
e « o ax
X It) VI 4J 01 U
01 VI 01 4J It)
U 01
ITS o a a vi u
*. vi TO 01
u u u at a
3 at T3 at at a
vi £ e v. a 3
a a vi
X vi 2
'u s 01 'fl e a
JITS ID o Xtr S
VI U. VI VI
4-1 1_ (TJ 01 4-> U
r~ VI C 3 01 01 -C 01
it) oi a VIM vi ia o> .c
U U VI U 4W
3 01 4J c u oioiitsvt
O) ITS HS-^ NIT) X «
JZ S- O> 4-1 4J » Ul OI U
watcr-uitsxcxits
Oa 30l3nSi - g
xot at eolrs 13:J
x i so:-o
-------
a:
^
i
Q.
O.
o
01 -J
3UI
C >
* *'J
s-
CJ UI
Q:
X
09 h>
i
r ui
ui £
|o
>-
CE
1
t-l
is
UI
«J
1 V
01 I/I I/I
u a O C
N .UI a t «
-.-rOt.ro
V) <<- O C VI X
i U VI U
3 U 3 « 0
vi vi o *- s .u»
I/I u- -o vt S o> S 01
VI ^ O -** I/I H- JJ Q.
00 J*~ S
JS ra VI ro
JJ U t- O VI
iJ 1-13 C 5 U
l-q j_i 01 O 01
-J
LU
>
UK
_J
*H
U
>
U
>.
H-
MH
>
h*
H-
(_)
<
C i- U U JJ
0» O C i
(J O r
C VI u (O *fc.
Of- U «
u o *J -o a
i. * c S *5
01 3 J= U r-
JJ & Ul U I/I vi O VI
c u u
01 JJ O 01 2
J-l U U r X 0)
i t I/I 01 S I/I
3 41 I- S 3T3
U I/I 41 U l/> O OJ
U 01 Q.J-1 £ >
w N Su u 01 S
rO ro f8 3 W «-
o. vi vi ew vi a. _i
§U 01
«- M
«- VI VI VI
L. e oi
VI O O 01 >-
c a
O VI W (J T S
i a <- t/i io
JJ O U JJ u 1/1
f9 > u W p-
v. i c ra 31-
ij at a vi 01
C JS U 01 JJ
0) c a i_<
U 01 O C
c > u -T xi.
O vi o oi 3
« 1/1 3 o u ax
rs 9 g. c i
3 VI 3 i (OIOU3
(J * *J U VI O O O
# o O « AJ *
^^ ^ ^ 4J JS E ^* ^
u i u u u S e oi
ro ra 3 JS O U
Q. JT T3 OL VI UZO.
j^
a
c
01
^*
S
5
(0 1
01 1 ai w o
J3 4J^- 0 0
o vi ja *>
(v lO fO i. u *»
r- 01 O
u -o .0 o
01 JJ *^" (O J^
U VI ra
O wraoix-«">o
Z CCft Z XUI J3<_)
VI
VI £ X
£ JJ U
*j e o
c g ' o
di o c io o
J3 E 1-0
S e o
x ra ja o
§0 = ro O
01 i-
u *»
U JJ 01
01 3 VI *J O
jj fr 01 ra jj
3 01 -0 W
tu o 01 o
= 00
vi go
u ~o ui s
0 *J U»
§-^- C CM
u 01 a *»
ra 01 E c
w a 01 S
X
e 01 i-
§01 O
u -< o
01 j= ra a
-^ O " U O
M w O -
- -» o a u>
l i. 71 < IO CM
^3 ^^ 01 41
ra ra 01 C I- O
a./ 20 o M
<0 3^>-~-
u u a o
i .wi C O
fa u g o
§o o S u
e O vn
a vi e M *»
^ jj eu "S 'e P
ra I &= S
u ^ ra VI S 01 U
3 VI U "O -OU_
a 01 1 01 o oi
= T3 01 » C 01 01
a « c "> fi
U E '- C
a u .uj o t/i ra
^- 01 *j 01 e 01 u
§1- 2 O 01 vi
X O S >> VI
3 TS a. u *- ^
-------
5.0 CASE STUDIES OF COMPOSITE SOURCE APPORTIONMENT METHODS
There are a significant number of examples where composite source
apportionment methods have been used effectively to answer questions on the
degree to which various source categories cause a nonattainment situation or
other air quality situation of interest. These studies cover the full
spectrum of complexity from Level I to Level III, Since the dividing lines
between the three levels of complexity are not clear cut, it follows that many
of the case studies may contain analyses that might logically be part of more
4
than one complexity level.
The following case studies exemplify the three levels of complexity of
composite source apportionment methods.
5.1 Use of Microscopy and Filter Analysis (Level I)
Drafts, et al., (I960) of the Illinois Institute of Technology Research
Institute (IITRI) have developed a protocol using microscopic identification
of particles collected on hi-vol filters supplemented by chemical analyses of
the filter deposits to apportion the collected particles to various source
categories.
Polarized light microscopy is the principal tool used to identify particle
sources based upon the previous microscopic analysis of particle samples
collected from various sources. X-ray diffraction may also be used to provide
positive identification of mineral species and scanning electron microscopy
may be used for identification of extremely small particles.
Low temperature ashing is used to determine the organic content of the
collected particles. Ion chromatography is used to determine the anions and
cations present, and depending on the nature of the study, elemental analysis
of samples may be carried out using x-ray fluorescence.
-140-
-------
Data are presented in terms of percent contribution of source categories
such as combustion, specific metallurgical operations, road traffic, and
biological aerosols.
5.2 Linn County, Iowa Non-Traditional Fugitive Dust Study (Level I)
A recently conducted study by Brookman (U.S. EPA, 1983d) investigated the
primary non-traditional sources affecting the air quality in and around Cedar
Rapids, Iowa (Linn County). The approach used for this study was,
essentially, the Level I protocol involving many of the analysis discussed in
Section 2.3.7.
The data base used for this study included existing source, meteorological
and ambient data. The source data were in the form of existing
microinventories around the five county monitoring stations and one rural
background station, a point source inventory performed by the local agency,
and a "first-cut" area source inventory (part of this study was to update this
inventory and then use the results for the nonattainment study). The source
data were supplemented with topographic maps, aerial photographs, and an
industrial fugitive source inventory (performed as part of the project). The
meteorological data were obtained from the Cedar Rapids Airport in the
standard form of LCD's (Local Climatological Data Summaries). The ambient
data were obtained from the State authorities and consisted of seven years
(1976 - 1982) of hi-vol data collected at the one background and five county
stations.
The first step of the protocol was to visit the monitoring sites to get a
"feel" for local contributing sources, topographic anomalies, siting
characteristics, etc. The second step was to evaluate the meteorological
data. This included determining wind persistence for each of the sampling
-141-
-------
days during the seven year period and noting precipitation and snow cover on
and preceding the sampling days. The third step was to determine the
background level for each year of the study period. This was done by using
the rural station hi-vol data. This level was found to be highly variable
from year to year, in direct correlation with the amount of precipitation for
that year (in fact, a linear relationship was shown to exist).
The next step in the protocol was to perform a trend analysis (see Section
2.3.7) and remove the background influence (thus effectively removing the
meteorological influence). The resulting trend plots for the five stations
showed definite anomalies at certain stations during certain years. Further
analyses were performed to try and discover the reasons for these anomalies.
First, the modeled point source influences were removed from the yearly
averages (the State has performed ISC modeling and information was available
on the relative monitor impact for each year). Second, pollution roses were
developed based on days with persistence greater than 0.71 (see Section
2.3.7). A pollution rose was also developed for the background station and
its influence was removed from the county station pollution roses.
Using the results of these basic analyses in conjunction with the
topographic maps, aerial photographs, site visits, discussions with County
officials, etc., probable causes for the nonattainment status of several
monitoring stations were uncovered. The principal findings of the study
showed that highway construction was the principal cause of TSP violations at
several stations throughout the years, industrial fugitive sources contributed
several micrograms per cubic meter to a few stations, and vehicle traffic on
paved and unpaved roads contributed heavily to most of the stations. The
results also showed that there was a general downward trend in the TSP levels
throughout the county throughout the years as a result of control programs
-142-
-------
(street cleaning, paving) and the economic recession (several large industries
had cut back their operations significantly).
The results of this study, which was conducted entirely using existing
data, were very helpful to the local authorities in planning future control
strategies (particularly in regard to construction projects). Example results
for one of the five monitoring sites are shown in Table 5-1. While the Level
I approach could not determine exact levels of source contributions or
distinguish between several fugitive sources in the same area that were
probably affecting a monitoring station, it did provide the authorities with
all of the information they desired. The next logical step for further source
contribution quantification would have to entail microscopic analyses of
filters and/or directional sampling.
TABLE 5-1
ESTIMATED SOURCE IMPACTS AT ONE LINN COUNTY, IOWA
MONITORING LOCATION (EQUIVALENT GEOMETRIC MEANS ug/m3)
Year
Source Type
Background
Traditional:
Stack
Fuel Combustion
Solid Waste Disposal
Auto Exhaust
Annual Recorded Mean
Non-Traditional Impact
1976
47.0
8.3
1.0
2.1
2.7
98.3
37.2
1977
38.6
8.8
1.0
2.1
2.7
109.3
56.1
5.3 Allegheny County Particulate Study
1978
37.9
3.3
1.0
2.0
2.7
90.6
38.2
(Levels
1979
35.8
3.8
1.0
1.8
2.7
95.0
44.9
I and
1980
40.7
3.3
1.0
1.7
2.8
106.5
51.5
ID
1981
36.5
3.3
1.0
1.5
2.8
80.0
29.4
1982
26.0
3.8
1.0
1.4
2.8
60.9
20.9
A study was carried out in Allegheny County (Yocom, et al., 1979, and
Brookman and Yocom, 1980) to investigate the particulate matter nonattainment
problem in the county and to determine the types of sources primarily
-143-
-------
responsible for the nonattainment situation. Of particular interest were the
relative contributions of the following types of sources:
Traditional - Industrial point and process fugitive sources.
Non-Traditional - In-plant open sources such as roads and material
piles and non-industrial sources such as public roads and
playgrounds.
Background - Natural sources and sources outside the county.
This study was an example of a composite source apportionment method
applied principally to available TSP data collected by the County Bureau of
Air Pollution Control. Most of the methods described in Section 2.3.7
(Preliminary or Qualitative Receptor Models) were applied to this 'data base.
Dispersion modeling using a version of PAL together with a detailed inventory
of traditional and non-traditional sources at the large, integrated steel
mills was used to predict specific contributions of these sources on selected
sampling sites.
Background levels of particulate matter were determined by the "composite
particulate rose" method described in Section 2.3.7 and applied equally to all
sampling sites in the county.
The assumption was made that the industrial component of measured
particulate matter at stations near the steel mills was made up entirely of
steel mill contributions. These stations constituted most of the
nonattainment stations. A variety of techniques (e.g., pollution roses,
wet/dry day analysis, microinventories, etc.) was used to infer the relative
contribution of traditional and non-traditional sources for the remaining
sampling stations in the county.
Special short-term sampling runs using Millipore filters were carried out
at several of the sampling sites under predetermined meteorological
-144-
-------
conditions. These samples were subjected to automated scanning electron
microscopic and energy dispersive x-ray analysis (SEM/EDAX) to determine the
chemistry of the selected particles in relation to particle size. By making
certain assumptions on the contributions of traditional and non-traditional
sources to specific particle chemistries it was possible to reconcile the
dispersion modeling-plus-background source apportionment results with those of
the particle identification studies.
Using estimated predictions of future reductions in emissions from
traditional and non-traditional industrial sources, the number of nonattaining
stations decreased and strategies could be developed for bringing the
remaining nonattaining stations into an attainment status by application of
controls on various categories of non-traditional sources.
The actual contracted cost for this study was about $60,000 in 1978
dollars. However it should be pointed out that the contractor served as a
consultant and management contractor and was asked to utilize much available
data and systems. One important feature was that 5 years of TSP and
meteorological data were already up on a large computer and were readily
accessible for a variety of manipulations. Furthermore, the dispersion
modeling and steel mill inventory were carried out by another contractor under
the sponsorship of a steel company consortium, and the special sampling and
SEM/EDAX work was funded directly by the county with yet another contractor
group. Thus the total costs for the project were estimated to be in the range
of $300-500,000.
5.4 Portland Aerosol Characterization Study (Level III)
The Portland Aerosol Characterization Study (PACS, Cooper, et al., 1979;
Watson, 1979; Core, et al., 1982) is a classic example of a Level III study.
-145-
-------
Its purpose was to quantify the major contributions to TSP concentrations
which were in violation of the NAAQS. It involved the following elements:
Creation and use of a site specific air quality dispersion model
(Pabrick and Sklarew, 1975).
Multiple site meteorological monitoring to create a complex wind
field.
A site specific gridded emissions inventory for one year.
A comprehensive study design and preparation (Mueller, et al.,
1977).
Meteorological regime stratification of sampling schedules to
minimize costs while still representing an entire year.
Characterization of all major source emissions with respect to
particle size and chemical composition.
Time, space, size, and chemical resolution of ambient suspended
particulate matter.
Identification and quantification of major sources by the mass
balance receptor model.
Comparison of mass balance and dispersion model results and
adjustments to each.
One of the unique aspects of the PACS was the interaction of the source
and receptor models to reinforce each other. The emissions inventory created
for the dispersion model was used to identify the most likely sources for
emissions characterization. By its estimates of likely ambient concentrations
and variability with wind direction, it allowed necessary analytical
requirements (detection limits) and sample durations to be estimated.
Initially, mass balance calculations did not agree with dispersion model
results. An examination of the emissions inventory resulted in corrections
and results of the two models did agree within reasonable margins for error
(Core, et al., 1982). Control strategy options were outlined which probably
would not have been proposed without the results of the PACS. Several of
these options are being implemented and it appears that Portland's air quality
is improving.
-146-
-------
Because of the in-kind donations and development work which went into the
PACS, it is difficult to estimate its total cost. Estimates range from
$300,000 to $1,000,000. The cost of misplaced controls is agreed to have been
much higher than the upper limit of the cost of the study.
-147-
-------
6.0 SUMMARY AND CONCLUSIONS
6.1 Summary
It should be apparent from this technical document that, considering the
state-of-the-art of receptor and source (dispersion) modeling and composite
source apportionment methods, there are no fool-proof step-by-step procedures
or recipes for studying particulate matter apportionment and making decisions
on the results. Nevertheless, there are a large number of useful tools to
assist the air resource manager in carrying out such studies. The selection
and application of these tools will depend on:
Objectives of the study.
Resources available.
Technical expertise and innovativeness of the project team.
Availability and accessibility of data bases.
Availability of sophisticated sampling and analytical equipment.
Some of the techniques described are well within the resources of even
small agencies (e.g., the methods described in Section 2.3.7 - Preliminary or
Qualitative Receptor Models) while others require highly trained statisticians
and computers of significant size (see Section 2.3.2 - Factor Analysis) and
meteorologists with considerable modeling experience (the ISC model described
in Section 2.2.2).
It should be clear, however, that in apportioning source contributions in
ambient particulate matter studies no single method (e.g., dispersion modeling
or a specific receptor model) will ordinarily suffice. Invariably some sort
of a composite methodology will be needed.
It has been reasonably well established that most dispersion models when
properly applied will predict source impact at a receptor within a factor of
two of the correct or measured value. As was pointed out earlier, the mass
balance model can apportion generalized source categories within this same
-148-
-------
range. If one can assume that a dispersion model operating on several sources
and a single receptor is a more accurate representation of relative
contribution than of total impact, it follows that combining the dispersion
and receptor models or two receptor models will produce a much more accurate
representation of source contributions than use of each independently. While
it is easy for anyone working in this area to believe this statement, the
state of technology is such that it is not possible to put error bars on the
results of any application of composite methods. Suffice it to say that the
results would be better than the use of a single method. The same is true for
the combined use of two receptor models.
6.2 Conclusions
An analysis of source contributions to particulate matter levels at one or
more receptor locations should start the analysis with elements of the Level I
approach (Section 4.3). The air resource specialist responsible for the
source apportionment study would need first to become intimately familiar with
available data on particulate matter, their trends, the adequacy of the
sampling array and the influence of basic meteorological factors as outlined
in Section 2.3.7. Furthermore, if an adequate emissions inventory for
particulate matter sources did not exist, preparing such an inventory would be
a first order of business. The same can be said for the assembly of existing
meteorological data.
In progressing to higher levels and more complex and costly approaches,
the analysts must carefully weigh the objectives of the analysis and the cost
effectiveness of more sophisticated techniques. If the source contribution is
needed in only general terms or, more likely, if the sources of nonattainment
are relatively straightforward, the techniques of a Level I approach may be
-149-
-------
adequate. If, on the other hand, specific contributions of selected sources
in a complex airshed are needed for control strategy development on a size
selective basis, and where emission controls will be costly, a Level II or III
approach will be needed. The cost range between Level I and III approaches
can vary by more than an order of magnitude (e.g., $10,000 to $500,000).
The application of composite source apportionment techniques is a
stepwise, and to some extent, an iterative process where the effectiveness of
each added technique is evaluated in terms of increasing the accuracy of the
results in relation to the added time and costs.
-150-
-------
7.0 REFERENCES
Alpert, D.J., and P.K. Hopke, 1980: A Quantitative Determination of
Sources in the Boston Urban Aerosol. Atmos. Environ., 14:1137.
Alpert, D.J., and P.K. Hopke, 1981: A Determination of the Sources of
Airborne Particles Collected During the Regional Air Pollution Study.
Atmos. Environ., 15:675-87.
Alpert, D.J., 1980: Quantitative Apportionment of Urban Aerosol Mass by
Factor Analysis. Ph.D. Dissertation, University of Illinois, Urban, IL.
American Society of Testing Materials, 1955: Alphabetical and Grouped
Numerical Index of X-Ray Diffraction Data. Spec. Tech. Publ. 48E.
Bowman, H.R., J.G. Conway, and F. Asaro, 1974: Atmospheric Lead and
Bromine Concentration in Berkeley, California (1963-70). Environ. Sci.
Technol., 8:558.
Britt, H.I., and R. Luecke, 1973: The Estimation of Parameters in
Nonlinear, Implicit Models. Technometrics, 15:233.
Brookman, E.T., and J.E. Yocom, 1980: Environmental Management: A Case
Study in the Use of Ambient Data for Source Assessment.
EPA-600/7-80-080. U.S. EPA, Research Triangle Park, NC 27711.
Casuccio, G.S., et al., 1983: The Use of Computer Controlled Scanning
Electron Microscopy in Environmental Studies. J. Air Pollut. Control
Assoc., 33:937-943.
Chow, J.C., J.G. Watson, J.J. Shah, and T.G. Pace, 1981: Source
Contributions to Inhalable Particulate Matter in Major U.S. Cities.
Presented at the 75th Annual Meeting of the Air Pollution Control
Association, New Orleans, LA.
Cooper, J.A., 1981: Review of the Chemical Receptor Model of Aerosol
Source Apportionment. Atmospheric Aerosols: Source/Air Quality
Relationships. American Chemical Society Symposium Series No. 167,
Washington, D.C.
Cooper, J.A., J.G. Watson, and J.J. Huntzicker, 1979: Summary of the
Portland Aerosol Characterization Study. Presented at the 72nd Annual
Meeting of the Air Pollution Control Association, Cincinnati, OH.
Cooper, J.A., and J.G. Watson, 1980: Receptor Oriented Methods of Air
Particulate Source Apportionment. J. Air Pollut. Control Assoc., 30:1116.
Core, J.E., J.A. Cooper, P.L. Hanrahan, and W.M. Cox, 1982: Particulate
Dispersion Model Estimation: A New Approach Using Receptor Models. J.
Air Pollut. Control Assoc., 32:1142.
Crutcher, E.R., 1982: Light Microscopy as an Analytical Approach to
Receptor Modeling. Presented at the APCA Specialty Conference on
Receptor Models Applied to Contemporary Air Pollution Problems, Danvers,
MA.
-151-
-------
Currie, L.A., et al., 1983: Interlaboratory Comparison of Source
Apportionment Procedures: Results for Simulated Data Sets. The Reports
from the Mathematical and Empirical Receptor Models Workshop (Quail Roost
II). Environmental Sciences Research Laboratory, U.S. EPA, Research
Triangle Park, NC 27711.
Dattner, S.L., and P.K. Hopke, 1982: Receptor Models Applied to
Contemporary Pollution Problems. Proceedings of the APCA Specialty
Conference, Pittsburgh, PA.
Draftz, R.G., J. Graf, E. Arnold, E. Grove, and E. Segers, 1980:
Allocating Fugitive and Point Source Contributions to TSP Non-Attainment
through Hi-Vol Analyses. Presented at the 73rd Annual Meeting of the Air
Pollution Control Association, Montreal, Quebec.
Duewer, D.L., B.R. Kowalski, and J.L. Fasching, 1976: Improving the
Reliability of Factor Analysis of Chemical Data by Utilizing the Measured
Analytical Uncertainty. Analytical Chemistry, 48:2002.
Dunker, A.M., 1979: A Method for Analyzing Data on the Elemental
Composition of Aerosols. Proceedings of the American Chemical Society,
September 12, Washington, D.C.
Dzubay, T.G., et al., 1983: Interlaboratory Comparison of Receptor Model
Results for Houston Aerosol. The Reports from the Mathematical and
Empirical Receptor Models Workshop (Quail Roost II). Environmental
Sciences Research Laboratory, U.S. EPA, Research Triangle Park, NC 27711.
Fabrick, A.J., and R.C. Sklarew, 1975: Oregon/Washington Diffusion
Modeling Study. Xonics, Inc., Van Nuys, CA.
a
Fox, D.G., 1981: Judging Air Quality Model Performance. Bull. Am.
Meteor. Soc., 62(5):599-609.
Friedlander, S.K., 1973: Chemical Element Balance and Identification of
Air Pollution Sources. Environ. Sci. Technol., 7(3):235-40.
Friedlander, S.K., 1977: Smoke, Dust, and Haze. Wiley Interscience, New
York, NY.
Gaynor, J.E., 1977: Acoustic Doppler Measurement of Atmospheric Boundary
Layer Velocity Structure Functions and Energy Dissipation Rates. J.
Appl. Meteorol., 16:148.
Gerlach, R.W., L.A. Currie, and C.W. Lewis, 1982: Review of the Quail
Roost II Receptor Model Simulation Exercise. Proceedings of the APCA
Specialty Conference on Receptor Models Applied to Contemporary Pollution
Problems. 96-109.
Gordon, G.E., 1980: Receptor Models. Environ. Sci. Technol., 14:792-800.
Gordon, G.E., W.H. Zoller, G.S. Kowalczyk, and S.M. Rheingrover, 1981:
Composition of Source Components Needed for Aerosol Receptor Models.
Atmospheric Aerosol: Source/Air Quality Relationships. American
Chemical Society Symposium Series No. 167, Washington, D.C.
-152-
-------
Gordon, G.E., et al., 1983: Considerations for Design of Source
Apportionment Studies. The Reports from the Mathematical and Empirical
Receptor Models Workshop (Quail Roost II). Environmental Sciences
Research Laboratory, U.S. EPA, Research Triangle Park, NC 27711.
Hanna, S.R., G.A. Briggs, and R.P. Hosker, Jr., 1982: Handbook on
Atmospheric Diffusion. DOE/TIC-11223. Office of Health and
Environmental Research, U.S. Department of Energy. National Technical
Information Service, Springfield, VA 22161.
Heidorn, K.C., 1978: An Index to Measure Consistency of the Wind
Direction for Periods Around One Day. Atmos. Environ., 12:993.
Henry, R.C., 1977: A Factor Model of Urban Air Pollution. Ph.D.
Dissertation, Oregon Graduate Center, Beaverton, OR.
Henry, R.C., 1982: Stability Analysis of Receptor Models that Use Least
Squares Pitting. Presented at the APCA Specialty Conference on Receptor
Models Applied to Contemporary Air Pollution Problems, Danvers, MA.
Henry, R.C., C.W. Lewis, Philip K. Hopke, and Hugh J. Williamson, 1983:
Review of Receptor Model Fundamentals. The Reports from the Mathematical
and Empirical Receptor Models Workshop (Quail Roost II}. Environmental
Sciences Research Laboratory, U.S. EPA, Research Triangle Park, NC 27711.
Holzworth, G.C., 1972: Mixing Heights, Wind Speeds, and Potential for
Urban Air Pollution Throughout the Contiguous United States. Pub, No.
AP-1Q1. U.S. EPA, Research Triangle Park, NC 27711.
Hopke, P.K., 1981: The Application of Factor Analysis to Urban Aerosol
Source Resolution. Atmospheric Aerosol: Source/Air Quality
Relationships. American Chemical Society Symposium No. 167, Washington,
D.C.
Hopke, P.K., 1982: Application and Verification Studies of Target
Transformation Factor Analysis as an Aerosol Receptor Model. Receptor
Models Applied to Contemporary Pollution Problems. Air Pollution Control
Association, Pittsburgh, PA 15230.
Hopke, P.K., D.J. Alpert, and B.A. Roscoe, 1983: FANTASIA - A Program
for Target Transformation Factor Analysis to Apportion Sources in
Environmental Samples. Computers and Chemistry, 7:149-155.
Houghland, E.S., 1933: Chemical Element Balance by Linear Programming.
Proceedings of the 76th Annual Meeting of the Air Pollution Control
Association, 83-14.7, Atlanta, GA.
John, W., et al., 1983: Validation of Samplers for Inhaled Particulate
Matter. California Department of Health Services, Berkeley. Air and
Industrial Hygiene Lab. Section. EPA-600/4-83-010.
-153-
-------
Johnson, D.L., at al., 1983: Chemical and Analytical Analyses of Houston
Aerosol for Interlaboratory Comparison of Source Apportionment
Procedures. The Reports from the Mathematical and Empirical Receptor
Models Workshop (Quail Roost II). Environmental Sciences Research
Laboratory, U.S. EPA, Research Triangle Park, NC 27711.
Kelly, J.P., R.J. Lee, and S. Lentz, 1980: Automated Characterization of
Fine Particulates. Scanning Electron Microscopy, 1:311.
Kerr, P.P., 1959: Optical Minerology. McGraw-Hill Book Company, Inc.,
New York, NY.
Kleinman, M.T., 1977: The Apportionment of Sources of Airborne
Particulate Matter. Ph.D. Dissertation, Mew York University, New York,
NY.
Kleinman, M.T., B.S. Pasternack, M. Eisenbud, and T.J. Kneip, 1980:
Identifying and Estimating the Relative Importance of Sources of Airborne
Particulates. Environ. Sci. Technol., 14:62-65.
Kneip, T.J., M.T. Kleinman, and M. Eisenbud, 1972: Relative Contribution
of Emission Sources to the Total Airborne Particulates in New York City.
Third IUAPPA Clean Air Congress.
Kneip, T.J., R.P. Mallon, and M.T. Kleinman, 1983: The Impact of
Changing Air Quality on Multiple Regression Models for Coarse and Fine
Particle Fractions. Atroos. Environ., 17;299-304.
Kuwana, T., 1980: Physical Methods in Modern Chemical Analysis, Vol. 2,
Academic Press, Inc.
Larsen, R.I., 1971: A Mathematical Model for Relating Air Quality
Measurements to Air Quality Standards. Pub. No. AP-89. U.S. EPA,
Research Triangle Park, NC 27711.
Lioy, P.J., and J.M. Daisey, et al., 1983: The New Jersey Project on
Airborne Toxic Elements and Organic Substances (ATEOS). A Summary of the
1981 Summer and 1982 Winter Studies. J. Air Pollut. Control Assoc.,
33:650-657.
Lippman, M., 1983: Sampling Aerosols by Filtration. In Air Sampling
Instruments for Evaluation of Atmospheric Contaminants, P.J. Lioy and
M.J. Lioy, ed., American Conference of Governmental Industrial
Hygienists, Cincinnati, OH,
Londergan, R.J., D.H. Minott, D.J. Wackter, R.R. Fizz, 1983: Evaluation
of Urban Air Quality Simulation Models. Report prepared for U.S. EPA,
Office of Air Quality Planning and Standards, Research Triangle Park, NC
27711.
Macias, E.S., and P.K. Hopke, 1981: Atmospheric Aerosol Source/Air
Quality Relationships. American Chemical Society Symposium Series No.
167, Washington, D.C.
-154-
-------
Mayrsohn, H. , and J.M. Crabtree, 1976: Source Reconciliation of
Atmospheric Hydrocarbons. Atmos. Environ., 10; 137.
McCrone, W.C. , and J.G. Delly, 1973: The Particle Atlas, 2nd Edition,
Ann Arbor Science Publishers Inc., Ann Arbor, MI.
Miller, M.S., S.K. Priedlander, and G.M. Hidy, 1972: A Chemical Element
Balance for the Pasadena Aerosol. J. Colloid Interface Science,
39<1):165.
Morandi, M.T., J.M. Daisey, and P.J. Lioy, 1983: A Receptor Source
Apportionment Model for Inhalable Particulate Matter in Newark, NJ.
Proceedings of the 76th Annual Meeting of the Air Pollution Control
Association, 83-14.2, Atlanta, GA.
Mueller, P.K. , S.L. Meisler, and S. Cohen, 1977: Design of the Portland
Aerosol Characterization Study and Associated Aspects of the Data Base
Improvement Project. P-5129.2, ERT, Inc., Westlake Village, CA.
Pace, T.G., 1978: An Empirical Approach for Relating Particulate
Microinventory Emissions Data, Siting Characteristics and Annual TSP
Concentrations. U.S. EPA, Research Triangle Park, NC 27711.
Pace, T.G. , 1983: Models to Develop Control Strategies for
Proceedings of the 76th Annual Meeting of the Air Pollution Control
Association, 83-14.3, Atlanta, GA.
Pierson, W.R. , and W.W. Brachaczek, 1976: Particulate Matter Associated
with Vehicles on the Road. SAE Automotive Engineering Congress and
Exposition, Detroit, MI.
Roscoe, B.A. , P.K. Hopke, S.L. Dattner, and J.M. Jenks, 1982: The Use of
Principal Component Analysis to Interpret Particulate Compositional Data
Sets. J. Air Pollut. Control Assoc. , 32:637-642.
Rozett, R.W. , and E.M. Peterson, 1975: Methods of Factor Analysis of
Mass Spectra. Analytical Chemistry, 47:1301.
Stevens, R.K. , and T.G. Pace, 1984: Overview of the Mathematical and
Empirical Receptor Models Workshop (Quail Roost II). Environmental
Sciences Research Laboratory, U.S. EPA, Research Triangle Park, NC 27711.
Strothmann, J.A. , and F.A. Schiermeier, 1979: Documentation of the
Regional Air Pollution Study. EPA 68-02-2093. U.S. EPA, Research
Triangle Park, NC 27711.
TRW Systems Group, 1969: Air Quality Display Model. PB189194. National
Technical Information Service, Springfield, VA 22161.
Taback, H.J., A.R. Brienya, J.F. Macho, and N. Brunety, 1979: Fine
Particle Emissions from Stationary and Miscellaneous Sources in the South
Coast Air Basin. Prepared for California Air Resources Board. KVB
Document 5806-783, Tustin, CA.
-155-
-------
Texas Air Control Board, 1979: User's Guide to the Texas Episodic
Model. Permits Section, Texas Air Control Board, 6330 Hwy 290 East,
Austin, TX 78723.
Texas Air Control Board, 1980: User's Guide to the Texas Climatological
Model. Permits Section, Texas Air Control Board, 6330 Hwy 290 East,
Austin, TX 78723.
Throgmorton, J.A., and K. Axtell, 1978: Digest of Ambient Particulate
Analysis and Assessment Methods. EPA-450/3-78-113. U.S. EPA, Research
Triangle Park, NC 27711.
Thurston, G.D., 1983: A Source Apportionment of Particulate Air
Pollution in Metropolitan Boston, Ph.D. Thesis, Harvard University,
Cambridge, MA 02138.
Turner, D.B., 1970: Workbook of Atmospheric Dispersion Estimates.
Office of Air Programs. Pub. No. AP-26. U.S. EPA, Research Triangle
Park, NC 27711.
Turner, D.B., 1979: Atmospheric Dispersion Modeling: A Critical
Review. J. Air Pollut. Control Assoc., 29:502.
U.S. EPA, 1971: User's Manual: SAROAD (Storage and Retrieval of
Aerometric Data). APTD-0663. Office of Air Programs, Research Triangle
Park, NC 27711.
U.S. EPA, 1973: User's Guide for the Climatological Dispersion Model.
EPA-RA-73-024.
U.S. EPA, 1977a: Guidelines for Air Quality Planning and Analysis Volume
10 (Revised): Procedures for Evaluating Air Quality Impact of New
Stationary Sources. EPA-450/4-77-001.
U.S. EPA, 1977b: User's Manual for Single-Source (CRSTER) Model.
EPA-450/2-77-013.
U.S. EPA, 1977c: Valley Model User's Guide. EPA-450/2-77-018.
U.S. EPA, 1977d: Addendum to User's Guide for Climatological Dispersion
Model. EPA-450/3-77-015.
U.S. EPA, 1978a: OAQPS Guideline Series: Guideline on Air Quality
Models. " EPA-450/2-78-027.
U.S. EPA, 1978b: User's Guide for PAL. EPA-600/4-78-013.
U.S. EPA, 1978c: User's Guide for RAM. EPA-600/8-78-016a.
U.S. EPA, 1978d. Digest of Ambient Particulate Analysis and Assessment
Methods. EPA 450/3-78-113.
U.S. EPA, 1979a: Industrial Source Complex (ISC) Dispersion Model User's
Guide. EPA-450/4-79-030.
-156-
-------
U.S. EPA, 1979b: Air Monitoring Strategy for State Implementation
Plans. EPA-450/2-77-010, as updated in 44 FR 27558. (Ambient Air
Quality Monitoring, Data Reporting and Surveillance Reporting, May 10,
1979)
U.S. EPA, 1980a: User's Guide for MPTER. EPA-600/8-80-016.
U.S. EPA, 1980b: Ambient Monitoring Guidelines for Prevention of
Significant Deterioration (PSD). EPA-450/4-80-012.
U.S. EPA, 1980c: Interim Guidelines and Specifications for Preparing
Quality Assurance Project Plans. QAMS-005/80. Office of Monitoring
Systems and Quality Assurance, Washington, DC 20460.
U.S. EPA, 1981a: Receptor Model Technical Series Volume I: Overview of
Receptor Model Application to Particulate Source Apportionment.
EPA-450/4-81-016a .
U.S. EPA, 1981b: Receptor Model Technical Series Volume II: Chemical
Mass Balance. EPA-450/4-81-016b.
U.S. EPA, 1981c: An Evaluation Study for the Industrial Source Complex
(ISC) Dispersion Model. EPA-450/4-8 1-002.
U.S. EPA, 1981d: Interim Procedures for Evaluating Air Quality Models.
Office of Air Quality Planning and Standards, Source Receptor Analysis
Branch, U.S. EPA, Research Triangle Park, NC 27711.
U.S. EPA, 1982a: PTPLU - A Single Source Gaussian Dispersion Algorithm.
EPA-600/8-82-014.
U.S. EPA, 1982b: Quality Assurance Handbook for Air Pollution
Measurement Systems: Volume IV. Meteorological Measurements.
EPA-600/ 4-82-060.
U.S. EPA, 1983a: Receptor Model Technical Series Volume III: User's
Manual for Chemical Mass Balance Model. EPA-45Q/4-83-014.
U.S. EPA, 1983b: Receptor Model Technical Series Volume IV: Summary of
Particle Identification Techniques. EPA-450/4-83-018.
U.S. EPA, 1983c: User's Network for Applied Modeling of Air Pollution
(UNAMAP) , Version 5" (Computer Programs on Tape). PB83-244368. National
Technical Information Service, Springfield, VA 22161.
U.S. EPA, 1983d: Linn County, Iowa Non-Traditional Fugitive Dust Study.
EPA-907/9-83-002 .
U.S. EPA, 1984: Receptor Model Source Composition Library, Draft. EPA-
Watson, J.G. , 1979: Chemical Element Balance Receptor Model Methodology
for Assessing the Sources of Fine and Total Suspended Particulate Matter
in Portland, Oregon. Ph.D. Dissertation, Oregon Graduate Center,
Beaverton, OR.
-157-
-------
Watson, J.G., 1982: Overview of Receptor Model Principles. Presented at
the APCA Specialty Conference on Receptor Models Applied to Contemporary
Air Pollution Problems, Danvers, MA.
Watson, J.G., P.J. Lioy, and P.K. Mueller, 1983: The Measurement
Process: Precision, Accuracy and Validity. Air Sampling Instruments for
Evaluation of Atmospheric Contaminants. 6th Edition. American
Conference of Governmental Industrial Hygienists, Cincinnati, OH.
Watson, J.G., R.C. Henry, J.A. Cooper, and E.S. Macias, 1981: The State
of the Art of Receptor Models Relating Ambient Suspended Particulate
Matter to Sources. Atmospheric Aerosol: Source/Air Quality
Relationships, American Chemical Society Symposium Series Mo. 167,
Washington, D.C.
Willard, H.H., L.L. Merritt, and J.A. Dean, 1974: Instrumental Methods
of Analysis. 5th Edition. D. Van Norstrand Company.
Yocom, J.E., E.T. Brookman, and B.I. Raffle, 1979: Strategies for
Control of Particulate Matter in Allegheny County. Final Report to
Allegheny County Bureau of Air Pollution Control, TRC Environmental
Consultants, Inc.
-158-
-------
APPENDIX A
RESULTS OF SELECTED MODEL VERIFICATION AND EVALUATION STUDIES
PERTAINING TO THE 6 ASSUMPTIONS EMPLOYED BY THE MASS BALANCE MODEL
-------
APPENDIX A
RESULTS OP SELECTED MODEL VERIFICATION AND EVALUATION STUDIES PERTAINING
TO THE 6 ASSUMPTIONS EMPLOYED BY THE MASS BALANCE MODEL
1. Source Composition
Source compositions will vary systematically and randomly. This
*
variability is caused by 1) transformations with transport time between source
and receptor (an example is the volatilization of bromine in auto exhaust;
e.g., Bowman, et al., 1974), 2) differences in the fuel type or operating
processes between similar sources or of the same source in time (an example is
the review of lead concentrations in auto exhaust from different automobiles
using different fuels by Pierson and Brachaczek, 1976), and 3) uncertainties
of the measurement process.
Watson (1979) and Currie, et al., (1983) introduced systematic errors into
source compositions to evaluate the effects on source contribution
calculations and found:
The error in the estimated source contributions due to biases
(i.e., proportional increases or decreases) in the source
compositions is in direct proportion to the magnitude of that bias
(Watson, 1979).
For random errors associated with estimates of average source
compositions, the magnitude of the source contribution errors
decreases as the number of components increases (Watson, 1979).
In comparing a number of mass balance solutions on simulated data
in which random errors and biases were introduced to the source
compositions, Currie, et al., (1983) found "Only the effective
variance [solution] explicitly included the effect of source
profile uncertainties, but these were treated as random rather
than as systematic error components."
A-l
-------
2. Components Add Linearly
No studies have been performed to evaluate deviations from this
assumption. While the deviations from this assumption are generally
considered to be small, the conversion of gases to particles and
reactions between particles are not inherently linear processes (e.g.,
Priedlander, 1977).
3. All Sources Have Been Identified
Watson (1979) systematically increased the number of sources
contributing to his simulated data from 4 to 8 contributors while solving
the mass balance equations assuming only four sources. He also examined
the case of including more sources in the least squares solution than
those which actually contribute. The results wera:
Underestimating the number of sources has little effect on the
calculated source contributions if the prominant properties
contributed by the source are excluded from the solution.
When the number of sources is underestimated and when prominent
properties of the omitted sources are included in the calculation
of source contributions, the contributions of sources with
properties in common with the omitted sources are overestimated.
* When source types which are actually present are excluded from the
effective variance least squares solution, ratios of calculated to
measured concentrations are often outside of the 0.5 to 2.0 range,
and the sum of the source contributions is much less than the
total measured mass. The low calculated/measured ratios indicate
which source compositions should be included.
When the number of sources is overestimated, the sources which are
not actually present yield contributions less than their
precisions if their properties are significantly distinct from
other sources. The over-specification of sources decreases the
precisions of the true source contribution estimates.
A-2
-------
4. Number of Sources Less Than Number of Components
It is likely that the number of individual sources contributing to
receptor concentrations is much larger than the number of properties which can
be measured. It is therefore necessary to group sources into source types of
similar compositions such that this assumption is met. For most commonly
measured aerosol properties, meeting Assumption 5 practically defines these
groupings. Henry (1977) speculates that the linear programming solution might
work if the number of sources is greater than the number of components, but he
offers no proof. None of the least squares solutions will work if this
assumption is not met.
5. Compositions are Linearly Independent
Watson (1979) examined effects of deviations from this assumption with
simulated data while Gordon, et al., (1981) studied it with ambient
measurements. Henry (1982) has devised an analytical method of determining
the degree of linear dependence in typical mass balance applications for which
only a few tests are available. His algorithm could be incorporated into the
routine mass balance model to identify linear independence at given precision
levels for all combinations of sources and components. The results of these
studies show:
With most commonly measured components (i.e., ions, elements,, and
carbon) and source types (e.g., motor vehicle, crustal, residual
oil, sea salt, steel production, wood burning, and various
industrial processes), from five to seven sources are linearly
independent of each other. Some of the source compositions used
in the solution may be better expressed as linear combinations of
measured source compositions (Henry, 1982).
Henry (1982) determined the modified source compositions which
would give the same results as a ridge regression solution and
noted that: "The apparent ability of ridge regression to solve
the multicollinearity problem is seen to be based on implicitly
changing the [mass balance] problem to one which is not physically
meaningful."
A-3
-------
Gordon, et al., (1981) found instabilities in the ordinary
weighted least squares solutions when they removed elemental
concentrations which were known to be unique to certain source
types. Using simulated data with known perturbations of 0 percent
to 20 percent, Watson (1979) found: "In the presence of likely
uncertainties, sources such as urban dust and continental
background dust cannot be adequately resolved by least squares
fitting, even though their compositions are not identical.
Several nearly unique ratios must exist for good separation.
6. Measurement Errors Random and Uncorrelated
These least squares solutions methods are derived from maximum likelihood
theory which requires this assumption. In reality, we 'know very little about
the distribution of errors for the source compositions and the ambient
concentrations. For small errors (i.e. <10 percent) the actual distribution
may not be important, but for large errors it probably is; a symmetric
distribution becomes less probable as the coefficient of variation of the
measurement increases. Though evaluation of deviation from this assumption
could be undertaken by generating simulated data perturbed by random numbers
drawn from lognonnal, uniform, and other distributions, these tests have never
been performed.
A-4
-------
APPENDIX B
SOURCE COMPOSITION REFERENCES
-------
APPENDIX B
SOURCE COMPOSITION REFERENCES
Air Conditioners;
Buchnea, D. & Buchnea, A. (1974) "Air Pollution by Aluminum Compounds
Resulting From Corrosion of Air Conditioners". ES&T, 8, 752.
Animal Feed and Wastes:
Capar, S.G., Tanner, J.T., Friedman, M.H., and Boyer, K.W. (1978)
"Multielement Analysis of Animal Feed, Animal Wastes, and Sewage Sludge".
ES&T, 12, 785.
Agricultural Burning;
Core, J.E. and Terraglio, F.P. (1978) "Field and Slash Burning Particulate
Characterization: The Search for Unique Natural Tracers". Annual Meeting of
Pacific Northwest International Section of APCA, Portland, Oregon.
Shum, Y.S. and Loveland, W.D. (1974) "Atmospheric Trace Element Concentrations
Associated with Agricultural Field Burning in the Willamette Valley of
Oregon". Atmospheric Environment, £, 645.
Auto Exhaust:
Ganley, J.T. and Springer, G.S. (1974) "Physical and Chemical Characteristics
of Particulates in Spark Ignition Engine Exhaust". ES&T, 8, 340.
Gillette, D.A. and Winchester, J.W. (1972) "A Study of Aging of Lead Aerosols
- - I". Atmospheric Environment, (5, 443,
Gillette, D.A. (1972) "A Study of Aging of Lead Aerosols II". A Numerical
Model Simulating Coagulation and Sedimentation of a Leaded Aerosol in the
Presence of an Unleaded Background Aerosol". Atmospheric Environment, 6, 451.
Habibi, K. (1973) "Characterization of Particulate Matter in Vehicle
Exhausts". ES&T, 7, 223.
Harrison, R.M. and Sturges, W.T. (1983) "The Measurement and Interpretation of
Br/Pb Ratios in Airborne Particles". Atmospheric Environment, 17, 311.
Hasanen, E., et al. "Benzene, Toluene and Xylene Concentrations in Car
Exhausts and in City Air". Atmospheric Environment, 15, 1755.
Hirschler, D.A. and Gilbert, L.F. (1964) "Nature of Lead in Automobile
Exhaust". Archives of Environmental Health, 8, 297.
Hirschler, D.A. et al. (1957) "Particulate Lead Compounds in Automobile
Exhaust Gas". Industrial and Engineering Chemistry, 49, 1131.
Holiday, E.P. and Parkinson, M.C. (1978). "Another Look at the Effects of
Manganese Fuel Additive (MMT) on Automobile Emission". APCA Meeting Houston.
B-l
-------
SOURCE COMPOSITION REFERENCES (continued)
Larsen, R.I. (1966) "Air Pollution from Motor Vehicles". Annals of the New
York Academy of Sciences, 136, 275.
McKee, H.C. (1970) "Discussion: Characterization of Particulate Lead in
Vehicle Exhaust: Experimental Techniques". ES&T, 4, 253.
McKee, H.C. and McMahan, W.A. (1960) "Automobile Exhaust Particulates - Source
and Variation". JAPCA, 10, 456.
Moran, J.B. et al. (1972) "Effect of Fuel Additives on the Chemical and
Physical Characteristics of Particulate Emissions in Automotive Exhaust".
U.S. EPA, Report EPA-R2-72-066.
Nielsen, T. (1979) "Determination of Polycyclic Aromatic Hydrocarbons in
Automobile Exhaust by Means of High-Performance Liquid Chromatography with
Fluoresence Detection". Journal of Chromatography, 170, 147.
Ninomeya, J.S. et al. (1970) "Automotive Particulate Emissions". Second
International Clean Air Conference, p. 663.
Ondov, J.M. et al. (1982) "Trace Element Emissions on Aerosols". ES&T, 16,
318.
Pierson, W.R. and Brachaczek, W.W. (1976) "Particulate Matter Associated with
Vehicles on the Road". SAE Automotive Engineering Congress and Exposition,
Detroit, Mich., Feb 23-27, 1976.
Pierson, W.R. et al. (1978) "Methylcyclopentadionyl Manganese Tricarbonyl
Effect on Manganese Emissions from Vehicles on the Road". JAPCA, 28.
Sampson, K.E. and Springer, G.S. (1973) "Effects of Exhaust Gas Temperature
and Fuel Composition on Particulate Emission from Spark Ignition Engines".
ES&T, 7, 55.
Ter Haar, G.L. and Boyard, M.A. (1971) "Composition of Airborne Lead
Particles" Nature, 232, 553.
Ter Haar, G.L. et al. (1972) "Composition, Size and Control of Automotive
Exhaust Particulates". JAPCA, 22, 39.
Von Lehmden, D.J. et al. (1974) "Determination of Trace Elements in Coal, Fly
Ash, Fuel Oil and Gasoline A Preliminary Comparison of Selected Analytical
Techniques". Analytical Chemistry, 46, 239.
Wilson, W.E. et al. (1973) "The Effect of Fuel Composition on Atmospheric
Aerosol Due to Auto Exhaust". JAPCA, 23, 949.
Carbon Black;
Serth, R.W. and Hughes, T.W. (1980) "Polycyclic Organic Matter (POM) and Trace
Element Contents of Carbon Black Vent Gas". ES&T, 14, 298.
B-2
-------
SOURCE COMPOSITION REFERENCES (continued)
Cement Production;
Cooper, J.A. at al. (1976) "Analysis of Portland Cement, Clinker, Raw Mix, and
Associated Ceramic Materials Using an Energy Dispersive X-Ray Fluorescence
Analyzer With Inter-Element Corrections". Advances in X-Ray Analysis, Vol.
19, Dubuque, Iowa, Kindall/Hunt Publishing Co.
Coal-Fired Boilers:
Bennett, R.L. and Knapp, K.T. (1978) "Sulfur and Trace Metal Emissions From
Combustion Sources". April 24-26, Southern Pines, N.C. Workshop on
Measurement Technology and Characterization of Primary Sulfur Oxides Emission
from Combustion Sources.
Block, Chantal and Dams, R. (1976) "Study of Fly Ash Emission During
Combustion of Coal". ES&T, 10, 1011.
Coles, D.G. et al. (1979) "Chemical Studies of Stack Fly Ash from a Coal-Fired
Power Plant" Environmental Research and Technology, 13, 455.
Conway, R.P. (1980) Chemical and Physical Characterization of Sulfates
Associated with Coal Fly Ash, Ph.D. Dissertation, Colorado State University,
Ft. Collins, CO.
Fisher, G.L. et al. (1978) "Physical and Morphological Studies of
Size-Classified Coal Fly Ash". ES&T, 12, 447.
Germani, M.S. (1980) Selected Studies of Four High Temperature Air Pollution
Sources Ph.D. Dissertation, Univ. of Maryland, College Park, MD.
Gladney, E.S. et al. (1976) "Composition and Size Distribution of In-Stack
Particulate Material at a Coal-Fired Power Plant". Atmospheric Environment,
10, 1071.
Homolya, J.B. and Cheney, J.L. (1978) "An Assessment of Sulfuric Acid and
Sulfate Emissions from the Combustion Fossil Fuels". Workshop Proceedings on
Pjcimary Sulfate Emissions from Combustion Sources Volume 2, Characterizations.
EPA Publication 600/9-78-036.
Klein, U.K. et al. (1975) "Pathways of Thirty-Seven Trace Elements Through a
Coal-Fired Power Plant". ES&T, 9, 973.
Nadkarni, R.A. (1975) "Multielement Analysis of Coal and Coal Fly Ash
Standards by Instrumental Neutron Activation Analysis". Radiochemical and
Radioanalysis Letters, 21, 161.
Ondov, J.M. et al. (1979) "Emissions and Particle Size Distributions of Minor
and Trace Elements at Two Western Coal-Fired Power Plants Equipped with
Cold-Side Electrostatic Precipitators". ES&T, 13, 947.
B-3
-------
SOURCE COMPOSITION REFERENCES (continued)
Ondov, J.M. et al. (1981) "Elemental Composition of Atmospheric Fine Particles
Emitted from Coal Burned in a Modern Electric Power Plant Equipped with a Flue
Gas Desulfurization System". Atmospheric Aerosol, Source/Air Quality
Relationships, edited by E.S. Macios and P.K. Hopke, American Chemical Society
Symposium Series 167, Washington, D.C.
Richards L.W. (1981) "The Chemistry Aerosol Physics, and Optical Properties of
a Western Coal-Fired Power Plant Plume". Atmospheric Environment, 15, 2111.
Small, J.A. (1976) An Elemental and Morphological Characterization of the
Emissions from the Dickerson and Chalk Point Coal-Fired Power Plants Ph.D.
Dissertation, University of Maryland, College Park, MD.
Small, M. (1979) Composition of Particulate Trace Elements in Plumes from
Industrial Sources Ph.D. Dissertation, University of Maryland, College Park,
MD.
Small, M. et al. (1981) "Airborne Plume Study of Emissions from the Processing
of Copper Ores in Southeastern Arizona". SS&T, 15, 293.
Smith, R.D. et al (1979) "Concentration Dependence Upon Particle Size of
Volatized Elements in Fly Ash". ES&T, 13, 553.
Surprenant, N.F. (1981) "Emissions Assessment of Conventional Stationary
Combustion Systems: Volume IV: Industrial Combustion Sources".
EPA-60Q/7-81-003C, Research Triangle Park, NC.
Taylor, D.D. and Flanagan, R.C. (1980) "Aerosols from a Laboratory Pulverized
Coal Combustor". Atmospheric Aerosol: Source/Air Quality Relationships,
edited by E.S. Kacias and P.K. Hopke, American Chemical Society Symposium
Series, No. 167, Washington, D.C.
Ulrich, G.D. (1976) "An Investigation of the Mechanism of Fly-Ash Formulation
in Coal-Fired Utility Boilers". U.S.-ERDA Report FE-2205-1.
Coke Ovens
Barret, R.E. et al. (1977) "Sampling and Analysis of Coke Oven Door
Emissions". EPA-600/2-77-213, Research Triangle Park, NC.
Copper Smelters
Germani, M.S. (1980) Selected Studies of Four High Temperature Air Pollution
Sources Ph.D. Dissertation, University of Maryland, College Park, MD.
Schwitzgebel, K. et al. (1978) "Trace Element Study at a Primary Copper
Smelter". EPA-600/2-78-065a, Research Triangle Park, NC.
Small M. (1979) Composition of Particulats Trace Elements in Plumes from
Industrial Sources Ph.D. Dissertation, University of Maryland, College Park,
MD.
B-4
-------
SOURCE COMPOSITION REFERENCES (continued)
Zoller, W.H. et al. (1978) "Atmospheric Trace Elements Emissions from Copper
Smelters". Presented at Division of Environmental Chemistry, American
Chemical Society, Miami, Fla.
Cotton Gin:
Lee, R.E. (1975) "Concentration and Size of Trace Metal Emissions from a Power
Plant, a Steel Plant, and a Cotton Gin". SS&T, 9, 543.
Diesel Exhaust:
Prey, J.W. and Corn, M. (1967) "Physical and Chemical Characteristics of
Particulates in a Diesel Exhaust". American Industrial Hygiene Association
Journal, 28, 468.
Schreck, R.M. et al (1978) "Characterization of Diesel Exhaust Particulate for
Mutagenic Testing". APCA Meeting, Houston.
Yergey, J.A. (1981) Chemical Characterization of Organic Adsorbates on Diesel
Particulate Matter, Ph.D. Dissertation, Pennsylvania State University.
NTIS (1982) "Diesel Exhaust Emissions. 1974-May, 1982". (Citations from the
American Petroleum Institute Data Base).
Glass Production;
Mamuro, T. et al. (1979) "Elemental Compositions of Suspended Particles
Released in Glass Manufacture". Annual Report of the Radiation Center of
Osaka Prefecture, 20. 29.
Jet Aircraft Exhaust:
Fordyce, J.S. and Shebley (1975) "Estimate of Contribution of Jet Aircraft
Operations to Trace Element Concentrations at or Near Airports". JAPCA, 25,
721.
Mamuno, T. et al. (1973) "Activation Analysis of Particulates Emitted from
Aircraft Jet Engines". Annual Report of the Radiation Center of Osaka
Prefecture, 14, 7.
Kraft Paper Mills;
Augustine, E. (1973) Airborne Sampling of Particles Emitted to the Atmosphere
from Kraft Paper Mill Processors and Their Characterization by Electron
Microscopy (Corvallis, OR: OSU Air Resources Center.)
Keith, L. (1976) "Identification of Organic Compounds in Unbleached Treated
Kraft Paper Mill Wastewaters". ES&T, 6, 555.
B-5
-------
SOURCE COMPOSITION REFERENCES (continued)
Nitrogen Containing Fuels;
Dubay, G.R. and Kites, R.A. (1978) "Cyano-arenes Produced by Combustion of
Nitrogen-Containing Fuel". ES&T, 12, 965.
Oil-Fired Boilers
Bennet, R.L. and Knapp, K.T. (1978) "Sulfur and Traca Metal Emissions from
Combustion Sources". April 24-26, Southern Pines, NC Workshop on Measurement
Technology and Characterization of Primary Sulfur Oxides Emission from
Combustion Sources.
Dietz, R.N. et al. (1978) "Operating Parameters Affecting Sulfate Emissions
from an Oil-Fired Power Unit". Workshop Proceedings on Primary Sulfate
Emissions from Combustion Sources, Volume 2 Characterization (EPA Publication,
500/9-78-0206).
Homolya, J.B. and Cheney, J.L. (1978) "An Assessment of Sulfuric Acid and
Sulfate Emissions from the Combustion of Fossils Fuels". Workshop Proceedings
on Primary Sulfate Emissions from Combustion Sources, Volume 2
Characterization (EPA Publication, 500/9-78-0206).
Mroz, E.J. (1976) The Study of the Elemental Composition of Particulate
Emissions From an Oil-Fired Power Plant Ph.D. Thesis (University of Maryland,
College Park, MD).
Mamuro, T. (1979) "Elemental Compositions of Suspended Particles Released from
Various Boilers". Annual Regort of the Radiation Center of Osaka Prefecture,
20, 29.
Van Lehmden, D.J. et al. (1974) "Determination of Trace Elements in Coal, Fly
Ash, Fuel Oil, and Gasoline - - A Preliminary Comparison of Selected
Analytical Techniques". Analytical Chemistry, 46, 239.
Zoller, W.M. et al. (1973) "The Sources and Distributions of Vanadium in the
Atmosphere". Trace Elements in the Environment Ed. E.L. Kothny (Washington,
D.C.: American Chemical Society).
Paste Plants;
Bjorseth A. and Lunde G. (1977) "Analysis of the Polycyclic Aromatic
Hydrocarbon Content of Airborne Particulate Pollutants in a Soderberg Paste
Plant". American Industrial Hygiene Association Journal, 38, 224.
Petroleum Pitch:
Grienke, R.A. and Lewis, I.C. (1975) "Development of a Gas Chromatographic
Ultraviolet Absorption Spectrometric Method of Monitoring Petroleum Pitch
Volatiles in the Environment". Analytical Chemistry, 47, 2151.
B-6
-------
SOURCE COMPOSITION REFERENCES (continued)
Refuse Combustion:
Campbell, W.J. et al. (1977) "Determination of Trace and Minor Elements in the
Combustible Fraction of Urban Refuse". Methods and Standards for
Environmental Measurement, (NBS Special Publication 264: Washington, D.C.), p.
157.
Clement, R.E. (1981) Development of Analytical Methodology for Airborne
Particulate Municipal Incinerator Ash, Ph.D. Dissertation, University of
Waterloo, Canada.
Eicemen, G.A. (1979) "Analysis of Fly Ash from Municipal Incinerators for
Trace Organic Compounds" Analytical Chemistry, 51, 2343.
Germani, M.S. (1980) Selected Studies of Four High Temperature Air Pollution
Sources Ph.D. Dissertation, University of Maryland, College Park, MD.
Greenberg, R.R. (1976) A Study of Trace Elements Emitted on Particles from
Municipal Incinerators Ph.D. Thesis (University of Maryland, College Park, MD).
Greenberg, R.R. et al. (1978) "Composition of Particles Emitted from the
Nicosia Municipal Incinerator". ES&T, 12, 1329.
Law, S.L. and Gordon, G.E. (1979) "Sources of Metals in Municipal Incinerator
Emissions". ES&T, 13, 433.
Mamuro, T. and Mizohato, A. (1978) "Elemental Composition of Suspended
Particles Released in Refuse Incineration". Annual Report of the Radiation
Center of Osaka Prefecture, 19, 15.
Shen, T.T. (1978) "Air Pollutants from Sewage Sludge Incineration". APCA
Meeting, Houston.
Road Dust;
Blumer, M. (1977) "Polycyclic Aromatic Hydrocarbons in Soils of a Mountain
Valley: Correlation with Highway Traffic and Cancer Influence". ES&T, 11,
1083.
Ciacco, L.L. (1974) "Composition of Organic Constituents in Breathable
Airborne Particulate Matter Near a Highway". ES&T, 2, 935.
Soil:
Rahn, K.A. (1976) "Silicon and Aluminum in Atmospheric Aerosols: Crust-Air
Fractionation?" Atmospheric Environment, 10, 597.
Taylor, S.R. (1964) "Abundance of Chemical Elements in the Continental Crust:
A New Table". Geochemica et Cosmochimia Acta, 28, 1273.
B-7
-------
SOURCE COMPOSITION REFERENCES (continued)
Thomae, S.C. (1977) Size and Composition of Atmospheric Particles in Areas
Near Washington Ph.D./ Thesis, University of Maryland, College Park, MD.
Steel Production;
Jacko, R.B. and Neuendorff (1977) "Trace Metal Particulate Emission Test
Results from a Number of Industrial and Municipal Point Sources". JAPCA, 21,
989.
Lee, R.E. et al. "Concentration and Size of Trace Metal Emissions from a Power
Plant, A Steel Plant and a Cotton Gin". ES&T, 9, 643.
Mamuro, T. et al. (1979) "Elemental Compositions of Suspended Particles
Released from Iron and Steel Works". Annual Report of_ the Radiation Center of
Osaka Prefecture, 20, 19.
Tire Dusts:
Dennis, M.L. (1974) "Rubber Dust from the Normal Wear of Tires". Rubber-
Chemistry and Technology, 47, 1011.
Pierson, W.R. and Brachaczek, W.W, (1974) "Airborne Particulate Debris from
Rubber Tires". Ecology Symposium, ACS Rubber Division, Toronto, Ontario May
7-10, 1974.
Pierson, W.R. and Brachaczek, W.W. (1975) "Airborne Particulate Debris from
Rubber Tires". Presented at the Conference on Environmental Aspects of
Chemical Use in Rubber Processing Operations, University of Akron, March
12-14, 1975.
Pierson, W.R. and Brachaczek, W.W. (1975) "In-Traffic Measurement of Airborne
Tire-Wear Particulate Debris". JAPCA, 25, 404.
Veneer Dryers;
Cronn, D.R. et al. (1983) "Chemical Characterization of Plywood Veneer Dryer
Emissions". Atmospheric Environment, 17, 201.
Volcanoes;
German, M.S. (1980) Selected Studies of Four High Temperature Air Pollution
Sources Ph.D. Dissertation, University of Maryland, College Park, MD.
Welding;
Akellson, K. et al. (1974) "Elemental Abundance Variation with Particle Size
in Aerosols from Welding Operations". Proceedings of the Second International
Conference on Nuclear Methods in Environmental Research (University of
Missouri), p. 385.
B-8
-------
SOURCE COMPOSITION REFERENCES (continued)
Wood Stoves:
Butcher, S.S. and Sorenson, E.M. (1979) "A Study of Wood Stove Particulate
Emissions". JAPCA, 29, 724.
Cooper, J.A. (1980) "Environmental Impact of Residential Wood Combustion
Emissions and its Implications". JAPCA, 30, 855.
B-9
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse befoie completing)
EPA-450/4-84-020
3. RECIPIENT'S ACCESSION-NO.
TITLE AND SUBTITLE Receptor Model Technical Series, Vol. V:
Source Apportionment Techniques And Considerations In
Combining Their Use
5. REP
6. PERFORMING ORGANIZATION CODE
7 AUTHORIS)
Michael K. Anderson, et al.
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
TRC Environmental Consultants, Inc.
East Hartford, CT 06108
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
Office Of Air Quality Planning And Standards
U. S. Environmental Protection Agency
Research Triangle, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
(MD 14)
14. SPONSORING AGENCY CODE
IS SUPPLEMENTARY NOTES
EPA Project Officer: Thompson G. Pace
16. ABSTRACT
This volume 1) discusses models which identify source contributions to receptor
concentrations, their input data, the assumptions on which they are based, and the
effects of typical deviations from those assumptions; 2) identifies measurements
which these models require, their availability, the additional assumptions imposed
by these measurements, and the effect of their precision and accuracy on modeling
results; and 3)presents approaches, pertaining to three levels of analysis detail,
for the optimum combinations of models and measurements in practical situations, and
illustrates these protocols with case studies.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lOENTIFIERS/OPEN ENDED TERMS C. COSATI Field/Group
13. DISTRIBUTION STATEMENT
19. SECURITY CLASS (This Reporlj
21. NO. OF PAGES
192
20. SECURITY CLASS (This page)
22. PRICE
EPA Form 2220-1 (9-73)
-------
yr-
U.S. Environmental Protection Agency
Region V, Library
230 South Dearborn Street
Chicago, Illinois 60604
-------