United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park NC 27711
EPA-450/3-78-113
September 1978
Air
oEPA
Digest of Ambient
Particulate Analysis
and Assessment Methods
-------
EPA-450/3-78-113
Digest of Ambient Particulate
Analysis and Assessment Methods
by
James A. Throgmorton and Kenneth Axetell
PEDCo Environmental, Inc.
2420 Pershing Road
Kansas City, Missouri 64108
Contract No. 68-02-2603
EPA Project Officer: Thompson G. Pace
Prepared for
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air, Noise, and Radiation
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
September 1978
-------
This report is issued by the Environmental Protection Agency to report
technical data of interest to a limited number of readers. Copies are
available free of charge to Federal employees, current contractors and
grantees, and nonprofit organizations in limited quantities from the
Library Services Office (MD35), Research Triangle Park, North Carolina
27711; or, for a fee, from the National Technical Information Service,
5285 Port Royal Road, Springfield, Virginia 22161.
This report was furnished to the Environmental Protection Agency by
PEDCo Environmental, Inc., 2420 Pershing Road, Kansas City, Missouri
64108-, in fulfillment of Contract No. 68-02-2603, Task Order No. 17. The
contents of this report are reproduced herein as received from PEDCo
Environmental, Inc. The opinions, findings, and conclusions expressed
are those of the author and not necessarily those of the Environmental
Protection Agency. Mention of company or product names is not to be
considered as an endorsement by the Environmental Protection Agency.
This document, in draft form, was the reference text for The Workshop on
Ambient Particulate Analysis and Assessment Methods, Raleigh, North
Carolina, July 19-20, 1978, Thompson G. Pace, Workshop Chairman.
Publication No. EPA-450/3-78-113
ii
-------
CONTENTS
Page
1.0 INTRODUCTION 1
2.0 ANALYZING TEMPORAL PATTERNS OF SUSPENDED 6
PARTICULATES
2.1 Long-Term Trends 6
2.2 Seasonal and Monthly Patterns 13
2.3 Weekday/Weekend and Daily Patterns 18
2.4 Diurnal Variation 20
2.5 Emission Pattern/Air Quality Relationships 25
3.0 ANALYZING SPATIAL PATTERNS OF SUSPENDED 30
PARTICULATES
3.1 Stratifying Concentrations by Site Environment 30
3.2 Intersite TSP Correlations 34
3.3 Pollution, Dosage, and Gradient Roses 38
3.4 Upwind/Downwind Relationships 45
4.0 ASSESSING THE EFFECT OF METEOROLOGICAL VARIATIONS 49
4.1 Correlation and Regression Techniques 49
4.2 Decision Tree Analysis 54
4.3 Analysis of Precipitation 58
4.4 Analysis of Wind Speed 63
4.5 Trajectory Analysis 68
5.0 ANALYZING EMISSIONS DATA 72
5.1 Emission Inventorying 72
5.2 Microinventorying 77
5.3 Diffusion Modeling 87
6.0 INTERPRETING CHEMICAL, ELEMENTAL, MORPHOLOGICAL 102
DATA
6.1 Temporal, Spatial, and Meteorologically- 102
Affected Patterns (TSM Techniques)
6.2 Enrichment Factor (E) 108
6.3 Chemical Element Balance (CEB) 114
6.4 Interspecies Correlations 127
111
-------
Paqe
6.5 Pattern Recognition 132
6.6 Factor Analysis 136
6.7 Interpretation of Morphological Data 146
7.0 INTERPRETING PARTICLE SIZE DATA 154
7.1 Frequency Distribution Comparisons 157
7.2 Species-Specific Size Distribution 159
Comparisons
8.0 DESIGNING STUDY AND INTERPRETING RESULTS 165
8.1 Selecting Techniques 165
8.2 Applying Techniques and Interpreting Results 170
8.3 Example Applications 173
9.0 REFERENCES 183
10.0 APPENDIX
A Techniques for Elemental Analysis 193
11.0 INDEX 196
iv
-------
FIGURES
No. Page
1-1 Ninety-Five Percent Confidence Belts for 5
Correlation Coefficient
2-1 Long-Term Trends at Composite Sites 9
2-2 Long-Term Trends, Seattle, Washington 12
2-3 Seasonal Patterns at Composite Sites 13
2-4 TSP Quarterly Averages, Quarterly Meteorology for 17
the Phoenix Area
2-5 Daily Variations in Traffic and TSP Levels 21
2-6 Comparison of TSP Concentration and Traffic 24
Volume at Broad and Spruce Streets
2-7 Traffic Volume versus Concentration for McGee 27
Street Sites, Kansas City
3-1 Locations of Sites To Be Intercorrelated 37
3-2 Locations of Sites and Pollution Roses 43
3-3 Gradient Roses " 44
3-4 Aggregate Storage Sampling Sites 47
4-1 AID Decision Tree for Forrest City, AR 59
4-2 Effect of Rainfall in Reducing TSP Concentrations 64
at Two Sites in Birmingham
4-3 Plots of Average TSP versus Wind Speed as a Function 67
of Wind Direction
4-4 Air Flow Trajectories for 10 Nonprecipitation 71
Days Terminating in Albany
5-1 Location of Major Sources 79
-------
No. Page
5-2 Converting Microinventory Area to Diffusion 86
Model Grid
5-3 One Mile Radius around Sampling Site 89
5-4 One-Quarter Mile Radius around Sampling Site 90
5-5 County X Particulate Air Quality in 1975 and 98
1985
6-1 Normalized Diurnal Variation in TSP and Selected 106
Elements
6-2 Enrichment Factors for Species in Desert 115
Background and Urban Particulate Matter
6-3 Enrichment Factors for Species in Desert 116
Background and Urban Small Particle
(<2 urn) Particulate Matter
6-4 Enrichment Factors for Species in Desert 116
Background and Urban Large Particle
(>2 um) Particulate Matter
6-5 Source Contributions to St. Louis Aerosol 126
6-6 Dendrogram of Feature Clustering for Desert 134
Urban Particulate Matter
7-1 Normalized Frequency Plots of Number, 156
Surface, and Volume Distributions for the
Grand Average October 1977 Measurements
at Denver's City Maintenance Yard
7-2 Particle Size Distribution for Alaskan and 160
Urban Sites
8-1 Screening Techniques for Example City No. 1 176
8-2 Screening Techniques for Example City No. 2 180
vi
-------
TABLES
No. Page
2-1 Statistical Methods Useful in Analyzing 7
Temporal Patterns
2-2 Effect of Traffic Volume on Concentrations 29
3-1 Example Site Classification 32
3-2 Example Stratification of TSP Concentrations 33
by Site Type
3-3 Example Correlations of Meteorological Data 33
for Site Types
3-4 Linear Correlation Coefficients between 24-h 39
TSP Levels at Philadelphia Monitoring Stations
in 1974, 1975, and 1976
3-5 Example Upwind/Downwind Concentrations 46
4-1 Meteorological Variables 50
4-2 Stepwise Multiple Regression Analysis with 1974 Data 55
4-3 Percent Variance Explained by AID Compared to 60
Percent Variance Explained by Multiple Regressions
against All 18 Meteorological Variables
5-1 Emission Source Categories 74
5-2 NEDS TSP Emissions for Philadelphia AQCR 78
5-3 Information Obtained in Presurvey 82
5-4 Land Use Categories and Classification Criteria 84
5-5 Description of Microinve'ntory Site 88
5-6 Microinventory Point Source Summary 91
5-7 Microinventory Area Source Summary 92
-------
No. Page
fe-8 Composite of Computed Air Quality for County X 99
from AQDM
"5-9 Source Contribution Analysis for County X 101
6-1 Dependence of Chemical Composition on 109
Meteorological Parameters (Urban Location)
6-2 Dependence of Chemical Composition on 110
Meteorological Parameters (Rural Location)
6-3 Sources of Primary Particulates in St. Louis 123
6-4 Source Concentrations of Particulate Matter 124
6-5 Results of Chemical Element Balance for 125
Elemental Concentrations
6-6 Linear Correlation Coefficients for Urban 131
Samples, Tucson, Arizona
6-7 Factor Loadings for an Urban Site 142
6-8 Eigenvalues of Correlation Matrix for an Urban Site 143
6-9 Element Loadings on Individual Factors and Possible 144
Explanations for Factor Significance
6-10 Composite Summary of Filter Analyses 148
6-11 Microscopic Analysis Results, Site 115004 152
7-1 Size Ranges Sampled by Current Methods 155
7-2 Classification of Airborne Trace Elements 163
According to Size
8-1 Summary of Techniques 167
8-2 Effectiveness Rankings 168
8-3 Cost-Effectiveness Rankings 170
8-4 Applying Techniques to Example City No. 1 177
8-5 Applying Techniques to Example City No. 2 181
A-l Techniques for Elemental Analysis 194
-------
ABBREVIATIONS FOR CHEMICAL ELEMENTS
Abbreviation
Ac
Al
Sb
Ar
As
At
Ba
Be
Bi
B
Br
Cd
Ca
C
Ce
Cs
Cl
Cr
Co
Cu
Dy
Er
Eu
F
Fr
Gd
Ga
Ge
Au
Hf
He
Ho
H
In
I
Ir
Fe
Kr
La
Pb
Li
-Lu
Mg
Mn
Hg
Mo
Chemical element
actinium
aluminum
antimony
argon
arsenic
astatine
barium
beryllium
bismuth
boron
bromine
cadmium
calcium
carbon
cerium
cesium
chlorine
chromium
cobalt
copper
dysprosium
erbium
europium
fluorine
francium
gadolinium
gallium
germanium
gold
hafnium
helium
holmium
hydrogen
indium
iodine
iridium
iron
krypton
lanthanum
lead
lithium
lutetium
magnesium
manganese
mercury
molybdenum
Abbreviation
Nd
Ne
Ni
Nb
N
Os
0
Pd
P
Pt
Po
K
Pr
Pm
Pa
Ra
Rn
Re
Rh
Rb
Ru
Sm
Sc
Se
Si
Ag
Na
Sr
S
Ta
Tc
Te
Tb
Tl
Th
Tm
Sn
Ti
W
U
V
Xe
Yb
Y
Zn
Zr
Chemical element
neodymium
neon
nickel
niobium
nitrogen
osmium
oxygen
palladium
phosphorus
platinum
polonium
potassium
praseodymium
promethium
protactinium
radium
radon
rehnium
rhodium
rubidium
ruthenium
samarium
scandium
selenium
silicon
silver
sodium
strontium
sulfur
tantalum
technetium
tellurium
terbium
thallium
thorium
thulium
tin
titanium
tungsten
uranium
vanadium
xenon
ytterbium
yttrium
zinc
zirconium
IX
-------
1.0 INTRODUCTION
The U.S. Environmental Protection Agency (EPA), state and
local air pollution control agencies, and independent research
scientists have used a wide variety of techniques for analyzing
and evaluating samples and data related to particulate matter
present in urban and rural atmospheres. Although descriptions
and examples of those techniques are available in the academic
and professional literature, it is difficult for a potential user
to learn about more than one technique at a time. As a result,
he may apply one technique to a given problem, not knowing that
another technique is more appropriate.
There is a need, then, for a compendium of techniques that
have been or can be used to analyze a suspended particulate
problem. Such a compendium should provide a brief and clear
description of each available technique, provide references which
will enable the user to obtain more information concerning the
technique and its application to air pollution control problems,
indicate the types of problems to which the technique can apply,
explain ways in which the various techniques can interrelate,
estimate the resources needed to implement each technique, and
provide a good, clear example of how the technique has been
applied in the literature. This is one set of purposes that will
be met by this digest.
A second purpose of this digest is to provide the user with
guidelines as to the use of these techniques in combination in
order to resolve a specific type of problem. In order to do
this, information is presented which summarizes and compares
available techniques in terms of their problem applicability,
resource requirements, and interrelationships.
-------
Three general types of problems are addressed: character-
izing the aerosol in terms of spatial extent, temporal patterns,
and composition; qualitatively identifying the source categories
(e.g., fuel combustion) that produce significant total suspended
particulate (TSP); and quantitatively defining the TSP impact of
specific sources (e.g., a large coal--fired power plant).
Resource requirements are assessed in terms of manpower,
skill, computer access, and data. Manpower requirements are
estimated for each individual technique and are assigned descrip-
tors of low, moderate, and high. These estimates do not include
the time required to produce the raw field or laboratory data.
Rather they pertain only to the time required to manipulate,
process, keypunch, and/or interpret the available data. Computer
requirements are specified in terms of whether a computer is not
necessary, optional, or necessary. The reader should be aware of
the interdependence of these two categories. Skill requirements
are described as being low, moderate, or high. Lastly, the types
of data required to implement each technique are identified. It
is assumed that the reader is aware of the different types of
instrumentation or laboratory methods that will produce the data
indicated.
Following the discussion of resources is a brief overview of
the ways in which the subject technique relates to the other
techniques presented in this digest. Three types of relation-
ships are identified: combinable, meaning that the subject
technique can be combined with another one to produce what is in
effect a third technique; parallel, meaning the techniques pro-
vide two separate ways of answering the same question; or depen-
dent, meaning the subject technique cannot be used without prior
use of another one. In one way or another, each of the tech-
niques discussed holds a parallel relationship with the others
presented in this digest.
The following 26 individual techniques, disaggregated into
six classes, are discussed in Chapters 2.0 through 7.0:
-------
0 Techniques assessing the temporal patterns of TSP
concentrations
0 Long-term trends
0 Seasonal and monthly patterns
0 Weekday versus weekend and daily patterns
0 Diurnal variations
0 Emission pattern/air quality relationships
0 Techniques assessing the spatial patterns of TSP
concentrations
0 Concentration versus site type
0 Pollution, gradient, and dosage rose
0 Intersite correlations
0 Upwind/downwind
0 Techniques assessing the effect of meteorological
variables
0 Regression and correlation analyses
0 Decision tree
0 Precipitation
0 Wind speed
0 Trajectory analysis
0 Emissions data analysis techniques
0 Emission inventorying
0 Microinventorying
Diffusion modeling
0 Techniques which interpret chemical, elemental, and
morphological data
0 Temporal, spatial, and meteorologically-affected
patterns
0 Enrichment factor
0 Chemical element balance
0 Interspecies correlations
0 Pattern recognition
0 Factor analysis
0 Microscopy
0 Techniques which interpret particle sizing data
0 Frequency distribution
0 Species-specific analysis
Although the techniques presented were selected following a
comprehensive overview of much of the air pollution control
literature, it should not be concluded that this digest is total-
ly inclusive. To the contrary, itTis stressed that there are
other techniques which may be available (e.g., intervention
analysis, population exposure analyses, Q-Q plots, box plots, and
graphical enhancement of scatter plots). The authors have simply
tried to identify those techniques which have been used most
commonly or which hold good potential. Similarly, the reader
-------
should not conclude that all possible interrelationships have
been pursued. The reader is encouraged to use his imagination to
create new and potentially rewarding combinations or interactions
of techniques.
While imaginative combinations are urged, the user is also
cautioned to be aware of two potential complications associated
with the use of these techniques. First, these techniques are
data manipulative and they assume that the data themselves are
valid. The user should be aware that some concern has been
expressed regarding the accuracy of the high volume sampler data
itself. Factors such as static deposition, wind directionality,
and artifact formation may tend to produce errors in measured
concentrations and may cause the analyst to make erroneous con-
clusions concerning the nature and extent of his particulate
problem. This subject is under active investigation by EPA. The
user should also investigate the data base to assess the degree
to which standard quality assurance procedures (such as those
described in EPA's Guidelines for Development of a Quality Assur-
ance Program series) and standard instrument siting procedures
1 2
are followed. ' Second, many of the techniques involve the
application of standard statistical measures such as linear
regression, correlation analyses, analysis of variance, stepwise
linear regression, multiple regression, and so on. "The user
should be alert to the fact that misuse of these techniques can
lead to erroneous interpretation of available data. For example,
one may find (as in some of the examples provided in the follow-
ing chapters) that a data set has a correlation coefficient (r)'
of +0.8. To understand the statistical significance of this
correlation, however.- it is equally important to know the number
of samples upon which that correlation is based. As Figure 1-1
indicates, a sample size of 5 indicates a true r of -0.20 to
+0.96 at the 95 percent confidence level, whereas a sample size
of 50 indicates a true r of +0.70 to +0.87. The user is urged to
refer to standard statistical texts prior to applying these
techniques.3'4'5'6'7
-------
-i.o
-10 -O.8 -0* -O.4 -0.8 0 0.2 04 04 0.8 U>
SAMPLE CORRELATION COEFFICIENT, r
Use explained in Reference 4. The numbers on the curves
indicate sample size for the case of a two-variable linear
regression.
Figure 1-1. Ninety-five percent confidence belts for
correlation coefficient.
Source: Reference 4
Reprinted by permission of the publisher.
-------
2.0 ANALYZING TEMPORAL PATTERNS OF SUSPENDED PARTICULATES
Analyzing temporal patterns of TSP concentrations is a
simple and moderately effective means of initiating a comprehen-
sive study of the causes, sources, and severity of a suspended
particulate air pollution problem. This chapter discusses five
techniques which have been commonly used to assess those trends
and patterns:
0 Long-term trends
0 Seasonal and monthly patterns
0 Weekday/weekend and daily patterns
0 Diurnal patterns
0 Emission pattern/air quality relationships
The analyst should exercise some caution when attempting to
apply these techniques. He should, for example, ensure that the
data are being compared for identical locations. Likewise he
should ensure that equivalent measurement procedures, sampler
configurations, and quality assurance checks have been employed.
Once he has determined that a trend analysis is indeed
desirable, the analyst should consider which statistical method
he will want to use. Some 'of the most commonly used methods,
their nature, uses, and limitations, are summarized in Table 2-1.
2.1 LONG-TERM TRENDS
2.1.1 Description of Technique
This technique is retrospective in nature in that it de-
scribes the historical record of air quality measurements obtain-
ed from single sites or a regional area. It is best used to
describe past data, but trend lines can be used (with some cau-
tion) to forecast future concentrations.
6
-------
Table 2-1. STATISTICAL METHODS USEFUL IN ANALYZING TEMPORAL PATTERNS
Method
Description
Uses
Limitations
Graphical
Moving average
Whittaker-
Henderson
formula
Daniel's test
for trends
using Spearman
rank correlation
Plot of TSP concen-
trations measured
over a specified
time period
Visual interpreta-
tion of data
May produce a plot
difficult to inter-
pret
Obtained by replac- Aids in determining
ing each value in more objective trend
a series of line by providing a
equally spaced data smoother plot of the
by the mean of itself original data
and a standard number
of values directly
preceding it
Mathematical equa-
tion which yields
a smoothed curve
of TSP concentration
vs time
Comparison of a
test statistic coef-
ficient with normally
distributed table
values for signifi-
cance; the data are
ranked in ascending
order when obtaining
the test statistic
Useful in determin-
ing an objective
trend curve for any
time period
Useful in classify-
ing a pattern as
upward/downward and
indicates consis-
tency of pattern
by the level of
significance
Generally requires a
computer
Not very powerful when
using a small sample
size (<8 observations)
-------
Table 2-1 (continued). STATISTICAL METHODS USEFUL IN ANALYZING TEMPORAL PATTERNS
Method
Description
Uses
Limitations
Parametric
correlation
Regression
analysis
00
Chi-square
analysis
Comparison of
means using F-
distribution
Comparison of a test
statistic with per-
centile of Students'
t statistics for
significance
Estimation of a
linear or expo-
nential model des-
cribing a set of
data and an esti-
mate of the level
of significance
Comparison of the
percent of observa-
tion above a given
level between two
time periods; com-
parison is between
the test statistic
and the quantiles of
a chi-square random
distribution
Testing of the hy-
pothesis that there
is no difference
between observed
means
Useful in classify-
ing a trend as
upward, downward, or
no change
To produce estimates
of the constant rate
of absolute or per-
cent change over
time
Useful test for a
change in extreme
values or short-term
statistics
Useful for distin-
guishing an actual
difference between
means and attach-
ing a level of sig-
nificance to results
Must assume data are
of a normal or log-
normal distribution
Can produce a standard
error of hypothesis
testing
Mininum of five obser-
vations for each time
period above and below
the comparison level;
observations must be
independent
Can produce a standard
error of hypothesis
testing
Source: References 3, 8, 9
-------
Examination of long-term trends is most easily accomplished .
by using simple graphical methods. In doing so, a variety of
averaging times may be used. Usually, plotting quarterly or
annual means over time will be sufficient to depict the basic
temporal pattern. Other parameters sometimes plotted include
second high 24-h concentrations or the number of violations of
the annual standard. A smoother plot of the basic data can be
obtained by using some of the techniques listed in Table 2-1.
The trend lines in Figure 2-1 were, for example, generated by the;
o
application of the Whittaker-Henderson formula.
900
ISO
i 100
.£
1
I
•
a
T
Urban
T
T
I
I
I
I
I
Nofiurban
v\
i
37
S» S* «0
61 42 M
Year
Figure 2-1. Long-term trends at composite sites.
Source: Reference 8
Reprinted by permission of the publisher.
Statistical techniques, such as those listed in Table 2-1,
are desirable when a trend is not clearly distinguishable from a
graphical plot of the data. They may be necessary to sort out
the real change in air quality from random variability produced
by such factors as wind speed or precipitation. Although sta-
tistical techniques are objective in the sense that they are
reproducible, they are subject to the errors of hypothesis test-
ing discussed in standard statistics texts.
It must be emphasized that an adequate data base must be
available in order to perform any statistically valid trend
analysis. For a three year running average, an absolute minimum
-------
of five years' data is required, although a longer period of data
acquisition (such as seven years) may be preferable.
2.1.2 Applicability of Technique
Long-term trend analysis characterizes the aerosol in terms
of its variability over time. It can also permit the analyst to
generate tentative propositions concerning the causes of the
observed trends. A slow steady decrease may, for example, reflect
the conversion of homes and commercial establishments from the
Use of coal to the use of natural gas as the primary heating
fuel. A sharp increase or decrease in concentrations can be
related to the start-up or shutdown of a source within the sampl-
;ing area, if it coincides in time with such an event.
Care must be taken when proposing causes of long-term
jtrends. Any proposition may be erroneous due to other random
factors, such as meteorological interference, which can also
affect the TSP concentration. This suggests that additional
investigation of probable causes of trends should be performed.
Used by itself, long-term trend analysis techniques yield little
ifisight into the causes of a trend or the determination of the
impact of minor changes in the amount or location of emissions.
2.1.3 Relationship to Other Techniques
Long-term trend analysis is a primary tool in the process of
generally defining a particulate problem. inferences as to the
causes, effects, and scope of the problem can be made when spa-
tial patterns and meteorological variations are also considered.
Long-term trends by site type (e.g., urban versus rural) can
assess the role of fugitive dust as the major contributing source
category or can suggest the existence of other dominating source
categories. Trends in intersite correlations can also be used to
generate propositions concerning the nature of the dominant
impacting sources.
A deviation from the historical trend can often be related
to the meteorology of the area. Changes in precipitation, wind
10
-------
speed, and/or wind direction from the historical average can
significantly affect measured TSP concentrations for the period
of the deviation.
Interpretation of physical, chemical, and elemental data can
help identify major source categories as well as determine the
impact of specific sources on the past air quality. Such data
are rarely contained in the historical record, however, and
cannot normally be included in the initial analysis of a particu-
late problem.
2.1.4 Resource Requirements
Resource requirements are summarized below:
Manpower Low
Skill Low
Computer Optional
Data Minimum of five years
historical TSP data
2.1.5 Example Application
An example can be drawn from a data summary generated for
Seattle, Washington. Data had been acquired continuously at
the Public Safety Building in Seattle since February 1965. The
12-month moving geometric mean plot in Figure 2-2 shows short-
term fluctuation, but it also depicts a long-term downward trend
which appears to level off just below the value of the annual
standard. This long-term trend is even more evident in the 24-
and 36-month moving geometric mean graphs. Assessment of a trend
based on isolated 12-month segments of the 12-month moving geo-
metric mean trace could easily be erroneous: the period from
July to November 1974 indicates a sharp upward trend and just the
opposite during the same period in 1975. From this analysis, it
was concluded that ^ambient levels of suspended particles were
decreasing in this major urban area.
11
-------
120
100
J 40
I »•*
u
1-yr moving average
*7 •• 0* 70 71 72 79 74 75
Ending Month
ISO
100
2-yr moving average
i i i
07 M 0* TO 71 72 JO 74 75
Ending Month
120
00
00
I 40
-
3-yr moving average
M O7
0*707l727*7»n
Ending Month
Figure 2-2. Long-term trends, Seattle, Washington.
Source: Reference 10
-------
2.1.6 References
3, 8, 9, 10, 11, 12, 13, 14, 15
2.2 SEASONAL AND MONTHLY PATTERNS
2.2.1 Description of Technique
This technique is very similar to the long-term trend
analysis described in Section 2.1.1. Usually, seasonal and
monthly averages are plotted versus time over a period of several
years. A quarterly or monthly moving average can also be used
when curve smoothing of the raw data is desired. A smoother plot
can also be obtained using quarterly or monthly moving averages
or other techniques such as the Whittaker-Henderson formula.
Figure 2-3, for example, is the same plot as Figure 2-1 with a
different value for the constant "a" in the Whittaker-Henderson
formula to approximate the seasonal trend curve.
200
I
1
»
I
Urfaon
I
I
I A
I I
I I
*a
M M
Figure 2-3. Seasonal patterns at composite sites.
Source: Reference 8
Reprinted by permission of the publisher.
Quarterly and monthly averages can also be computed for a
number of years in combination. . A minimum of three years' data
is recommended if the data are to be used in this fashion. This
will tighten the confidence limits around the computed means.
13
-------
tend to factor out the effect of large-scale variations in meteo-
rological conditions, and minimize the variations in industrial,
fuel combustion, or other man-related emissions that have oc-
curred.
Statistical techniques presented in Table 2-1 can also aid
in the analysis.
2.2.2 Applicability of Technique
As in the case of long-term trends, this technique char-
acterizes the aerosol in terms of its temporal fluctuation.
Indirect identification of the major sources or source categories
contributing to the measured TSP levels is also possible. As an
example, consider the seasonal patterns depicted in Figure 2-3.
The urban sites' tendency to produce peak values in the winter
season may be largely attributable to the use of large amounts of
fuel (other than gas) for space heating. The nonurban sites'
tendency toward higher values in the summer may be closely related
to fugitive and windblown dust from cyclical activities in rural
areas.
Care must be taken when proposing causes of seasonal or
monthly trends, however, since many other random factors can also
affect the TSP concentrations. Precipitation, wind speeds, and
wind direction can all play an important role in causing TSP
concentrations. As with long-term trends, additional investiga-
tion of probable causes is often warranted.
2.2.3 Relationship to Other Techniques
Relatively little information is obtainable from a seasonal
or monthly trend analysis when used alone. But this technique
can suggest areas where further analyses are warranted. When
used in conjunction with other techniques, more detailed analyses
can be performed, such as identifying the location of the partic-
ulate problem, factors affecting TSP concentrations, and char-
acteristic source categories contributing to those concentra-
tions. Through the use of spatial pattern techniques (such as
14
-------
comparing urban and nonurban sites), certain inferences can be
made as to the role of fugitive dust sources. Seasonal pollution
roses can be used to indicate possible contributing sources or to
determine if the problem is due to more than one source. The
scope of the problem can be refined through the use of intersite
correlations over time. Given seasonal or monthly trends from
several sites, intersite correlations can be performed to deter-
mine if the problem is regional or site specific for all seasons
of the year.
The effects of meteorological conditions are easily assessed
when used in conjunction with seasonal and monthly trends.
Often, high correlation coefficients can be obtained when com-
paring wind speed, wind direction, or precipitation data to the
seasonal trend. This comparison can suggest whether fugitive
dust, industrial facilities, or other identifiable sources cate-
gories are culpable.
Interpretation of physical, chemical, and elemental data can
aid in identifying contributing sources if the temporal and
spatial patterns of these data coincide with the seasonal or
monthly TSP patterns. An example of this is agricultural activ-
ity. If a tracer element related to agricultural operations is
determined to have the same seasonal pattern as the TSP concen-
tration, then the dominant role of that source category can be
inferred.
2.2.4 Resource Requirements
Resource requirements are summarized below:
Manpower Low
Skill Low
Computer Optional
Data Minimum of three years
historical TSP data
15
-------
2.2.5 Example Application
An example of seasonal variations and their interrelation-
ship with other techniques is drawn from Phoenix, Arizona.
This example consists of an evaluation of seasonal patterns of
particulate levels and the associated seasonal meteorology af-
fecting the levels.
Figure 2-4 presents the quarterly averages of concentrations
of TSP at various high volume monitoring sites throughout the
Phoenix area. It is suggested that the seasonal patterns and
magnitudes fluctuate considerably from year to year due to varia-
tions in seasonal meteorology and statistical limitations stem-
ming from the small number of samples per quarter. The authors
concluded that:
0 Concentrations of particulates did not appear to
exhibit a consistent seasonal pattern at any of the
stations
0 For any given year, a similar seasonal pattern was
observed at the different stations
The authors then investigated the relationship between these
patterns and various meteorological parameters and found that:
0 Measured levels of TSP were inversely proportional to
rainfall frequency for any given season of the year
0 Levels of TSP appeared to be less sensitive to rainfall
frequency in the winter months than in the warmer
summer months
0 Levels of TSP were generally highest in the fourth and
first quarters for a given number of rainfall
days
0 TSP concentrations were relatively the same in all but
the second quarter when rainfall frequency increased
0 TSP concentrations were inversely related to wind speed
0 The most severe seasonal TSP concentrations were
apparently caused by fugitive dust arising from human
activities during periods when wind speeds were minimal
16
-------
M
O
-H
4J
a
43
0)
u
o
u
200
Central
Phoenix 3
TSP, ug/m
200-
South
Phoenix
TSP, ug/mj
'R
'B
1
19
1
i
73
!
1
I
1
I
1
1
Stas
v\\\K\l
1974
1
1
\
^
5;
19
75
1 1
i
I
97:
1
1
1-7
1
1
5
I
1973
1974
200'
West
Phoenix . 100
TSP, ug/m
\X^
1973
^
^
1974
North
Phoenix
TSP, ug/m
200
3 100-•
1973
1974
(0
o
-H
•P
•H
T)
O
U
n>
u
•H
o>
O
iH
O
M
O
10
4J
Quarterly
Rainfall,
Inches
2 •
V •
S&2
1973
10
Average
Wind Speed, 5
mph
197;
1974
1975
ft
1975
1975
1973-75
1973-71
Historical Avg.
Figure 2-4. TSP quarterly averages, quarterly meteorology
for the Phoenix area.
Source: Reference 16
17
-------
2.2.6 References
8, 9, 10, 11, 12, 14, 15, 16, 17, 18
2.3 WEEKDAY/WEEKEND AND DAILY PATTERNS
2.3.1 Description of Technique
Historical data generally contain an equal number of samples
collected for each day of the week. Averaging these data for
each day or for weekdays/weekends reveals the average daily or
weekday/weekend variations for a particular site or area. Since
one year of historical data contains only a small number of
samples for each day (typically nine per site), several years'
data are normally necessary to derive any statistically valid
inferences concerning sources or locations of emissions. The
standard statistical techniques presented in Table 2-1 can be
used to indicate significant differences and/or causes of the
variations.
2.3.2 Applicability of Technique
Again, this technique characterizes the aerosol in terms of
its temporal fluctuation. However, given an adequate number of
data points, hypotheses indicating the major contributing sources
or source categories can be proposed. High correlation coeffi-
cients between daily TSP levels and traffic patterns can often be
obtained, thereby suggesting traffic as the major source. With
certain assumptions pertaining to background levels and relative
contributions from different sources, the qualitative impact of a
specific source or source category can be estimated. Variations
in emission rates of contributing sources can also be suggested.
2.3.3 Relationship to Other Techniques
Several other techniques can be used with the daily average
variation analysis to enhance the investigator's understanding of
the nature of the measured particulate levels. The spatial
pattern analyses can yield additional information useful in
18
-------
determining the locations of sources. Concentration versus site
type for each day of the week and intersite correlations can
indicate whether the levels are region wide or site specific.
They can also suggest whether the TSP levels are due to fugitive
or windblown dust or due to man's activities (e.g., industrial
processes), if emissions from such sources are shown to vary over
time in a pattern similar to that of TSP levels. By using an
accelerated sampling frequency instead of the standard one out of
six days, pollution roses for each day of the week can be used to
indicate the direction of the particulate emissions and possibly
to aid in locating a source. Care should be taken to account for
statistical limitations imposed by the number of data points
used, however.
It is difficult to correlate average daily TSP levels with
meteorological conditions, although this analysis may yield some
useful information. An average daily pattern of rainfall, wind
speed, or wind direction will most likely not be found.
Physical, chemical, and elemental data analysis can offer
additional insight as to the sources of particulate matter. With
such data, TSP levels can be attributed to traffic, fugitive
dust, or industrial processes emitting unique tracer elements.
The pattern recognition technique can be structured so as to use
average daily values as a basic input parameter.
2.3.4 Resource Requirements
Resource requirements are summarized below:
Manpower Low
Skill Low
Computer Optional
Data Data base adequate to attach
statistical significance
to results (minimum
of three years' TSP data)
19
-------
2.3.5 Example Application
An example of a daily variation analysis is taken from
12
Philadelphia, Pennsylvania. Daily TSP levels had been moni-
tored at several Philadelphia sites for a number of years, there-
by allowing for a statistically significant analysis of the
variations during the weekly cycle of human activity. Normalized
TSP values were plotted by day of week (see Figure 2-5), taking
background concentrations into account. These normalized values
were then compared with city-wide traffic volume by day of the
week.
Figure 2-5 implies that fluctuations in daily TSP were
closely related to daily changes in city-wide traffic volumes.
This analysis tended to indicate a close correlation between TSP
levels and traffic volumes. It was observed, though, that daily
fluctuations of other emission sources could also be causing much
of the noted correlation.
2.3.6 References
9, 11, 12, 14, 15, 16, 17
2.4 DIURNAL VARIATION
2.4.1 Description of Technique
Variations in TSP levels during the day in any given area
can be determined using this analysis technique. The data re-
quired for such an analysis are not generally contained in the
historical record and must therefore be generated as part of a
special study. The data are usually generated in terms of one-,
two-, or four-h high volume samples collected over several days.
TSP concentrations versus time of day can then be plotted, and
the diurnal pattern determined. Sampling can be done over any
number of days, weeks, or months. For longer sampling durations,
less statistical weight is attached to each individual data point
and missing data do not greatly affect the significance of the
analysis.
20
-------
tsJ
1
*
W
W
I
i
8
3
H
*
fe
AMS Lab
(Avg. 1972-74 TSP)
City-Wide Traffic
(1973)
500 S. Broad
(1972 TSP)
Source;
DAY OF THE WEEK
Figure 2-5. Daily variations in traffic and TSP levels.
Reference 12
-------
2.4.2 Applicability of Technique
By defining the typical variation in TSP concentrations
during the day, this technique partially characterizes the
aerosol. In addition, it is feasible to identify major source
categories contributing to the observed concentrations if infor-
mation is available on the emission patterns in the area. For
example, reference 19 presents data relating to diurnal varia-
tions in TSP levels and compares these data to the diurnal traf-
fic and power demand patterns. Through this comparison, it was
suggested that the TSP levels closely followed the traffic and
power demand patterns and were most likely affected by them. If
there are no specific sources or source categories which display
an unusual diurnal emission pattern, it is unlikely that this
technique could be used to identify dominant sources.
2.4.3 Relationship to Other Techniques
The short time period under investigation in a diurnal
variation analysis does not lend itself to use with many other
techniques. General information is obtainable when this analysis
is performed at several sites. Source categories contributing to
the TSP levels can be suggested by locating samplers at different
site types. Intersite correlations can be used to determine if
there are nearby contributing sources and how dominating sources
vary throughout the day. Wind speed, wind direction, and mixing
height data can be used to assess their effects upon TSP levels i
throughout the sampling period.
By interpreting chemical, physical, and elemental data in
the diurnal cycle, more specific information can be obtained,
such as an identification of major contributing sources or source
categories. Filter analysis techniques, such as microscopy and
X-ray diffraction, can also be used to identify variations in
impacting sources during each sampling period.
22
-------
2.4.4 Resource Requirements
Resource requirements are summarized below:
Manpower Moderate
Skill Low
Computer Not required
Data One- to 4-h TSP averages
of sufficient quantity
to attach statistical
significance to results
a-Manpower requirements are such due to special data and analysis
requirements.
2.4.5 Example Application
An excellent example can be drawn from Philadelphia, Penn-
sylvania, where diurnal TSP levels were compared to diurnal
traffic patterns. It was hypothesized that the TSP levels
measured at a particular intersection were traffic related and
possibly were due to reentrainment of road dust. Four sampling
periods were specified, and 13 periods were sampled, starting on
a Friday night and continuing until the following Monday morning.
The hypothesis of reentrainment of the redistributed dried
particles by vehicular traffic was supported by Figure 2-6,
which presents the measured TSP concentrations and traffic vol-
umes for each sampling period. Two samples were affected by
sandblasting operations (displayed on the plot as the last two
data points). Calculation of the linear regression correlation
coefficients for the remaining 11 pairs yielded a value of +0.79,
which is significant at the one percent level. Additional data
revealed that a significant difference in TSP levels prior to,
during, and after street flushing also existed, thus statisti-
cally supporting the hypothesis.
2.4.6 References
9, 11, 17, 19
23
-------
833
10
(A
o
z
<
VI
3
O
Z
V)
•c
o
6
(A
*
W
i 4
o
z
ui
3
O
6/17 6/18
(F) (Sq)
I I
6/19
(Su)
I
20- OO- 04- 08- 12- 16- 20- 00- 04- 08- 12-
24 04 08 12 16 20 24 04 08 12 20
SAMPLING PERIOD
O -1
4-HOUR
TRAFFIC
VOLUME
6/20
(M)
300
250
i
>»
9
200
w
u
§
u
ISO
- 20-
0830
100
Figure 2-6. Comparison of TSP concentration and traffic
volume at Broad and Spruce Streets.
Source: Reference 11
24
-------
2.5 EMISSION PATTERN/AIR QUALITY RELATIONSHIPS
2.5.1 Description of Technique
In concept/ this is a very simple technique. Emission
patterns or activity patterns related to emissions from a spe-
cific source or source category are compared by standard statis-
tical techniques to TSP data patterns covering comparable time
periods. In practice/ the technique is complicated by the com-
monly observed absence of detailed emission/activity data and the
limitations imposed by the standard TSP sampling schedule of
every sixth day- Emission/activity patterns investigated by this
technique have included the following:•
0 Traffic counts for a street immediately adjacent to a
sampling site
0 Heating degree-days (related to space heating emis-
sions)
0 Product output rates from specific industrial processes
0 Open field burning
2.5.2 Applicability of Technique
This technique does not help to characterize the aerosol,
but it can lead to an identification of source categories impact-
ing upon the sampler. The primary utility of the technique is
that it helps to define qualitatively the TSP impact of a spe-
cific source or source category.
2.5.3 Relationship to Other Techniques
Generally, there is a direct relationship between this
technique and those which analyze the temporal variations in air
quality. One determines the existence of a temporal pattern for
one parameter (TSP or emission data), then looks for confirmation
of a meaningful relationship by analyzing the temporal variation
in the other.
25
-------
Similarly, propositions derived from the analysis of TSP
spatial patterns or the effect of meteorological variables can be
further investigated by analyzing emission patterns. Thus, if
the shape of a pollution rose is inclined toward a nearby roadway
and TSP values at that site are inversely related to rainfall,
then the potential significance of street dust can be substan-
tiated by a correlation analysis between TSP values and traffic
volumes on the nearby street.
Particle size, element, chemical, and morphological data can
be used to investigate further the suspicions aroused by the
application of this technique. Thus, variations in Pb and Br
concentrations could be correlated with traffic volume to extend
the example discussed above.
2.5.4 Resource Requirements
Resource requirements vary with the specific situation to
which the technique is applied. The estimates given below are
based upon the application of the traffic volume/ TSP correlation
method:
Manpower Low
Skill Low
Computer Not required
Data TSP measurements and
traffic
2.5.5 Example Application
This example is drawn from a study of the impact of reen-
trained dust upon TSP concentrations in Kansas City, Missouri.
Two sites were located on either side of a relatively well-
traveled street. High volume samplers were placed at two dif-
ferent heights at each location. Data were collected over a
three month period. A linear regression analysis between TSP
values at one of the samplers and traffic counts on the street
was performed. As Figure 2-7 indicates, a correlation of approx-
imately +0.8'was found. The effect of each additional vehicle
26
-------
300
e
c
0
•H
4J
id
M
0
c
0
u
0)
tp
(0
M
(U
>
(0
,5
I
^r
(M
M
0
(N
100
• tit* on «ail tide of »fr««»
X til* on w«il tide of street
10OU 20UU
bUUU
7000 8000
10000
Traffic volume, 12- or 24-h
Figure 2-7. Traffic volume versus concentration for McGee
Street sites, Kansas City.
Source: Reference 20
27
-------
upon TSP concentrations was also investigated by comparing con-
centrations and traffic volumes for different time periods at two
different sites. As Table 2-2 suggests, the effect on the sites
was not consistent. Qualifications attached to the results of
this study were that emissions from other sources may have inter-
fered with the results and that the emission rates per vehicle
may actually have varied considerably from site to site, de-
pending upon roadway conditions and other factors.
2.5.6 References
11, 12, 20, 21, 22, 23, 24
28
-------
Table 2-2. EFFECT OF TRAFFIC VOLUME ON CONCENTRATIONS
Location/
period
McGee Street,
Kansas City
Weekday
Weekend
8 a.m. to 8 p.m.
(day)
8 p.m. to 8 a.m.
(night)
Hamilton Ave.,
Cincinnati
Weekday
Weekend
8 a.m. to 8 p.m.
(day)
8 p.m. to 8 a.m.
(night)
Average
traffic
volume for
period
5360
1560
3190
863
17792
17325
12682
5110
Arithmel
average partj
concentration ,
Ground level
(5ft) sites
140.4
92.2
196.4
100.9
70.6
66.6
72.2
59.8
:ic
Lculate.,
ug/m
Elevated
(20ft) sites
129.6
81.4
177.3
80.6
63.7
60.8
63.7
56.7
All samplers are 35 ft from edge of street.
Source: Reference 20
29
-------
3.0 ANALYZING SPATIAL PATTERNS OF SUSPENDED PARTICULATES
In this chapter, four techniques for analyzing the spatial
pattern of suspended particulates are discussed:
0 Stratifying concentrations by site environment
0 Intersite TSP correlations
0 Pollution, dosage, and gradient roses
0 Upwind/downwind relationships
3.1 STRATIFYING CONCENTRATIONS BY SITE ENVIRONMENT
3.1.1 Description of Technique
The environments within which monitoring sites are located
can be divided into a variety of classes and can be used to
account for the spatial variation in air quality. Examples of
classification schemes that have been reported in the literature
n , .. _ ,.. . 13,16,21,25,26
include the following: ' ' ' '
Urban Central city residential/ River valley
Rural commercial (fugitive dust High density
sources) Town
Central city residential Suburban
(no fugitive dust) Background
Traffic Rural/residential (fugi-
Non-traffic tive dust sources)
Suburban residential Non-urban
(fugitive dust sources) Residential
Rural residential (no Commercial
fugitive dust) Industrial
Remote
The technique itself is rather simple. Once the sites are
classified by environment, TSP data associated with those sites*
are aggregated and summarized statistically in terms of their
30
-------
average geometric mean. These data can then be analyzed for
temporal variations/ or for the effect of meteorological vari-
ables.
3.1.2 Applicability of Technique
This technique can be useful in generating a more accurate
understanding of the regional distribution of suspended particu-
late air quality whenever topographical conditions negate the
level terrain assumptions utilized in most diffusion models or
whenever the major impacting emission sources are fugitive dust
related. However, the technique cannot be used to characterize
the aerosol in any other way-
Major impacting source categories can be qualitatively
suggested by this technique insofar as the site classification
scheme is developed objectively and not in a manner calculated to
prove the culpability of a given source. Such source categories
can be implicated by examining each high TSP site type for common
attributes. No quantitative assessment of source impact is
possible with this technique, however.
No determination of the quantitative impact of a single
source has been demonstrated with this technique. No deter-
mination of the impact of changes in emissions has been demon-
strated.
3.1.3 Relationships to Other Techniques
The temporal variation in suspended particulate concentra-
tions and the effect of meteorological conditions upon those
concentrations can be assessed for the various site classes. In
combination, these techniques can assess the role of fugitive
dust as the dominating source category or can suggest the exis-
tence of a dominating group of sources. Likewise, when used in
conjunction with other spatial pattern techniques, site classi-
fication can suggest the location of dominant sources. When
applied to chemical/elemental data, this technique can be used to
identify key source categories even more precisely (see Section
6.1) .
31
-------
3.1.4 Resource Requirements
Resource requirements are summarized below:
Manpower Low
Skill Low
Computer Optional
Data Site descriptions, TSP
measurements
3.1.5 Example Application
In a study recently prepared for the Kansas City region, 31
monitoring sites were divided into five different site exposures
13
as shown in Table 3-1 below:
Table 3-1. EXAMPLE SITE CLASSIFICATION
Class Description
Background Sites outside the central city
and remote from obvious local
man-made sources of particulate
matter
Town A town outside the central city
Suburban Areas principally residential
in character outside commercial-
industrial areas
High density Locations outside river valleys
but having a high level of
commercial or industrial
development
River valley Low elevation sites in river
valleys with surrounding higher
elevation
Stratified by their relationships to the primary annual
NAAQS, 12 of these sites were found to exceed that standard (see
Table 3-2 below).
32
-------
Table 3-2.
EXAMPLE STRATIFICATION OF TSP CONCENTRATIONS
BY SITE TYPE
Class
Background
Town
Suburban
High density
River valley
<75 ug/m
2
5
8
1
3
>75 ug/m
0
3
1
3
5
Quarterly geometric means were then computed for each of
these site classes, yielding a graphical plot showing distinct
seasonal changes in TSP concentrations. According to the au-
thors, high concentrations measured during the winter tended to
refute the hypothesis that fugitive dust sources (excluding
reentrained dust) had a major impact, since fugitive dust impact
should be highest in the dry summer and fall. The prevalence of
high wind speeds and the occurrence of street sanding during the
winter tended to support a reentrained dust impact hypothesis.
Correlation coefficients were calculated for particulate
concentrations versus wind speed, mixing height, days since
precipitation, and barometric pressure for each of the five site
classes. The correlation coefficients shown in Table 3-3 were
noted; however the confidence limits around those values were not
reported.
Table 3-3.
EXAMPLE CORRELATIONS OF METEOROLOGICAL DATA
FOR SITE TYPES
Correlation coefficient
Class
Background
Town
Suburban
High density
River valley
Wind
speed
-.033
-.024
-.264
-.169
-.157
Mixing
height
.221
.315
.267
-.050
.085
Days since
precipitation
.555
.290
.286,
.237
.426
Barometric
pressure
.051
-.160
.040
.121
-.031
33
-------
The authors concluded that the four meteorological parameters
which were investigated failed to account adequately for the
observed daily variations in TSP concentrations. The differences
in correlations from site type to site type were considerable,
however, and suggested that the site classification scheme was a
meaningful conceptualizing device.
It should be noted that the authors should have assessed
their results in terms of probability or confidence limits.
Failure to do so may have resulted in a misinterpretation of the
results.
3.1.6 References
13, 16, 21, 25, 26
3.2 INTERSITE TSP CORRELATIONS
3.2.1 Description of Technique
Correlation coefficients can be computed between daily TSP
concentrations measured at one site and those measured on the
same days at each other site in the region. Two sites which are
relatively close together (i.e., less than two miles apart) can
be expected to correlate highly if they are being impacted by the
same dominant source or source category and are in a location
with unobstructed wind flow. The correlation analysis can be
performed for a data set of one year or more and improves in
accuracy as the number of measurements in the data set increases.
Procedures for computing correlation coefficients are amply
described in standard statistical texts and need not be described
in detail here. In general, the correlation coefficient is
defined as:
34
-------
r =
" - - 2 - 2 1/2
[Z(x±-x) * Z(y±-y) Z] 17^
where x . = the ith value of the variable x
E
_ x-
and x = —^- = mean of all x values
N
In this case, x is the TSP value at one site and y is the TSP
value on the same day at another nearby site. The best possible
correlation is given by r = ±1. As r approaches zero, the vari-
ables become more randomly related. If r = +1, x and y are
directly proportional; if r = -1, they are inversely proportion-
al. Correlation coefficients indicate how well two variables
vary together, not necessarily that there is or is not a causal
relationship between the two variables.
3.2.2 Applicability of Technique
This technique can characterize the aerosol in terms of its
spatial distribution. This is done primarily in terms of assess-
ing the uniformity of fluctuation in TSP concentrations from site
to site .
Where nearby sites intercorrelate highly, the presence of a
common dominating set of impacting source categories can be
inferred. However, it is not possible to identify what those
source categories are or what their respective contributions are.
The existence of a unique source or source category can be in-
ferred when one site correlates poorly with a set of sites that
otherwise intercorrelate highly.
A large negative correlation implies that one site has large
values when the other has small values. This could occur when a
large point source is located between the sites.
No determination of the impact of changes in emissions has
been demonstrated with this technique.
35
-------
3.2.3 Relationship to Other Techniques
Given a sufficient number of samples, temporal patterns in
intersite correlations can be investigated. If the correlation
coefficient changes dramatically, it can be inferred that domi-
nating sources vary over time.
Again assuming a sufficient number of samples are available,
measured concentrations can be stratified by meteorological
conditions and intersite correlations can be computed therefor.
The nature of dominant sources can be inferred from changes in
the resulting correlations.
Intersite correlation data can also be computed for differ-
ent species (e.g., Pb, SO,, NO_) and compared to TSP intersite
correlations. Where TSP intersite correlations are high and
species correlations are high, it can be inferred that the two
sites are measuring essentially the same air mass.
3.2.4 Resource Requirements
Resource requirements are summarized below:
Manpower Low
Skill Moderate
Computer Preferable, depends
upon number of sites
Data TSP measurements at
more than one site
(adequate number of
measurements to assure
statistical significance)
3.2.5 Example Application
An example can be drawn from a recent study of the TSP
22
attainment status in Houston, Texas. Two sites are located on
the north side of the central downtown area. Less than a mile
apart, both of these sites have severe wind flow obstructions to
the east but they are somewhat dissimilar in their immediate
environments (see Figure 3-1). The daily TSP concentrations
correlate well (r = 0.86). Pollution roses for each site have
36
-------
Figure 3-1. Locations of sites to be intercorrelated,
Source: Reference 22
37
-------
similar shapes. Correlations of TSP and Pb values by wind direc-
tion indicate strong relationships when persistent winds come
from the north and northwest. Correlations between the two sites
(in terms of nitrates, sulfates, ashable organics, Mn, and Cu are
L-all greater than +0.75.
' From these comparisons, the authors of the study concluded
that the two sites measured essentially the same air mass despite
the differences in their immediate environments. An analysis of
sources in the vicinity of the two sites led the authors to
conclude that traffic-related sources to the north and northwest
of the sites and industrial sources to the southwest were the
primary identifiable contributors to the polluted air mass sam-
pled at those sites.
, A matrix display of a typical set of intersite TSP correla-
tions from a recent study in Philadelphia is shown in Table 3-4.11
In this example, daily TSP concentrations at 13 sampling sites
are intercorrelated for each of three years. As that table
indicates, there was considerable variation in correlations from
site pair to site pair and from year to year. Some sites, such
as BEL and FRI correlated well from year to year, whereas others,
.such as S/E and ALL correlated well one year and poorly the-next.
;Again it should be noted that the utility of this matrix would
be enhanced, were the results to be expressed in terms of prob-
abilities or assigned confidence limits.
3.2.6 References
11, 22
3.3 POLLUTION, DOSAGE, AND GRADIENT ROSES
3.3.1 Description of Techniques
The three techniques grouped into this category u-se essen-
tially the same types of information but apply them- slightly
differently.. A pollution rose is produced by directionally
analyzing air quality data at a given sampling station. The data
38
-------
Table 3-4. LINEAR CORRELATION COEFFICIENTS BETWEEN 24-H TSP
LEVELS AT PHILADELPHIA MONITORING STATIONS IN
1974, 1975 AND 1976
Site DBF ALL INT BEL ROX N/E NBR FRI*
DBF -
ALL -
INT -
BEL -
ROX -
N/E -
NBR -
FRI -
LAB -
SBR -
S/E -
500 -
AFS -
'76
'75
•74
'76
'75
•74
•76
•75
•74
•76
'75
'74
'76
•75
'74
'76
'75
•74
'76
•75
•74
'76
•75
'74
•76
'75
'74
'76
'75
'74
'76
'75
'7.4
'76
'75
'74
'76
'75
'74
O.dl 0.69 0.70 0.49 0.76 0.65 0.
0.61 0.56 0.72 O.C4 0.65 0.66 0.
0.64 0.72 0.59 0.56 0.63 - 0.
0.57 0.66 0.45 0.60 0.66 0.
75
68
74
75
0.52 0.45 0.36 0.66 0.47 0.68
0.50 0.50 0.44 0.51 - 0.
0.37 0.50 0.70 0.55 0.
0.66 0.51 0.56 0.41 0.
0.39 0.44 0.60 - 0.
0.88 0.78 0.85 0.
0.89 0.80 0.87 0.
0.56 0.55 - 0.
0.65 0.60 0.
0.68 0.66 0.
0.68 - 0.
0.63 0.
0.43 0.
0.
0.
0.
74
62
60
56
91
85
72
67
65
60
70
7-2
54
76
56
-
LAB
0
0
0
0
0
0
0
.78
.78
.79
.70
.66
.62
.74
0.63
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.54
.78
.86
.61
.66
.78
.61
.85
.91
.68
.78
.66
-
.75
.80
.70
SBR
0.
0.
0.
0.
78
75
—
76
55
-
0.67
0.
0.
0.
0.
0.
0.
0.
44
—
81
86
—
62
64
-
71
69
-
0.89
0.
0.
0.
0.
0.
72
-
85
74
-
81
81
-
S/E
0.89
0.59
0.80
0.81
0.27
0.40
0.61
0.47
0.67
0.73
0.48
0.53
0.46
0.46
0.19
0.75
0.54
0.47
0.60
0.30
-
0.78
0.49
0.67
0.68
0.54
0.54
0.74
0.34
-
500
0.79
0.67
0.78
0.85
0.47
0.62
0.66
0.43
0.58
0.95
0.76
0.62
0.59
0.68
0.52
0.74
0.65
0.53
0.75
0.66
-
0.83
0.63
0.80
0.34
0.78
0.68
0.81
0.65
-
0.70
0.49
0.53
AFS Average
0.77
0.84
—
0.75
0.66
-
0.59
0.58
-
0.56
0.87
-
0.67
0.54
-
0.71
0.73
-
0.72
0.62
-
0.68
0.79
-
0.87 .
0.82
-
0.79
0.83
-
0.84
0.50
0.77
0.73
0.74
0.68
0.69
0.70
0.53
0.55
0.61
0.53
0.56
0.75
0.76
0.56
0.60
0.62
0.51
0.72
0.67
0.58
0.70
0.58
-
0.75
0.68
0.67
0.77
0.75
0.64
0.77
0.67
-
0.72
0.46
0.53
0.77
0.63
0.63
0.73
0.71
*CAMP station in 1974; roof-top station in 1976.
Source: Reference 11
39
-------
are segregated according to the wind direction observed while
each sample was being collected, and samples with like wind
direction are grouped together. Using degree increments of the
360
compass for direction, the result is —- subsets (where x is the
X
number of degrees in each radial subset), each containing the
data points recorded with corresponding wind directions within
that degree increment. For each subset, one then calculates
X
an appropriate measure such as geometric mean or arithmetic mean.
The resultant numbers indicate the distribution of pollutant
readings versus direction. These numbers may be conveniently
plotted and interpreted in polar form. A variation of this tech-
nique involves computing a pollution rose only for those data
that exceed some preselected 24-h concentration (e.g., 75 ug/m ).
Since the wind blows from some directions more frequently
than from others, the impact of source emissions will be weighted
not only by the magnitude of the readings associated with that
source, but also by the frequency of occurrence of those readings.
A dosage rose can take this fact into account. A dosage rose is
produced by plotting the total dosage received at the sampling
site in terms of concentration-days (ug-days/m ) rather than the
average TSP concentrations.
A gradient rose is similar to a pollution rose, except that
the difference in TSP levels between two sites is displayed as a
function of wind direction instead of the average TSP level
versus wind direction. It is produced by selecting only those
days when winds blow persistently toward one sampler from the
direction of a nearby (e.g., <8 miles distant) sampler, calcu-
lating TSP concentrations at the two sites, and recording the
difference. The resulting values are displayed on a polar plot.
A negative value is recorded for the first site if its measured
mean concentration is lower than that of the nearby site.
3.3.2 Applicability of Technique
Other than by partially describing the spatial variation in
TSP concentrations, this technique does not lead to a character-
ization of the aerosol. .-•;•
40
-------
This technique identifies major impacting source categories
only insofar as a distinct set of such source catgories can be
pinpointed by the shapes of pollution roses generated at nearby
sites.
A pollution rose can pinpoint the existence of a major
source if the shape of the rose clearly points toward an iden-
tifiable source. A semiquantitative assessment of the contri-
bution from that source can be made by comparing the average
concentrations for other directions with the one pinpointed by
the rose. The dosage rose will do the same, but it will also
take the impact of dominant wind directions into account.
The gradient rose will add to the understanding provided by
the pollutant/dosage roses. If there is a large positive gra-
dient in a particular direction/ on the average, TSP levels are
higher at the site under study than at the nearby upwind site.
This large increase indicates a localized source between the two
samplers. A small gradient indicates an area source or the
presence of many point sources which affect the two sampling
sites approximately equally. A large negative gradient indicates
that either an effective sink for airborne particulates is located
between the two sampling sites or a large source is located
upwind of the upwind sampling site.
A key potential limitation associated with this technique is
that the wind direction data upon which it relies may have been
generated from a location which is not representative of the -
location at which the TSP concentrations were measured. The best
way to avoid this problem is to generate site-specific wind data
whenever possible.
3.3.3 Relationship to Other Techniques
Given a sufficient number of samples to make a statistically
valid comparison, pollution roses can be computed for differing
time periods. The resulting roses can be related to available
knowledge concerning the types, magnitude, and seasonal varia-
tions in emissions from nearby sources. Similar computations can
41
-------
The made for different sets of meteorological conditions. The
;possible presence of dominating nontraditional sources can be
investigated through the use of this latter combination of tech-
niques .
'_ Data from the most current emission inventory can be used to
select sampling sites that can be investigated through the use of
pollution roses. Likewise, microinventories can be applied to
sites which have inexplicable pollution rose shapes. Given
i
iadequate data, particle size distributions for various chemicals
'and elements can also be stratified by wind direction (see Sec-
'tion 7.2).
3.3.4 Resource Requirements
i Resource requirements are summarized below:
Manpower Low
Skill Low
Computer Not required
Data TSP and site-representative
wind direction measure-
ments
3.3.5 Example Application
Three sites in the northeast part of the Houston, Texas
22
metropolitan area are shown in Figure 3-2. Pollution roses for
those sites, also shown in Figure 3-2, indicate that the highest
concentrations at each of these sites occur when winds blow from
an industrial area to the south and southwest. Examination of
the gradient roses (see Figure 3-3) indicates that higher concen-
trations are measured at the center site than at the other two
when winds blow through the residential area. There is a dra-
matic increase at the center site when winds are from the south-
west across both the industrial area and a four-lane highway.
The study from which this example is drawn concluded that al-
though the impact of the industrial area was evident, it was not
known whether point sources or fugitive emissions from that
industrial area were the cause of that impact.
42
-------
to
1. Gulf Chemical & Metallurgical
2. Union Carbide Corp.
3. Amoco Oil Co.
4. Marathon Oil Co.
5. Texas City Refining
6. Monsanto Co.
Figure 3-2
Source: Reference 22
Location of sites and pollution roses.
-------
1. Gulf Chemical & Metallurgical
2. Union Carbide Corp.
3. Amoco Oil Co.
4. Marathon Oil Co.
5. Texas City Refining
6. Monsanto Co.
Source: Reference 22
Figure 3-3. Gradient
roses,
-------
3.3.6 References
22, 21, 28, 29
3.4 UPWIND/DOWNWIND RELATIONSHIPS
3.4.1 Description of Technique
This technique can be applied on two different scales:
regional or local. In both cases the concept is simple: to
measure concentrations upwind and downwind, from a suspected major
source of particulate matter and to compute the difference in
concentrations for that period when winds were consistently from
the upwind direction. On a regional scale, this technique can be
used to determine the relative contributions of natural and
urban-generated aerosols. On a local scale, it can be used to
estimate the impact of a specific source.
3.4.2 Applicability of Technique
Other than revealing the spatial variation in air quality,
this technique does not shed any light on the character of the
aerosol. But, the regional scale upwind/downwind analysis can
provide some information as to the role of naturally occurring
windblown dust combined with the long-range transport of sec-
ondary particulates.
The primary utility of the upwind/downwind technique is that
it can accurately quantify the impact of identifiable sources.
This capability is especially useful in conducting special studies
of the impact of a single source. (See example below.)
Once a source's relationship to upwind/downwind measurements
has been satisfactorily quantified, then the effect of changes in_
that source's emissions can be estimated by using linear rollback
calculations.
3.4.3 Relationship to Other Techniques
For the upwind/downwind technique to be meaningful, it is
necessary to analyze only those data generated when the wind
45
-------
"persistently blew from the upwind direction. This can be accom-
plished either by sampling only during periods of such winds (as
with directionally actuated samplers) or by extracting and sta-
tistically analyzing only days of direction-relevant winds. It
*is possible, however, that unknown squrces are contributing to
measured differences in upwind and downwind concentrations.
Microinventories of the areas immediately surrounding the sam-
plers can be helpful in investigating this possibility. More
useful yet is to compare the upwind/downwind relationships re-
vealed by this TSP analysis with those revealed by the analysis
jof species concentrations at the two sites, taking into account
•the types of particulate matter emitted by the suspected source.
[Likewise, microscopic analysis of material collected on indi-
vidual upwind and downwind samples will add supporting evidence
to source origin hypotheses.
3.4.4 Resource Requirements
i Resource requirements are summarized below:
i
^ Manpower Low
/ Skill Low
Computer Not necessary
Data TSP and site-representative
wind direction measure-
ments
3.4.5 Example Application •
An example can be drawn from a recent effort to quantify the
rate of dust emissions from aggregate storage piles. The
stockpiling operations at a sand and gravel pit were selected for
testing. Prevailing winds were from the southwest and south.
The storage area shown in Figure 3-4 covers approximately 17
acres. Fifteen stockpiles were located in this area. Other
sources, such as crushing and screening, were either located
north of the stockpile area or else were inactive during the
sampling period.
46
-------
I)/® SAMPLING SITES
NO. 6 STO*
PIT
NO. S
STONE
CRUSHING AND
SCREENING PLANT
NO. 8
GRAVEL
NO. 57 GRAVEL
30'H
MAINTENANC
( SCREENING
I CRUSHING I
HI (31'H
WEIGH STATION
BY-PASS US 50-126
Figure 3-4. Aggregate storage sampling sites.
Source: Reference 15
-------
Field testing was conducted for a one month period and
consisted of eleven 24-h runs and eight 12-h runs. Conventional
high volume samplers with wind direction activators were used.
A 180 degree sector of sampling was employed, so that any wind
with a southerly component activated all of the samplers. Five
of these samplers were located north (downwind) of the storage
area, and one was located to the south (upwind). Wind speed and
direction data were also measured and recorded at the site.
As indicated in Table 3-5 below, the test results indicated
that the upwind sampler (station 1) recorded lower average con-
centrations than did the downwind sites:
Table 3-5. EXAMPLE UPWIND/DOWNWIND CONCENTRATIONS
Site Arith mean,
ug/m
1 (upwind) 73
2 124
3 76
4 108
5 86
6 107
An evaluation of the effects of four different factors on
the emission rates was also performed. Those factors were rain-
fall, wind speed, type of aggegrate material, and amount of
activity in the piles.
3.4.6 References
15, 30, 31, 32
48
-------
4.0 ASSESSING THE EFFECT OF METEOROLOGICAL VARIABLES
Many commonly measured meteorological variables may affect
TSP concentrations. This chapter discusses techniques which have
been used to determine the relationships between individual
meteorological parameters and TSP concentrations. The five
techniques discussed are:
0 Correlation and regression techniques
0 Decision tree analysis
0 Analysis of precipitation
0 Analysis of wind speed
0 Trajectory analysis
4.1 CORRELATION AND REGRESSION TECHNIQUES
4.1.1 Description of Techniques
The most widely used technique to isolate the relationship
between an individual meteorological variable and associated TSP
concentrations is simple correlation of daily measurements of the1
two variables (see Section 3.2.1 for discussion of correlation).
Several meteorological variables can be evaluated at once and the
paired correlations reported in matrix form. Meteorological
parameters commonly used as independent variables are listed in
Table 4-1. Statistical tables of significance, based on sample
size, provide comparative "r" values to determine whether the
calculated correlations are significant.
TSP measurements at several different sites in an urban area
or region can be placed in the matrix and correlated with the
meteorological variables and other TSP measurements. Indicators
of particulate air quality other than TSP, such as light scat-
tering or citizen complaints, may also be included as parameters
49
-------
Table 4-1. METEOROLOGICAL VARIABLES
Meteorological
parameters
Alternative forms
Precipitation
Wind speed
i
!
I
Wind direction
!
I
Visibility
Inversion
i
^Temperature
I
pew point temp
i
Cloud cover
-*9
Sunshine
Amount, duration (h), days since rain, log of
days since rain, 3-day accumulated, amount on
preceding day or days
24-h average, max h, min h, resultant, ratio of
resultant to average, average through mixing
layer
Frequency of each of 8 dir plus calms, total
miles of wind from each of 8 dir, most per-
sistent, resultant, variability in percent
24-h average, max h, min h
Height, base (lower height), intensity, duration,
a.m. mixing height
24-h average, max h, min h, wet bulb or dry bulb
24-h average, max h, min h
Average percent
Hours
50
-------
[in this analysis. Also, significant correlations between two
meteorological variables may be used to indicate related meteo-
rological conditions (e.g., winds from the south may correlate
Well with precipitation).
The time periods for the TSP and meteorological measurements
may be offset slightly in a second analysis and the resulting
correlations compared with those in the first analysis. Higher
correlations reveal a time lead or lag in the relationship be-
tween TSP and those meteorological variables.
Many studies have employed the correlation matrix as a
screening technique to reduce the number of variables to be
considered in a subsequent stepwise multiple regression analysis.
Available computer programs such as IBM's subroutine STPRG (Sci-
entific Subroutine Package) report the independent variables in
the order of their importance and the total variance in TSP
concentration at a site due to all the meteorological variables
that increase the multiple correlation coefficient. A stepwise
multiple linear regression program also calculates an F-value for
analysis of variance, to determine the statistical significance
of each additional variable in the multiple regression.
The higher the multiple correlation coefficient, the more
the meteorological variables explain or relate to TSP concentra-
tions. Also, the analysis can provide an equation relating TSP,
the dependent variable, to all the significant meteorological
variables. This equation has predictive capabilities.
4.1.2 Applicability of Techniques
If the spatial distribution of sources and receptors remains
fixed, then it is assumed that variations in daily TSP concentra-
tions are a function of meteorological parameters and source
strengths. Correlation and regression techniques attempt to
explain as much of the variations as possible in terms of meteo-
rological conditions. If some emission parameters are included
in the correlation matrix, this technique can determine whether
these parameters are related to TSP concentrations more or less
so than the selected meteorological variables.
51
-------
These techniques do not characterize the aerosol. A com-
parison of the most important variables at different sites can
lead to indirect, qualitative judgments as to impacting source
categories (e.g., fugitive dust sources are more affected by
precipitation than other sources). Wind direction as a meteoro-
logical variable can reveal directions associated with high and
low TSP concentrations. Higher correlations with a time lag may
indicate the type of source or relative distances of impacting
sources.
No quantitative assessment of the impact of a specific
source or of a change in emission rates can be made with these
techniques.
4.1.3 Relationship to Other Techniques
Correlation and regression analyses of meteorological data
can explain observed seasonal patterns, but not shorter-term
patterns or long-term trends.
In analyzing spatial patterns, a correlation matrix may show
natural groupings of sites which relate similarly to changes in
wind speed, precipitation, temperature, and other meteorological
parameters and therefore may be affected by the same air masses.
Also, a correlation matrix for meteorological variables may be
done simultaneously with intersite correlation analysis. Use of
wind direction as a meteorological variable produces an analysis
similar to, but potentially more comprehensive than, the pollu-
tion rose technique.
Neither emission inventory nor diffusion modeling techniques
have obvious relationships to analysis of meteorological vari-
ables. However, source emission patterns can be used in a corre-
lation matrix or in stepwise multiple regression along with air
quality and meteorological parameters if the emission patterns
can be described on a daily basis (e.g., daily traffic volumes or
days with coal firing in a dual fuel power plant).
52
-------
4.1.4 Resource Requirements
Data requirements for these techniques are quite high, with
a resulting need for extensive data handling. Resource require-
ments are as follows:
Manpower Moderate
Skill Moderate
Computer Required
Data TSP averages, equivalent
meteorological variable
measurements
4.1.5 Example Application
The example is taken from a study of the impact of field [
burning on particulate air quality in the Eugene-Springfield area
24
of Oregon. Correlation analyses were followed by stepwise
multiple regression analyses. Air quality, emissions, and mete-
orological data for the 1974, 1975, and 1976 field burning sea-
sons were used, with separate runs for.each season's data. ,
Both types of analyses included 31 variables: 8 measures ofj
air quality (3 daily hi vol concentrations, 2 daily average light
scattering measurements, visibility, smoke readings, and daily. i
complaints); 10 emission variables (number of acres of fields and
number of tons of slash burned per day by quadrant—N, E, S, and
W—plus total field burning and total slash burning) ; and 13 '•.
meteorological variables (daily average temperature, rainfall, j
logarithm of number of days since rain, relative humidity, wind i
frequencies from each of eight compass directions, and calm I
winds).
The 31 x 31 correlation matrices by year showed that the i
three hi vols correlated well (0.70-0.96) with each other for all
three seasons. TSP concentrations correlated with light scat-
tering (0.33-0.75) for all sites and correlated somewhat with
smoke observations (0.50-0.59), slash burning activity in 1976
(0.16-0.65), logarithm of days since rain (0.45-0.68), and rela-
tive humidity (-0.36 to -0.61). There was no correlation in any (
season between field burning activity and TSP concentrations.' I
*»
53
-------
The regression analyses with air quality parameters as
dependent variables can be summarized as follows: in all three
seasons, air quality variables (TSP, light scattering, visibil-
ity, and smoke observations) were most strongly related to meteo-
rological variables. Slash burning had a greater effect than
field burning. The only exception was complaints, which were
definitely affected by increased field burning. The ranked
independent variables for the regression analyses with 1974 data
are shown in Table 4-2.
4.1.6 References
13, 16, 18, 24, 29, 33, 34, 35
4.2 DECISION TREE ANALYSIS
4.2.1 Description of Technique
This technique utilizes the AID decision tree program
developed at the University of Michigan Institute for Social
Research. Starting with a large number of daily observations
of TSP (dependent variable) and several meteorological variables
(independent variables), the AID program sequentially splits the
sets of daily observations into two subgroups of an independent
variable, at each split choosing the partitipn of the data that
maximizes the difference in TSP concentrations as measured by the
residual sum of squares (RSS).* The partitioning process is
repeated on successively smaller subgroups of daily observation
sets until one of the following circumstances occurs:
n - 2
RSS = £ (A. - A) , where
j=l 3
A = TSP concentration for one day
n = number of days of data in subgroup
ARSS = RSS. .. - , - RSS.,, - RSS.,0
initial Gl G2
54
-------
Table 4-2.
STEPWISE MULTIPLE REGRESSION ANALYSIS
WITH 1974 DATA
Dependent variable
TSP Eugene Airport
TSP Eugene Commerce
TSP Springfield Library
B . Eugene
scat ^
B , . Springfield
scat
Visibility
Smoke observations
Complaints
Independent
variables
Rel humidity
Calm winds
W winds
SE winds
Log of days
NW winds
N winds
Rel humidity
W winds
Calm winds
Log of days
S fields
W slash
Rel humidity
Calm winds
W winds
Log of days
S fields
NW winds
W slash
W winds
W slash
N fields
Rel humidity
NW winds
NE winds
Calm winds
N winds
SE winds
W winds
N fields
Calm winds
NE winds
Av temperature
NW winds
Log of days
Rainfall
SE winds
NE winds
Rainfall
N fields
W winds
SE winds
Av temperature
Total slash
Av temperature
SW winds
N slash
NE winds
NE winds
N fields
S winds
Cum mult
correlation
0.484
0.576
0.634
0.660
0.678
0.692
0.703
0.606
0.695
0.752
0.768
0.775
0.784
0.541
0.660
0.714
0.735
0.751
0.768
0.783
0.383
0.476
0.5SO
0.593
0.628
0.664
0.689
0.702
0.709
0.400
0.560
0.602
0.641
0.669
0.690
0.709
0.718
0.725
0.340
0.436
0.519
0.555
0.569
0.581
0.590
0.294
0.395
0.438
0.461
0.610
0.662
0.689
Sign-
-
+
-
+ -
+
• —
+ -
-
—
+
+
+
+
-i-
-
+
+
-
+
~
T
-
-
+
+
-
+
-
-f
+
+
+
—
+
+
-
—
-
+
-
+
—
+
-
+
+
+
+
+
Source: Reference 24
55
-------
0 No subgroups can be split to achieve a ARSS above a
preset lower bound, e.g., 1 percent of original RSS of
TSP concentrations
0 Further division would produce a subgroup with less
than a preset number of days of data, e.g., 3
0 The number of terminal subgroups reaches a preset
limit, e.g., 10
The independent variables (meteorological parameters) in
this program must be represented by discrete numbers. To achieve
discreteness, the raw meteorological data are usually divided
into ranges. For example, if the variable Z represents wind
speed, the investigator may assign the following values for Z:
Z^ Wind speed, mph
1 0-3
2 3-7
3 7-15
4 15-25
5 > 25
The version of AID used in TSP analyses is simplified so that
meteorological variables must be assigned discrete numbers rep-
resenting monotonic intervals (as in the wind speed example
above). Also, the simplified program performs only monotonic
splits of the data (i.e., Gl = 1, 2, 3 and G2 = 4, 5 but not Gl =
1, 4 and G2 = 2, 3, 5).
The result of decision tree analysis is a listing of the
independent variables in order of importance according to the
percent of variance in TSP concentrations that each explains.
The analysis also identifies the ranges of meteorological con-
ditions that account for large percent variances in TSP. It has
been found that the AID program generally explains more of the
variance in TSP data than does multiple linear regression, be-
cause it does not involve restrictive assumptions such as lin-
earity or additivity.
56
-------
4.2.2 Applicability of Technique
Decision tree analysis of meteorological data has the same
applications as correlation and regression techniques. It also
has the same relationships with other techniques and similar
resource requirements.
As with correlation and regression, emission parameters can
be included along with meteorological parameters as independent
variables in the analysis. This allows determination of relative
effects of emissions and meteorology on TSP variance.
The meteorological classes used in AID provide more infor-
mation on TSP-meteorology relationships than regression coef-
ficients. Also, these meteorological classes can be used to
normalize TSP trend data for differing meteorological conditions.
4.2.3 Example Application
The example is an application of the AID program to TSP and
meteorological data at 25 hi vol sites in EPA Region VI.
for the analysis were from the years 1973-1975.
rological variables were input:
Data
Eighteen meteo-
Month of the year
Average visibility
No. observations blowing
dust
Average wind speed
Resultant wind dir
Average relative humidity
Wind variability
a.m. mixing height
a.m. average wind speed
thru mixing layer
p.m. average wind speed
thru mixing layer
Max temperature
Min temperature
Amt of precipitation
Max wind speed
Three-day accumulated
prec ipitation
No. of days since last
precipitation
No. of precipitation
observations
For each meteorological parameter, six to nine data ranges were
specified. Termination criteria for the program were the same as
discussed previously: ARSS <1 percent RSSQ, n<3, or 10 terminal
subgroups.
57
-------
An example of the program output is presented in Figure 4-1.
The percent variance explained at each of the 25 sites is shown
in Table 4-3; values range from 28 to 72 percent and average 51
percent. These are equivalent to total correlation coefficients
of 0.53 to 0.85 which average 0.71.
The variables that were found to be most important in
explaining TSP variance were month of year (appearing at 16
sites), the two mixing height variables (at least one at 20
sites), number of days since rain (14 sites), wind variability
(14 si-tes), average relative humidity (13 sites), maximum daily
temperature (13 sites), resultant wind direction (12 sites), and
three-day accumulated precipitation (12 sites).
To provide an assessment of the decision tree technique's
performance, multiple regression analyses were also run on the 25
data sets. The percents of TSP variance explained by linear and
log-linear multiple regression are presented in Table 4-3 for
purposes of comparison. The results indicate that the decision
tree technique is considerably better than multiple regression in
explaining TSP levels in terms of meteorological variables. The
performance of this technique is particularly notable because the
decision trees use only a few of the meteorological parameters
while the regression equations use all 18 parameters. One pos-
sible explanation for the increased capability is that the AID
program discerns significant nonlinear dependencies, whereas
regressions are limited to linear or log-linear relationships.
4.2.4 References
36, 37
4.3 ANALYSIS OF PRECIPITATION
4.3.1 Description of Technique
In most cases, precipitation is considered to have the
greatest impact of the meteorological variables on TSP concentra-
tions. Precipitation has three effects, all of which tend to
58
-------
FORREST CITY, AR
1 N = 166
Y = 99.1
PMMHslOOO
1000<60
/
12 N = 22
Y = 101.7
55
123.0
11 N = 5
Y = 172.0
\
60
-------
Table 4-3. PERCENT VARIANCE EXPLAINED BY AID COMPARED TO
PERCENT VARIANCE EXPLAINED BY MULTIPLE REGRESSIONS
AGAINST ALL 18 METEOROLOGICAL VARIABLES
Site*
Forrest City, AR
Helena, AR
Jonesboro, AR
Springdale, AR
Stuttgart (1), AR
Stuttgart (2), AR
Albuquerque (1), NM
Albuquerque (2), NM
Dona Ana, NM
Las Cruces, NM
Raton, NM
Bethany, OK'
Muskogee, OK
Oklahoma City, OK
Roger Mrlls County, OK
Sequoyah, OK
Tulsa (1), OK
Tulsa (2), OK
Corpus Christi (1), TX
Corpus Christi (2), TX
Corpus Christi (3), TX
Harlingen, TX
Lubbock, TX
McAllen, TX
San Bern" to, TX
Percent variance explained
AID program Regression
1 S groups 10 groups Linear Loq-linear
50%
60%
60%
63%
61%
67%
70%
73%
63%
62%
52%
51%
36%
55%
47%
69%
69%
44%
50%
67%
63%
71%
76%
47%
59%
42%
52%
51%
52%
53%
62%
62%
59%
54%
56%
44%
39%
28%
50%
45%
61%
59%
40%
43%
497,
54%
61%
72%
38%
55%
31%
31%
42%
38%
29%
49%
56%
42%
58%
33%
19%
31%
24%
37%
32%
50%
28%
24%
23%
13V-
40%
60%
66/'>
24%
20%
44%
42%
47%
43%
41%
52%
45%
31%
63%
50%
23%
41%
29%
43%
40%
54%
29%
13%
18%
27%
28'/.
57%
53%
24%
40%
Mean
Source: Reference 37
59%
51%
36%
39%
60
-------
reduce TSP levels: (1) it condenses on particles within clouds
(rainout) and washes out particles as it falls; (2) it wets
surfaces and thereby suppresses generation of fugitive dust; and
(3) it is often associated with frontal passages which bring in
relatively clean air.
A large number of analytical techniques and graphical data
presentations have been employed to quantify the effect of rain-
fall in reducing concentrations. Some of the more useful tech-
niques are described briefly below.
0 Correlation/Regression. As discussed in Section 4.1,
one or more precipitation-related variables are usu-
ally involved in correlation and regression analyses.
0 Daily TSP concentration plotted against amount of
precipitation. Time periods for the precipitation
values can be the sampling day/ the day before sam-
pling, the previous 48 hours, sampling days with no
rain on prior day, and weekday or weekend sampling days
only.
0 Comparative frequency distributions of TSP.concentra-
tions for rain/no rain data subgroups. Possible sub-
groups are no yain within 48 h, no rain within 7 days,
rain >.01 in. within 48 h, rain >.10 in. within 48 h,
and rain">.25 in. within 48 h.
0 Average TSP concentration plotted against days since
rain. If TSP data from several sites are used, they
should be normalized. This analysis provides better
results if hi vol samples are obtained daily rather
than every sixth day. Different amounts of minimum
rainfall can be specified (e.g., days since .25 in.
rainfall) to produce a series of curves. The effect of
different rainfall intensities is shown by the interval
between the curves.
0 Time plots of average TSP on the same graph as inverse
precipitation (e.g., average monthly rain/rain for
month), using monthly, quarterly, or^annual averaging
periods. To the extent that precipitation affects TSP
concentrations, the two curves should track one an-
other. A correlation coefficient can measure the
correspondence of the two curves.
0 Quarterly average TSP concentrations plotted against
number of days of rain in quarter.
61
-------
0 A precipitation rose. This can be compared with a
pollution rose for the same site and may explain some
of the apparent directional impacts.
4.3.2 Applicability of Technique
Rainfall is usually associated with a reduction in fugitive
dust emission rates. With this association, the relative effects
of rainfall at different sites or in different urban areas can be
used to identify areas with high fugitive dust impacts. Because
of the washout action of rainfall, quantitative relationships
between rainfall-related TSP reductions and fugitive dust impact
cannot be established.
Analysis of precipitation cannot be used to characterize the
aerosol or to estimate the impact of a specific source or of a
change in emissions.
4.3.3 Relationship to Other Techniques
Seasonal or monthly TSP patterns can be explained with
rainfall analyses. Data for some analyses (such as weekday/
weekend, upwind/downwind, or source emission patterns) can be
stratified for days with rain and days with no rain. By com-
paring the two data subsets, some additional information on
source contributions can be obtained.
Precipitation roses can be compared with wind roses, pollu-
tion roses, element roses, etc. to further clarify directional
impacts. With sufficient data, the effect of rainfall on par-
ticle size distributions (cumulative or species-specific) can
also be assessed.
4.3.4 Resource Requirements
These techniques are relatively simple and do not require
extensive data. Resource requirements are summarized below:
62
-------
Manpower Low
Skill Low
Computer Optional
Data TSP and various
measurements of
precipitation
4.3.5 Example Application
The example is an analysis of TSP concentrations versus time
since rain. TSP readings for 1974 at two sites in Birmingham
were sorted into six classifications of days since rain, then
21
averaged. To facilitate comparison with results from other
areas, the TSP concentrations were normalized by dividing by the
five-week running average concentration (samples were taken every
day). The results are shown in Figure 4-2. Calculations were
performed twice, using alternate amounts of precipitation of 0.01
in. and 0.25 in. to classify a day as one with precipitation.
The curves in Figure 4-2 indicate that concentrations return to
near normal in one day after rain in downtown Birmingham but are
depressed for a second day at an industrial exposure in North
Birmingham.
4.3.6 References
11, 12, 13, 15, 16, 21, 24
4.4 ANALYSIS OF WIND SPEED
4.4.1 Description of Technique
Wind speed potentially affects TSP concentrations in two
contrary manners. As wind speed increases, the effective volume
of air available for dilution increases and, for constant source
strengths, downwind concentrations are inversely proportional to
wind speed. However, emission rates also increase with wind
speed because this is the agent by which soil and dust particles
are naturally entrained. At instantaneous wind speeds up to 12
or 13 mph, almost no dust is entrained, but at higher speeds the
emission rates can become substantial.
63
-------
I.SO
140
ISO
O 1.20
<
* 1.10
w
S 1.00
K
^ o»o
0.80
0.70
oeo
(10
T46)
I 2 9 4 5 OR MORE
NUMBER OF DAYS AFTER PRECIPITATION
a) NORTH BIRMINGHAM
040
I Z 3 4 5 OR MORE
NUMBER OF DAYS AFTER PRECIPITATION
b) DOWNTOWN BIRMINGHAM
Number of observations shown in parentheses
Figure 4-2. Effect of rainfall in reducing TSP concen-
trations at two sites in Birmingham.
Source: Reference 21
64
-------
The main interest in analysis of wind speed is to determine
the relative effects of dilution and wind erosion generation on
TSP concentrations. This is done most simply by plotting TSP
concentration versus 24-h average wind speed for all samples at a
site or for selected data subsets. The characteristic curve
shows high TSP values at extreme high and low wind speeds and
lower TSP values with moderate wind speeds. ' ' ' Dis-
tortions from this shape can be interpreted in terms of the
relative impact of wind-generated sources.
The analysis can be further refined by plotting TSP concen-
tration versus 24-h average wind speed separately for each wind
direction. This analysis provides an indication of wind-gener-
ated emissions by direction. One requirement for directional
analysis is that only data taken on days with persistent wind
directions be used. In this case, wind persistence is defined as
a ratio of resultant wind speed to average wind speed equal to or
greater than 0.9. Variable wind directipn (ratio <0.9) can be an
additional subset for plotting TSP versus wind speed.
Wind speed is often used as a meteorological variable in
correlation and regression analyses. However, because of the
characteristic U-shaped TSP versus wind speed curve, correlations
are usually low and may be insignificant. In some studies, TSP
concentrations have been correlated with wind speed by speed
range.
At least one other technique has been employed to analyze
38
wind speed impact. For a set of two sampling sites, one in an
urban area and one a background sampler upwind of the city, the
ratio of TSP concentrations is calculated for each sampling day.
The sampling days are then ranked according to increasing wind
speed. If the ratio increases consistently with wind speed, this
has been interpreted to mean that extra-urban particulate is a
major contributor to concentrations. Conversely, if the ratio
decreases consistently, major contributors are located between
the two samplers.
65
-------
4.4.2 Applicability of Technique
As with other meteorological analyses, wind speed techniques
assist in identifying source categories but cannot characterize
the aerosol, estimate the impact of a specific source, or esti-
mate the impact of changes in emissions from a source.
4.4.3 Relationship to Other Techniques
If wind speed varies substantially on a seasonal or diurnal
basis, it may explain some of the temporal TSP patterns. The
wind speed variable has been combined with wind direction in some
correlation/regression analyses by using total miles of wind
blown from each direction as independent variables. This is
related to the pollution rose concept.
Because wind speed affects the deposition rate of large
particles, an anlysis which combines wind speed and particle
sizing may provide more information on the impact of wind speed.
4.4.4 Resource Requirements
Wind speed analysis techniques are relatively simple and do
not require extensive data. Resource requirements are as follows:
Manpower Low
Skill Low
Computer Optional
Data TSP and various wind speed
measurements
4.4.5 Example Application
TSP data from Houston and Dallas-Fort Worth were processed
22
into several preliminary displays for further analysis. One
of these displays was plots of average TSP concentrations versus
average wind speed as a function of wind direction. In addition
to plots for the eight compass directions at each hi vol site,
additional plots were generated for all days with variable winds
and for all persistent winds. The plots for one site are shown
in Figure 4-3.
66
-------
N -
i 1
I I T
NE -
I I I I I I
i r
i i i i i
SH -
i r
5E -
S -
I I I
V -
fl i 0 i c > 0 ' E ' F '
flVG TSP VS flVG WIND SPEED
91V SPMPLES 173P SflMPLES
HIVOL 40600Q2B NWS 12960
MONTHS JRN - DEC YEflRS 70 71 72 73
Figure 4-3. Plots of average TSP versus wind speed as a function
of wind direction.
Source: Reference 22
67
-------
4.4.6 References
13, 15, 16, 21, 29, 35, 37, 38, 39
4.5 TRAJECTORY ANALYSIS
4.5.1 Description of Technique
This technique attempts to identify the approximate origin
of air parcels sampled and to determine portions of the TSP
sample that have been transported. A backward trajectory from
the sampling site is calculated, usually with upper air (850 or
900 millibar level) rather than surface wind speed and direction
measurements.
For 24-h TSP samples, there are actually a number of dif-
ferent trajectories that reach the sampler throughout this time
period. However- none of the techniques to date have assumed
more than a single trajectory; the resulting air parcel that is
traced is assumed to arrive at the site at either the midpoint or
end of the 24 hours.
Upper air wind measurements are not taken continuously, so
some interpolation between the 12-h interval reported readings
must be made. Also, because of the distances involved in 24-h
air movements, wind data from several upper air meteorological
stations should be used and appropriate values interpolated for
locations between these stations. The National Weather Service
prepares and distributes via teletype and facsimile circuits 6-h
trajectory vectors at the 1000, 850, and 700 mb levels for-vari-
ous stations in the U.S. EPA has a trajectory analysis program
available which linearly interpolates between the three closest
meteorological stations, computes backward trajectories in 2-h
increments, and gives values for the u and v components of the
wind at each intermediate location.
After identifying the approximate area of origin of the air
parcel, the next step is to find TSP and possibly sulfates and
nitrates data at a nearby location. These ambient concentrations
should be for the day prior to the downwind sample. If a number
68
-------
of data pairs of these upwindbackground versus downwind-impacted
areas samples are generated, correlations and other comparisons
of the data pairs can be performed.
This technique is based on several estimates and assump-
tions. Therefore, its results should only be used qualitatively-
4.5.2 Applicability of Technique
Trajectory analysis can estimate the portion of a measured
TSP concentration due to long-range transport, which helps de-
scribe the aerosol. Also, if transported particulate is con-
sidered as a source category, it can help to determine contrib-
uting source categories. However, it cannot quantitatively
estimate the impact of a specific source or the impact of changes
in emissions from a source.
4.5.3 Relationship to Other Techniques
Transported particles are those in the small particle size
range, generally less than 2 to 3 microns diameter. Therefore,
particle sizing and possibly chemical analysis of the <3 urn
fraction may increase the information gained by trajectory anal-
ysis. Also, much of the transported material may be secondary
particulate (sulfates, nitrates, ammonium, and organic com-
pounds) , so chemical analyses of upwind and downwind samples
would aid in establishing the amount of transport.
Trajectory analysis is similar in concept to correlation/
regression analyses with wind directions as independent vari-
ables. If performed in a comprehensive manner, the latter anal-
yses are much more quantitative.
4.5.4 Resource Requirements
Availability of wind measurements that appropriately reflect
air mass movements within the mixing layer in the vicinity of the
sampling site is the key to this technique. Evaluation of data
by a qualified meteorologist is a prerequisite. Other resource
requirements can be summarized as follows:
69
-------
•Manpower Moderate
Skill Moderate
Computer Probably required
Data TSP and upper air wind
speed and direction
measurements
4.5.5 Example Application
This "descriptive" trajectory analysis was done after a
correlation analysis of TSP concentrations with wind direction in
Albany, New York showed that higher particulate concentrations
were associated with southerly winds and that lower concentra-
39
tions were associated with winds from the northwest. Twenty-
four hour trajectories for 10 days during the study period,
representative of various wind directions, were generated using
National Weather Service 6-h vectors. In all 10 cases, predicted
trajectories were verified satisfactorily by comparison with
local wind records for the days in question.
The resulting trajectories and corresponding TSP concen-
trations are shown in Figure 4-4. They showed that air masses
from the south had passed through the Washington, D.C.-Phila-
delphia-Allentown-Bethlehem areas or New York-Long Island during
the preceding 24 hours, and that air masses from the Northwest
passed through low population areas of Uupper New York State and
Canada.
4.5.6 References
33, 38, 39, 40
70
-------
TERMINUS SUSPENDED
DATE PARTICULATES
6/12/72
6/13/72
6/14/72
7/02/72
7/01/72
7/04/72
6/O3/72
6/16/72
6/20/72
6/19/72
Figure 4-4. Air flow trajectories for 10 nonprecipitation
days terminating in Albany.
Source: Reference 39
Reprinted by permission of the publisher.
71
-------
5.0 ANALYZING EMISSIONS DATA
Previous chapters have presented several techniques for
analyzing suspended particulate data. Such techniques can
provide evidence implicating source categories possessing certain
emission characteristics, but additional techniques must be used
to quantify those emission characteristics. In this chapter
three such techniques are discussed:
0 Emission inventorying
0 Microinventorying
0 Diffusion modeling
5.1 EMISSION INVENTORYING
5.1.1 Description of Technique
Procedures for compiling an emission inventory are amply
described in a number of EPA publications and need not be re-
28 41 42 43
peated in detail here. ' ' ' In brief, the technique is to
inventory the major sources of particulate emissions in terms of
their magnitude, degree of control, and location.
Emission sources have been classified a number of different
ways. One of the most common has been to use the following
classification scheme:
Industrial process
Fuel combustion
Transportation
Electricity generation
Incineration
Miscellaneous
72
-------
Table 5-1 lists subdivisions for each of these source categories
and the type of source (point, area, or line) included in each.
Point sources are any stationary source emitting more than some
designated minimum (usually 25 ton/yr) of particulate matter. An
area source is a collection of sources whose individual emission
rates are small but whose collective impact may be large. A line
source is a source that can be geometrically described best as a
line (e.g., a highway). The line source description is used
primarily in carbon monoxide microscale analyses and is rarely
used in particulate emission inventories.
In addition, there has been a growing trend toward making
finer distinctions among various types of "unconventional sources"
of particulates. Fugitive emissions are particulate matter
emitted from industrial facilities without the benefit of flow or
direction control. Fugitive dust, on the other hand, is a type
of particulate emission driven airborne by man's activity or by
the effect of wind erosion upon exposed.surfaces. A summary of
the various significant categories of fugitive dust sources is
also presented in Table 5-1.
The major deficiencies of the technique include the follow-
ing:
0 Emission factors represent typical emission rates and
may not apply to the specific sources of concern
0 Methods for apportioning area and fugitive dust sources
to geographical subdivisions of a region are subject to
large potential errors
0 Inventorying, by itself, tells the analyst nothing
about air quality; other techniques, such as diffusion
modeling, must also be used
5.1.2 Applicability of Technique
Emission inventorying, by itself, does not help to charac-
terize the aerosol. Rather, the fundamental utility of the
technique is to identify the major sources and source categories
which may contribute to measured or predicted TSP concentrations.
73
-------
Table 5-1. EMISSION SOURCE CATEGORIES
Source category
Subdivisions
Source
description
Industrial process
Fuel combustion
Transportation
Electricity generation
Incineration
Miscellaneous
Chemical manufacture
Food/Agriculture
Primary metals
Secondary metals
Mineral products
Petroleum industry
Wood products
Evaporation
Metal fabrication
Leather products
Textiles
In-process fuel
Fugitive emissions
Other
Internal combustion
External combustion
Highway vehicles
Lt duty gas autos
Lt duty gas trucks
Motorcycles
Hvy duty gas trucks
Hvy duty diesel trucks
Off-highway vehicles
Rail locomotives
Vessels
Aircraft
Point, area
Point, area
Area, line
Point
Solvent evaporation
Fires
Fugitive dust
Anthropogenic (unpaved
roads, agri tilling,
const activities,
street dust, off-road
mtr veh activities,
inactive tailing piles)
Wind erosion (unpaved rds,
agri fields, disturbed
soil surfaces)
Point, area
Point, area
Source: Adapted from References 41 and 44.
-------
Recognizing the unique and vital role that this technique plays
in any effort to control TSP concentrations, there are still a
number of deficiencies associated with this technique that the
analyst needs to be aware of. The most important are the fol-
lowing :
0 The inventory is usually limited to man-related sources
and may downplay the role of natural sources. Ques-
tions which are still the subject of research concern
the roles of naturally occurring windblown dust and sea
spray, and secondary particulates transported over long
distances.
0 The types of sources which are inventoried have changed
over time as new information has become available
(e.g., industrial fugitive emissions and fugitive dust
from such source categories as paved roads have only
recently been included in inventories).
0 Actual emissions from industrial facilities may differ
from estimated emissions by a rather large amount.
Fluctuating loads may cause actual collection control
efficiencies to differ from measured efficiencies.
Likewise, malfunctions and deterioration in equipment
may occur and cause significant deviations from cal-
culated emissions.
This technique is a prerequisite for determining the impact
of specific sources upon air quality, for that impact cannot be
determined without first identifying the locations and approxi-
mate magnitudes of those sources' emissions. Similarly, the
technique is a prerequisite for assessing the effect of changes
in emissions upon TSP air quality.
5.1.3 Relationship to Other Techniques
Most emission inventory data are prepared in terms of annual
emissions. As a result, quantitative comparisons between air
quality and emission data are normally possible (depending upon
the availability of data) only in terms of annual trends. How-
ever, in some cases data on temporal variations in emissions from
specific sources or source categories may be available. Tech-
niques for relating short-term variations in air quality and
emissions are discussed in Section 1.5.
75
-------
Emission inventorying relates to the analysis of spatial
patterns of air quality, in that upwind/downwind analyses begin
with the assumption that a certain source may be an excessive
contributor to measured air quality and in that the shapes of
pollutant roses initially lead the analyst to check the current
emission inventory for major particulate sources which are lo-
cated in the direction suggested by those shapes.
The major relationships between this technique and the
analysis of the effect of meteorological variables relates to
rainfall. Certain emission sources, such as fugitive dust sources!
vary in magnitude inversely with rainfall. Where rainfall is
shown to have a major effect upon TSP values at a specific site,
then it could be inferred that fugitive dust sources are a con-
tributing source.
Diffusion modeling is dependent upon emission inventory data
since the emission rates from modeled sources are necessary input
values.
Emission inventories are also used in the chemical element
balance technique to approximate the contribution of sources for
which suitable tracers cannot be found. Most industrial pro-
cesses fall within this category. Likewise, where detailed
elemental concentrations of a source's emissions are available,
such data can be used with element roses and interspecies cor-
relations at a given sampling site to verify that source's impact
upon TSP values at that site. Occasionally, as in the case of
windblown dust, data concerning the particle size distribution of
a source category is available. When such is the case, sampling
site particle size distribution data can be used to implicate
major contributing source categories.
5.1.4 Resource Requirements
Resource requirements for this technique are summarized
below:
76
-------
Manpower High
Skill Low-moderate
Computer Optional
Data Emission factors, source
activity rates and locations,
climatological data
5.1.5 Example Application
Examples of the application of the emission inventorying
techniques can be found in any major city. As a result, a de-
tailed example will not be presented. Rather, one brief example
from Philadelphia will be discussed.12'30
The Metropolitan Philadelphia Interstate AQCR is a major
industrial area encompassing over 1,300 individual plants in
Pennsylvania, New Jersey, and Delaware. National emission data
system (NEDS) data for 1972 provided a breakdown for most of the
source categories in the AQCR (see Table 5-2). Locations of the
50 largest sources in the Philadelphia area are graphically
displayed in Figure 5-1.
The NEDS data did not include any estimate of the emissions
from possible fugitive dust sources. The authors used data from
other recent studies of dirt roads, dirt airstrips, construction,
and agricultural tilling. Subsequent calculations indicated that
the fugitive dust sources contributed over 10 times the total
tonnage as the traditional inventoried sources. The majority of
fugitive emissions were calculated to be contributed by dirt
roads and construction activity (see Table 5-2).
5.1.6 References
12, 28, 41, 42, 43, 44, 45
5.2 MICROINVENTORYING
5.2.1 Description of Technique
The term "microinventory" refers to the procedure of esti-
mating annual particulate emissions in a relatively small area
surrounding a high volume sampler site. When first applied, this
77
-------
Table 5-2. NEDS TSP EMISSIONS FOR PHILADELPHIA AQCR
Source category
Fuel combustion
External
Residential
Electrical
Industrial
Commercial-Institutional
Internal
Industrial process
Chemical
Food/ Agriculture
Metals
Mineral
Petroleum
Other
Solid waste disposal
Government
Residential
Commercial-Institutional
Industrial
Transportation
Gasoline
Diesel
Aircraft
Vessels
Emissions, ton/yr
Point Area Total
(133,440)
133,410
-
12,954
114,341
6,115
30
(61,322)
1,810
1,665
9,195
22,521
26,116
15
(8,989)
8,651
-
5
333
_
-
-
-
—
(31,463)
31,463
6,334
-
19,729
5,400
—
—
-
-
—
-
-
—
(6,989)
-
1,415
4,906
668
(23,074)
17,857
2,912
1,022
1,283
(164,903)
164,873
6,334
12,954
134,070
11,515
30
(61,322)
1,810
1,665
9,195
22,521
26,116
15
(15,978)
8,651
1,415
4,911
1,001
(23,074)
17,857
2,912
1,022
1,283
Fugitive emissions
Dirt roads
Dirt airstrips
Construction
Agricultural tilling
Totals
203,751
(2,808,640)
1,954,629
2
850,594
3,415
2,870,166
(2,808,640)
1,954,629
2
850,594
3,415
3,073,917
Source: Reference 30
78
-------
4443 -
4440 -
4493 -
4490 -
4423 -
4490 -
4413 -
KEY
Q >1000 TPY
B 300-1000TPY1
H 10O-5OO TPY
4410 -
473
410
4*3
I
490
I
4*3
I
300
I
SOS
Figure 5-1. Location of major sources,
Source: Reference 12
79
-------
technique was rather simple, but it has since evolved into a de-
tailed compilation of sources and their distance and direction
46,47
from a sampler.
Although a relatively small area is inventoried, the area
must be large enough to include the location of most of the
sources significantly affecting air quality at the site being
investigated. The most recent approach uses a five mile radius
(from the sampling site) for point sources and one mile for area
sources.
Distances and compass directions from the sampling site to
impacting sources should also be calculated. For point sources,
this is done out to the five mile radius limit. For area sources,
however, such an approach is impractical. Instead, the area
within a one mile radius of the sampling site is divided into
four 90 degree quadrants and three radial distances (as shown
below).
Within Sector 1 (1/4 mile radius from the site), the major
sources very close to the monitor are described as to their
distance and direction from the monitor. Emissions from all
sources in that sector are then tabulated. Similar tabulations
are made for sources within the 1/4 to 1/2 mile radii and the 1/2
to 1 mile radii with these radii being divided into quadrants.
Thus, a total of nine sectors are considered separately for the
inventory of area source emissions.
Fugitive dust sources are generally the dominant area
source within one mile. These sources include dust from paved
roads, unpaved roads, parking lots, cleared areas, construction,
rail yards, agricultural tilling, and so on. Industrial processes
80
-------
smaller than the point source cutoff of 25 ton/yr actual emis-
sions are also included. The area source inventory also includes
combustion sources (such as space heating, small boilers, and
incinerators), which were not included in the point source inven-
tory, and transportation exhaust sources.
A procedure for determining the locations of the sources and
their activity levels should be developed for each urban area to
minimize time required in the field. A presurvey is suggested if
the sites to be inventoried are not located in the same region as
the staff performing the inventory. This presurvey requires a
one- to two-day general tour of the sites to be inventoried.
Table 5-3 indicates the information obtained during the pre-
survey .
Following this presurvey, a 1/4 mile radius map of the road
network is drawn to scale using USGS maps. Then the distances
and directions from the sites to the point sources are calcu-
lated. Aerial photos can be used to prelocate or preinventory
cleared areas or other fugitive dust sources. The return, or
survey visit, consists of a detailed visual inspection of the
vicinity of the site. The condition of the streets (dirt level
and presence of paving, curbs, etc.) is described, so that appro-
priate emission factors can be selected later. The location and
activity level of unpaved or gravel parking, for example, is
estimated by the best available means (talks with the owner or
attendant, or estimated lot capacity and turnover rate). The
locations of very close point sources (as calculated from UTM
coordinates) are confirmed. The mile area is checked for omis-
sions from the point source inventory. Any obvious, unusual
characteristics that may not have been considered accurately in
the area source allocation scheme (such as a concentration of
coal or wood fired space heaters and fireplaces) are noted, and
appropriate adjustments are made to the inventory. The survey
usually requires two persons for about one-third to one-half day
in the field per site, depending upon the complexity of the area
and the number of fugitive sources.
81
-------
Table 5-3. INFORMATION OBTAINED IN PRESURVEY
U.S. Geological Survey maps of the area from a local engi-
neering supply store
Land use maps from the local Council of Governments (COG)
Traffic data from local COG or city/state transportation
agency
TSP annual air quality data and the site UTM coordinates
from the local air pollution control agency
Gridded inventory of area combustion sources from the local
air pollution control agency
Point source inventory, including annual TSP emissions over
25 ton/yr and UTM coordinates, from the local air pollution
control agency
Precise location of hi vol including height and distance to
all major sources (such as roads, parking lots, etc., within
60 m) obtained during site tour
Photographs of site and surrounding area obtained during
site tour
Aerial photographs (optional) usually available through COG
Source: Reference 45
82
-------
Standard emission factors are then applied to activity
factors such as vehicle miles traveled (VMT), acres of tilled
land, etc./ to compute the emission estimates for each sector.
Predominant land use in each sector can be estimated by
visual inspection or through land use maps. For further accu-
racy, the 1/4 mile radius (Sector 1) can be divided into quad-
rants with assigned land use classifications. Such information
may be useful for future planning and projection activities.
Table 5-4 indicates the land use categories and classification
criteria which have been found to be useful.
5.2.2 Applicability of Technique
This technique does not contribute to an improved charac-
terization of the aerosol. However, it can qualitatively esti-
mate the relative impacts of fugitive dust sources and point
sources upon measured air quality. This can be accomplished by
performing intersite comparisons of emissions for sites with
equivalent TSP concentrations. One recent application used
microinventory data at several sites to produce a multivariable,
45
empirically-based predictive equation.
When combined with standard diffusion models to provide
refined detail in the vicinity of receptors with measured TSP
data, the technique can be used to assess the impact of a change
in emissions from given sources. But, by itself, the technique
will neither determine the impact of a specific source or changes
in emissions from a specific source.
5.2.3 Relationship to Other Techniques
No attempt has yet been made to compare temporal variations
in TSP concentrations with temporal variations of emissions
within microinventoried areas, and the state-of-the-art suggests
that such comparisons will not be possible except in terms of
long-term trends.
The technique is potentially quite supportive of and com-
plementary to techniques which analyze spatial patterns of TSP.
83
-------
Table 5-4. LAND USE CATEGORIES AND
CLASSIFICATION CRITERIA
Descriptor
Characteristics
Undeveloped
Agriculture
Light residential
Dense residential
Suburban commercial
Central commercial
Light industry
General industry
Heavy industry
Airport
No significant activity;
includes parks and pasture
land
Active farming
Four or less dwelling units/
acre
Greater than four dwelling
units/acre
Retail businesses in strip
development or shopping
centers
Central business district
Metal fabrication, warehousing,
trucking, etc.
Controlled industrial processes
Steel mills, foundries, coking,
etc.
Municipal airport
Source: Reference 45
84
-------
For example, this technique can be used to classify sites by type
for stratification by concentration. Likewise, is can be used to
search for sources implied by the shapes of pollutant roses.
Intersite correlations can be cross-checked with this technique.
Where a correlation is high but the microinventories are quite
dissimilar, then a reasonable inference would be that sources
outside the microinventoried area are the primary cause of the
problem. Where correlations are high and microinventoried emis-
sions are equivalent, then sources within that area are probably
the primary causes.
Other than being used in conjunction with seasonal varia-
tions in activity patterns (in terms of their effect upon fugi-
tive dust emission factors), this technique bears no strong
relationship with techniques which assess the effect of meteoro-
logical variables.
The resolution of diffusion modeling grid systems could
theoretically be improved by using microinventories which are
45
converted to small detailed grid systems as shown in Figure 5-2.
A successful application of this technique has not yet been
reported in the literature, however.
No attempts have yet been made to relate microinventory data
to particle size, chemical, elemental, or morphological data;
however, it is possible (although challenging) that the technique
could be refined to use element-specific emission factors to
produce element-specific emission totals for the microinventoried
areas. These totals could then be correlated with measured
concentrations for that element at a series of sampling sites.
5.2.4 Resource Requirements
Resource requirements for the microinventory technique are
summarized -below:
85
-------
4
/
f
/
/
1
1
1
1
\
\
1
I
i
\
V
\
\
\
N
4
X1
X
,
/
1
I
\
\
\
X
.^ -
^ —
1
V
\
N
•s.
^». «,
V
"*" *>,
X
-~ ^_
***,
™" X
\
\
t
J
• **
.,
\
\
\
1
1
f
f
^''
^
\
V
\
\
\
\
\
t
1
1
1
1
1
j
/
/
*
9 O.BK 0.5K
1.0 K
Figure 5-2. Converting microinventory area to diffusion
model grid.
Source: Reference 45
86
-------
Manpower Moderate-high
Skill Low
Computer Not required
Data Same as for emission inventory
plus more detail in vicinity
of samplers
5.2.5 Example Application
The example is drawn from a recent report produced for
EPA. The example microinventory is summarized in Tables 5-5
through 5-7 and Figures 5-3 and 5-4 in terms of the following
data:
0 Description of the site
0 SAROAD code
0 Location
0 Monitor height
0 Land use description by sector
0 Localized sources within 200 feet of the sampler
0 Air quality data
0 A USGS map of the one mile radius area around the
sampler
0 A map of the 1/4 mile radius area around the sampler
0 Point source summary in terms of emissions, distance,
and compass direction from the sampler
0 Area source summary in terms of activity rate and emis-
sions by sector for each source category
5.2.6 References
13, 45, 46, 47, 48, 49, 50
5.3 DIFFUSION MODELING
5.3.1 Description of Technique
An atmospheric simulation model can be defined as a mathe-
matical description of the transport, dispersion, and trans-
formation processes that occur in the atmosphere, in its sim-
plest form, such a model relates pollutant concentrations (x) to
pollutant emission rates (Q) and a background concentration (b),
as in the following equation:
87
-------
Table 5-5. DESCRIPTION OF MICROINVENTORY SITE
DESCRIPTION OF SITE
SAROAD code - 26 2380 005 HOI
Location - 6402 East 37th,
Kansas City, Missouri
Monitor height - 35 ft, on roof of
fire station
Land use, by sector -
la light industry
Ib light residential
Ic undeveloped
Id light residential
2 light industry
3 general industry
4 undeveloped
5 light residential
6 light industry
7 light industry
8 dense residential
9 dense residential
10 general industry
11 dense residential
12 dense residential
13 dense residential
Localized sources, within 200 ft of monitor -
Source Distance Description
Stadium Drive
Unpaved parking lot
Unpaved alley
Fremont
55 ft 7870 ADT, dirty, uncurbed
105 ft 0.1 acres, 10 cars
100 ft 5 cars
60 ft 75 ADT, dirty, uncurbed
Air quality data -
Year
Annual geometric mean, ug/m'
Source: Reference 46
No. of samples
1977
1976
1975
1974
1973
85
89
86
89
101
53
54
47
51
88
-------
Figure 5-3. One mile radius around sampling site,
Source: Reference 46
89
-------
1,7. Unpaved pkg lot, 0
2. Cleared area, 0.1 ac
3,6. Unpaved pkg lot, 0.1 ac,
4. Unpaved storage area
5. Unpaved pkg lot, 0.3 ac,
8. Unpaved pkg lot, 1.7 ac,
9. Unpaved pkg lot, 1.5 ac,
10. Unpaved storage area, 2
ac, 10 cars
5 cars
0.2 ac
25 cars
35 trucks
150 cars
0 ac
11. Unpaved storage area, 8.1 ac
Scale 1" = 400
Unpaved pkg lot, 0.8 ac, 20 cat
Unpaved pkg lot, 0.7 ac, 50 cai
Unpaved storage area, 3.4 ac
Unpaved storage area, 2.3 ac
Unpaved road, 400 ft, 5 cars
Unpaved road, 300 ft, 10 cars
18,19. Unpaved alley, 750 ft, 5 ci
20. Unpaved road, 675 ft,-5 cars
21. Unpaved road, 600 ft, 5 cars
12
13
14
15
16
17
Figure 5-4. One-quarter mile radius around sampling site.
Source: Reference 46
90
-------
Table 5-6. MICROINVENTORY POINT -SOURCE SUMMARY
POINT SOURCE SUMMARY
Site: 6402 East 37th
Plant Emission Distance Compass
number level, t/yr from site, mi direction,
o
26 48 .39 180
25
24
23
22
16
15
13
14
9
893
98
43
1,233
249
41
341
327
46
1
2
2
3
4
4
4
4
5
.7
.4
.7
.5
.3
.4
.7
.8
.1
35
25
20
10
5
0
350
'o
330
Source: Reference 46
91
-------
Table 5-7. MICROINVENTORY AREA SOURCE SUMMARY
AREA SOURCE SUMMARY
Site: 6402 East 37th
Source category
COMBUSTION :
Residential fuel
Coiran/Ind fuel
Incinerators
Rail/Air
Auto exhaust
INDUS PROCESSES!
Ind processes
FUGITIVE DUST:
Railroad yards
Clean streets
Contti streets
Unpaved roads
Cleared areas
construction
Agriculture
Storage areas
Unpaved pkg lots
Activity
rate
9491 pop
8081 emp
97.4xl06 VMT
4 sources
78 ac
94.5xlOC VMT
3.0xlO«> VMT
2.7x10* VMT
0.1 ac
16 ac
25.4 ac
Emissions by sector, ton/yr
1
0.2
0.5
1.9
1
0.4
10.6
1.6
1.2
nea
3.2
4.8
2
0.1
0.4
0.6
2.0
3.6
3
0.1
0.4
0.6
10
1.2
3.6
4
0.1
0.4
0.2
1.4
5
0.1
0.4
0.6
3.6
6
0.5
1.5
27.0
3.0
152.3
9.6
12.0
7
0.5
1.5
16.7
0.5
95.6
8
0.5
1.5
3.2
1.6
0.7
14.1
19.1
7.4
9
0.5
1.5
13.5
1.0
70.8
19.1
Total
2.l>
B.I
64.3
13.6
7.6
355.6
49.4
1.2
3.2
24.2
vo
to
Recap
COMBUSTION
JND PROCESSES
FUGITIVE DUST
mci^ml
Emissions by sector, t/yr
1
2.6
1.0
21.8
as.*
2-5
4.0
10.0
15.4
20. -I
6-9
68.4
2.6
404.2
«!7£..2i
Total
75.0
13.6
441.4
sao.o
-------
x = kQ+b (eq.2)
The constant k is a function of atmospheric conditions and the
spatial relationship between source and receptor. Atmospheric
simulation models are ultimately concerned with the variability
of k, and of emission rates and their impact on pollutant con-
centrations .
Modeling procedures have been generally categorized into
four generic classes: Gaussian, numerical, statistical, and
physical. Gaussian models are generally considered to be
state-of-the-art techniques for estimating the impact of non-
reactive pollutants such as total suspended particulates.
The extent to which a specific model is suitable for use
depends upon a number of factors, among which are the following:
0 The detail and accuracy of the data base
0 The meteorological and topographic complexities of the
area
0 The technical competence of the persons performing
such modeling
0 The resources available
0 The situation being modeled (i.e., the type and
number of sources, and the time frame of concern)
For assessing the impact of point sources over all typical
averaging times, a number of models are available. These models
are summarized in three recent EPA publications, and need not be
repeated in detail here. ' ' Where refined analyses are
required and no significant meteorological or topographical
complexities exist, EPA has recommended the use of the Single
Source (CRSTER) Model. Where such complexities do exist, EPA has
recommended that each complex situation be treated on a case-by-
case basis with the assistance of expert advice.
The Climatological Dispersion Model (CDM/CDMQC), the Air
Quality Display Model (AQDM), and the Texas Climatological Model
(TCM) have been recommended for evaluating the long-term impact
of urban multisource complexes. In the case of multisource
93
-------
short-term average situations, the Gaussian Plume Multiple Source
Air Quality Algorithm (RAM) and the Texas Episodic Model (TEM)
have been recommended. The statistical conversion mechanisms of
COM and AQDM can also be used in urban multisource areas; how-
ever, their use has not been recommended for situations dominated
by large point sources.
AQDM, CDM, and the other above-described models may also be
inappropriate for areas which may be dominated by fugitive dust
sources. In such locations, models such as the Hanna-Gifford
Model, a modified CDM/Rollback model, the Atmospheric Transport
and Diffusion Model, or other techniques may be more applic-
28
able. A fact which must be considered is that models with a
wide applicability (such as AQDM and CDM) are not generally
available for dealing with long-range transport, deposition,
windblown particulate matter, and certain unique meteorological
circumstances.
Although specific model impact requirements vary with the
model being applied, the general input requirements can be clas-
sified as source, meteorological, receptor, and background air
quality data. Typical point source emission data includes the
emission rate, stack height, stack diameter, stack exit velocity
and temperature, and location. Typical area source data include
the emission rate prorated over a predetermined rectangular grid
network, a representative average emission height, and locational
coordinates. Meteorological data typically include such vari-
ables as wind direction, wind speed, atmospheric stability, and
mixing height. Efforts must be made to assure that these mete-
orological data are as representative of the transport and
dispersion conditions in the modeled area as possible. Receptor
sites are usually chosen to provide an adequate degree of spatial
resolution to TSP values and/or to provide an estimate of the
peak concentration that would be caused by the modeled source(s) •
It is commonly assumed that the annual mean background TSP
concentration is 30-40 ug/m over much of the Eastern United
States. However, such an assumption is not necessarily valid for
94
-------
short-term situations or for other parts of the U.S. Methods for
determining appropriate background concentrations are discussed
in a recent EPA publication.
Once estimated values are generated, the applied model
should be validated and/or calibrated. As one aspect of the
validation process, statistical methods including skill scores,
contingency tables, correlation analysis, time series and spatial
analysis, and others are often attempted. Calibration of a
model, the process of identifying systematic errors and applying
a correction factor, usually involves the application of regres-
sion analysis or a similar statistical technique. The statis-
tical reliability of such procedures is limited by uncertainties
associated with input variables and the normally small number of
"calibration" sites. Calibration may be the only alternative for
improving estimated concentrations, but it should only be applied
to long-term models at the present time.
5.3.2 Applicability of Technique
Commonly used models characterize the aerosol only in terms
of absolute TSP concentrations over varying time periods at a
number of receptor locations. However, there has been a recent
trend toward developing models which incorporate a fallout func-
28
tion for larger particles.
In theory, one of the attributes of the diffusion model is
its ability to calculate the contribution of various source
categories to predicted air quality. This capability is limited
in practice, however, by the facts that not all major emission
sources categories are always identified (e.g., street dust prior
to circa 1975), that the emission factors and activity rates
associated with those sources are of varying degrees of accuracy,
and that the procedures for allocating area source emissions to
gridded areas can be subject to significant error.
One of the most important uses of diffusion modeling is to
assess the air quality impact of emissions or changes in emis-
sions from a specific source or identified set of sources. The
95
-------
increasing use of diffusion models, and the limitations asso-
ciated with the other techniques discussed in this digest, sug-
gests that there are no better techniques currently available for
performing this task.
5.3.3 Relationship to Other Techniques
Diffusion modeling, largely because of the resources re-
quired for implementation, is rarely related to techniques which
analyze the temporal patterns of particulate air quality. There
is, however, a relationship with those techniques which analyze
spatial patterns. It has already been noted that most diffusion
models assume level terrain conditions. Other techniques, such
as intersite correlations and pollution roses, help to compensate
for the errors this assumption induces by providing supplementary
information. No quantitative relationships have yet been devel-
oped , though.
Meteorological conditions are, of course, used as input
variables to diffusion models, but they bear no other significant
relationship to that technique. Similarly, emission inventory
data are required as a model input but bear no other distinct
relationship. Microinventories have the potential of providing
refined emission data in the immediate vicinity of samplers.
In one recent study, the Hanna-Gifford diffusion model was
used as an ingredient in the development of a new "factor model"
54
of air pollution. No other relationships between diffusion
modeling and physical, elemental, chemical, or morphological data
have been demonstrated. The techniques provide dissimilar ways
of approaching the same problem and can be used to supplement
each other.
5.3.4 Resource Requirements
Resource requirements for diffusion modeling vary consid-
erably depending upon the specific model used. Screening pro-
cedures discussed in a recent EPA publication are quite simple
96
-------
and are not resource intensive. More sophisticated models,
such as the RAM, are much more resource intensive. The following
summarizes the requirements related to urban multisource models:
Manpower Moderate
Skill High
Computer Necessary
Data Emission inventory, meteor-
ological data, TSP measurements
Assumes that emission inventory data have already been obtained.
5.3.5 Example Application
Examples of the application of specific diffusion models can
be found for any major city. What follows is a very brief de-
scription of how one model was applied to a hypothetical situa-
tion. The example, an application of AQDM, is drawn from the EPA
publication entitled Air Quality Analysis Workshops: Volume I—
i 44
Manual.
The Air Quality Display Model (AQDM) was used to perform the
air quality analysis for County X. Figure 5-5 illustrates the
particulate air quality computed for 1975 and 1985 as plotted by
a computerized routine. Table 5-8 is an example of the computed
air quality from the AQDM output tables. In making these com-
putations, it was assumed that no new control programs were in
force. Only existing regulations and Federal new source per-
formance standards were assumed to be in effect. Source com-
pliance data, where available, were used to determine actual
emissions.
It is evident from Figure 5-5 that there were several areas
in the county exceeding the primary NAAQS for particulates in
1975 and that there were much more widely spread violations of
3
the secondary standard of 60 ug/m . By 1985, growth and devel-
opment has caused significant increases in the geographical
extent of primary and secondary standard violations.
97
-------
1975
00
1985
Figure 5-5. County X particulate air quality in 1975 and 1985,
Source: Reference 44
-------
Table 5-8.
COMPOSITE OF COMPUTED AIR QUALITY FOR
COUNTY X FROM AQDM
RECEI-TOR COHCEMTRATJGN DATA
EXPECTED ARITHMETIC MEAN
RECEPTOR
NUMBER
1
2
3
4
5
6
7
e
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
33
39
40
RECEPTOR LOCATION
(KILOMETERS)
HOf-.TZ VERT
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
700.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
705.0
710.0
710.0
710.0
710.0
710.0
710.0
710.0
710.0
3710.0
3715.0
3720.0
3725.0
3730.0
3735.0
3740.0
3745.0
3750.0
3755.0
3760.0
3765.0
3770.0
3773.0
2730.0
3785.0
3710.0
3715.0
3720.0
3725.0
3730.0
3735.0
3740.0
3745.0
3750.0
3755.0
3760.0
3765. d
3770.0
3775.0
3780.0
3785.0
3710.0
3715.0
3720.0
3725.0
3730.0
3735.0
3740.0
3745.0
(MICKOGRAMS/CU .
1975
PARTICULARS
42.
43.
*43.
44.
44.
45.
4C.
46.
45.
44.
44.
43.
43.
42.
42.
42.
43.
43.
44.
44.
45.
46.
47.
47.
46.
45.
44.
44.
43.
43.
42.
42.
44.
44.
45.
45.
46.
48.
49.
43.
KITER)
1980
P7.RTICULATES
44.
45.
47.
47.
49.
50.
50.
50.
48.
48.
47.
46.
46.
45.
44.
44.
45.
46.
47.
48.
49.
51.
52.
51.
50.
49.
48.
47.
46.
46.
45.
44.
47.
48.
49.
50.
50.
53.
55.
53.
1985
PARTICULATES
45.
46.
47.
47.
48.
50.
50.
50.
49.
48.
47.
46.
46.
45.
44.
44.
46.
47.
48.
48.
49.
51.
52.
51.
50.
49.
48.
47.
46.
46.
45.
44.
48.
49.
49.
50.
51.
53.
55.
53.
Source: Reference 44
99
-------
An analysis of the TSP contribution by the various sources
at five of the receptors used in the example is given in Table
5-9.
5.3.6 References
11, 13, 28, 44, 51, 52, 53, 54
100
-------
Table 5-9. SOURCE CONTRIBUTION ANALYSIS FOR COUNTY X
Point Source
Compliance Status
Receptor 119
Calculated Air
Quality (ug/m3)
* Reduction for:
Primary Standard
Secondary .Standard
Source Contributions (%)
Points
Stone Quarry *1
2
3
4
5
6
7
Steel Mill #1
2
Lead Smelter
Steam Plant 01
2
Brick Plant #1
2
Concrete Plant #4
6
Areas
»10
77
83
108
112
Background
1975
70
-
15
1.7
1.5
27.2
2.1
1.2
1.4
3.2
1.0
3.6
2.2
42.6
87.7
1980 1985
86 86
13 12
30 30
2.0 2.0
1.7 1.8
35.1 32.4
1.1
1.3 1.5
3.0 3.7
3.8 3.8
2.4 2.7
34.9 35.0
84.2 84.0
Receptor 132
1975 1980 1985
57 66 68
-
10 12
8.4 10.0 9.8
1.2 1.4 1.4
1.6 1.8 1.8
1.0
13.7 18.0 16.1
3.2 2.8 3.7
1.5 1.6 1.6
5.4 5.3 6.6
1.0 1.0
52.5 45.2 44.3
87.5 88.1 86.3
Receptor
1975
73
-
17
2.6
3.2
2.0
9.5
1.4
1.0
7.5
7.8
2.2
2.2
1.8
1.9
1.4
1.1
41.3
86.9
1980
82
8
26
4.4
5.2
2.6
13.6
1.3
1.0
5.1
5.3
1.6
2.3
1.1
1.4
1.6
2.0
1.9
1.5
36.8
86.7
151
1985
83
10
28
4.3
5.0
2.6
12.2
1.2
1.3
5.8
6.0
1.3
2.7
1.0
1.4
1.6
2.4
1.8
1.5
35.9
88.0
Receptor 171
1975 1980 1985
55 62 63
.
3 S
1.3 1.6 1.6
6.1 7.5 7.4
1.4 1.7 1.7
4.9 6.6 5.9
8.0 S'.O 8.9
1.0
1.4 1.5 1.7
7.3 9.1 9.0
5.8 7.3 7.2
55.0 48.4 47.6
92.4 91.7 92.0
Receptor
1975
77
3
22
3.0
1.8
1.9
12.5
1.0
1.1
3.1
1.5
2.7
8.0
1.8
1.2
1.0
2.1
1.4
1.2
1.0
41.2
87.5
1980
84
11
29
4.7
2.3
2.4
17.3
1.0
2.0
1.0
1.9
7.8
1.4
1.1
1.5
2.2
1.2
1.0
1.3
1.1
35.6
86.8
233
1985
86
13
31
4.6
2.2
2.4
15.4
1.4
2.3
1.1
1.6
9.3
1.4
1.1
l.S
2.6
1.2
1.0
1.3
1.1
34.7
86.2
Baseline
Emissions
T/yr
7620
7297
5495
5896
16350
5280
6315
3446
3446
2530
2322
1205
3660
3660
50
8
Allowable
Emissions3
T/yr
6265
4130
3115
3847
9240
2974
3570
3222
3222
38
1103
382
74
74
b
b
aBased on current SIP regulations.
In compliance.
Source: Reference 44
-------
6.0 INTERPRETING CHEMICAL, ELEMENTAL, AND MORPHOLOGICAL DATA
Recent advances in techniques which measure the elemental or
chemical concentration (by total mass or particle size range) of
the aerosol and the morphology and chemical composition of indi-
vidual particles have led to concurrent advances in the analyst's
ability to characterize the ambient aerosol, identify major
contributing sources and source categories, and further define
and solve air pollution control problems.
This chapter discusses several techniques that have seen
progressively more common usage:
0 Temporal, spatial, and meteorologically-affected (TSM)
patterns
0 Enrichment factor (E)
0 Chemical element balance (CEB)
0 Interspecies correlations
0 Pattern recognition
0 Factor analysis
0 Interpretation of morphological data
6.1 TEMPORAL, SPATIAL, AND METEOROLOGICALLY-AFFECTED
PATTERNS (TSM TECHNIQUES)
6.1.1 Description of Techniques
Depending upon data availability, the TSM techniques de-
scribed in Chapters 2, 3, and 4 can be applied to elements and
chemical compounds. Normally, however, the analyst will not have
access to enough data to assess temporal patterns in a meaningful
manner. Thus, for the most part, the analysis of temporal varia-
tions in element and compound concentrations will focus upon
short-term time intervals. Pollution, dosage, gradient roses,
and upwind/downwind techniques can easily be adapted to use for
102
-------
analyzing spatial variations in element and compound concentra-
tions. Intercorrelations within and among sites can also be
adapted to such uses (see Section 6.4 for a discussion of inter-
species correlations).
Similarly, the techniques described in Chapter 4 for ana-
lyzing the effect of meteorological variables can be easily
adapted for use with chemical/element data.
6.1.2 Applicability of Techniques
In combination, these TSM techniques characterize the
aerosol to the extent that they document the degree to which that
aerosol varies in composition over time and space.
Source categories impacting upon a sampling site can be
identified to an extent with this technique. Specifically, data
on the spatial and temporal variation of elements and their
relationships to meteorological variables can be used to inves-
tigate hypotheses formed by a preceding analysis of TSP data.
Such an analysis may show, for example, that Si, Al, Mg, K, and
other elements normally associated with soil dust are covariant
over time with TSP. Similarly, a weekday versus weekend analysis
may reveal higher concentrations of Pb and Br (automotive-related)
on weekdays than weekends.
The impact of a specific source can be assessed through use
of the pollution, dosage, and gradient roses, and upwind/down-
wind techniques discussed in Chapter 3. The only difference is
that one investigates the variation in specific elements/com-
pounds rather than in TSP.
The ability of these techniques to assess the effect of
changes in emissions has not yet been successfully demonstrated.
6.1.3 Relationship to Other Techniques
As indicated in the previous subsection, there is an inte-
gral relationship between these techniques and those discussed in
Chapters 2 through 4. Trajectory analysis is especially amenable
to analysis for variations in SO and NO_ concentrations.
103
-------
These techniques can be combined with various types of
diffusion modeling and emission inventorying techniques. For
example, upwind/downwind measurements of Pb concentrations have
been used in conjunction with basic Gaussian diffusion equations
to back-calculate Pb emissions from motor vehicles. Likewise,
elemental concentrations upwind and downwind of urban areas have
been used to estimate overall emission rates for those elements.
Spatial variations in concentrations can be used with the
chemical element balance technique to assess spatial variations
57
in source contributions. Particle size distribution data can
be used with this set of techniques to substantiate or invalidate
the relative impact of fugitive dust sources. In general, this
technique is closely related to or supportive of all chemical/
elemental/morphological techniques.
6.1.4 Resource Requirements
Resources required to implement these techniques are much
greater than those for similar techniques which relate to TSP
concentrations due to the much larger data set. As a result,
resources required to implement these techniques are largely a
function of the availability of standard computer programs and
the number of sites and chemicals/elements to be investigated.
The following estimates are based upon a 10 site, 20 chemical/
element analysis involving the application of only one of the.
specific techniques (adapted to chemical/element use) described
in detail in Chapters 2 through 4:
Manpower Low
Skill Moderate
Computer Necessary
Data Chemical/element measurements,
other measurements dependent
upon specific application*3
_ . -
Does not consider the equivalent laboratory time or skills
, required to utilize the instrumentation indicated.
Specific choice of instrumentation depends upon specific ele-
ments investigated.
104
-------
6.1.5 Example Application
Due to the large number of possible techniques that have
been subsumed under this section, four different examples will be
presented. One will demonstrate the analysis of diurnal varia-
tions in various elements, another two will address the element
rose and upwind/downwind techniques, and a fourth will exemplify
the use of linear regression as a way of quantifying the rela-
tionship between chemical/element concentrations and meteoro-
logical variables.
6.1.5.1 Diurnal Variations - The example is drawn from a recent
study performed for the New York metropolitan area. Four-hour
nucleopore samples were collected six times daily at one site for
a period of 13 days (minus invalid data) during December. These
samples were then analyzed with X-ray fluorescence (XRF) for Si,
S, Ti, V, Mn, Fe, Zn, As, Br, and Pb. In addition, three nu-
cleopore samples of street dust and four of local soil were also
collected and analyzed with XRF. The authors suspected .that
there were major analytical errors for Ti, As, and Mn; thus,
those elements are not addressed herein.
Figure 6-1 presents plots of TSP and element data for the
sampling period normalized to the first 4-h period (2200-0200)
concentration. Since meteorological conditions were identical
for TSP and element measurements, any deviation from the TSP
concentration pattern could be attributed to either sampling
error or differing emission patterns for the sources of the
various elements. As Figure 6-1 indicates, Si, V, Fe, and Zn are
covariant with TSP. Pb and Br, on the other hand, are covariant
with each other but differ significantly from TSP. S shows a
unique ^pattern.
6.1.5.2 Element Rose - The element rose technique was used in a
58 ' " •
recent study for Miami, Florida. Samples were collected' at
three^different sites. The wind direction at each of these sites
was structured into 36 separate 10 degree groups, and average
concentrations for each of 15 elements were calculated.
105
-------
2.0
1.5
to
C
o
•H
4J
id
M
-P
C
0)
O
c
o
o
TJ
0)
N
1.0
LEGEND
Si
TSP-
S ---
Fe- .........
Br
Pb 9 9 9
1
5
1400-1800
6
1800-22C
2200-0200 0200-0600 0600-1000 1000-1400
; Sampling period
Figure 6-1. Normalized diurnal variation in TSP and selected
elements.
Source: Reference 17
106
-------
According to the authors of the study, the most striking
example of the usefulness of the method was provided by the
distribution of Pb and Br at a background site. The peak in each
case was toward the southwest, pointing toward a large Dade
County-operated vehicle service station.
6.1.5.3 Upwind/Downwind Analysis - The example for this tech-
59
nique is drawn from Phoenix, Arizona. Element data were col-
lected for five 6-h sampling periods at sites upwind and downwind
of the Phoenix urbanized area. Four-stage Andersen size frac-
tionating sampler and dichotomous sampler filters were subjected
to elemental analysis by X-ray fluorescence.
Downwind samples of two of the six elements investigated
(Br, Pb) were generally about twice as great as upwind ones.
Elevated values of those elements were attributed to inversion
conditions. Cu, Zn, and As concentrations were slightly higher
downwind than upwind. It was suspected that mixing of air masses
introduced Cu emissions from nearby smelters into the Phoenix
area. Fe concentrations were similar at both types of sites.
6.1.5.4 Linear Regression for Meteorological Variables - Drawn
from a study of Tucson, Arizona, this example demonstrates the
use of linear regression analysis to assess the impact of meteo-
rological variables upon chemical/element concentrations.
Initial variables included the following:
0 Relative humidity (day before sampling)
0 Dew point (day before sampling)
0 Wind speed (day before sampling)
0 Most persistent and resultant wind direction
0 Hours of sunshine and cloud cover
0 Precipitation (occurrence and amount)
0 Maximum, minimum, and average visibility
Samples were divided into two categories according to the
two dominant resultant directions. When their chemical/element
concentrations were compared, no significant differences were
107
-------
found. The authors concluded that in order to investigate the
effect of wind direction in more detail, shorter sampling periods
would be necessary.
Selected meteorological parameters were correlated with
chemical/element data at two sites, one urban and one rural.
Results of the analysis for the urban site, shown in Table 6-1,
indicated to the authors that TSP and Al (and the other soil-
related elements) showed a direct dependence upon wind speed.
TSP and Al were inversely related with humidity and dew point.
No apparent relationship between precipitation and chemicals/
elements was revealed. TSP and Al were inversely related with
visibility. No significant relationships were demonstrated for
non-soil species like SO4 and Pb. Results of the rural site
analysis, shown in Table 6-2, were similar to those of the urban
site with two exceptions: at the rural site, the inverse rela-
tionship between Al and humidity and dew point was much stronger,
and the dependence of TSP and Al with wind speed was less cer-
tain. The authors attributed these results to the greater effect
of soil moisture upon soil dust than fugitive dust, and to the
possible unrepresentative nature of the wind data used for the
rural sampling site.
6.1.6 References
17, 22, 30, 34, 55, 56, 57, 58, 59
6.2 ENRICHMENT FACTOR (E)
6.2.1 Description of Technique
The enrichment factor method uses data concerning the
elemental concentrations of suspended particulates collected fron
the atmosphere and of the continental crust or sea to produce
estimates as to the degree to which a given element is "enriched"
*
hence produced by non-natural sources.
* An excellent discussion of the enrichment factor technique can
be found in Reference 60. Much of the material herein is based
upon that reference.
108
-------
Table 6-1. DEPENDENCE OF CHEMICAL COMPOSITION ON
METEOROLOGICAL PARAMETERS (URBAN LOCATION)
ws
RH
DP
T
P
VIS
WS(DB)
RH(DB)
DP(DB)
MASS
.28(.02)*
-.48(.00)
-.46(.00)
.00(.99)
-.41(.00)
-.34(.01)
.42(.00)
-.48(.00)
-.52(.00)
Al
.18(.14)
-.55(.00)
-.39(.00
.18(.i4)
-.54(.00)
-.26(.03)
.36(.01)
-.57(.00)
-.39(.001)
*Y
-.1K.40)
.03(.84)
-.17(.17)
-.29(.02)
.3K.01)
-.2K.09)
-.14(.255)
.17(.18)
-.10(.42)
N03"
-.12(.34)
.01(.94)
-.13(.30)
-.12(.34)
.12(.34)
-.28(.02)
.02(.87)
.06 (.66)
-.05(.69)
WS
RH
DP
T
P
VIS
WS(DB)
RH(DB)
DP(DB)
Pb
-.27(.02)
.07(.58)
-.3K.OO)
-.42 (.00)
.39(.00)
-.01(.94)
-.21(.IO)
.15(.23)
-.25(.04)
Zn
-.15(.24)
.07(.58)
-.05(.72)
-.20(.12)
.30 (.01 5)
-.10(.45)
-.21(.09)
.06(.66>
-.03(.80)
Cu
-.15(.22)
.10(.42)
.08C.52)
-.06 (.66)
.26(.05)
-,01(.94)
-.21(.10)
.13(.29)
.09(.50)
Cd
-.15(.24)
-.03(.82)
-.02(.88)
.07(.60)
.13(.29)
,00(.99)
-.16(.20)
-.04(.73)
.01(.91)
Vali£s in parentheses indicate the probability of true
cowelation coefficient being zero (i.e., no correlation exists)
Wind Speed
Relative Humidity
Dew Point
Temperature
Sourc^: Reference 34
P: Barometric Pressure
VIS: Visibility
WS(DB): Wind Speed (day before sampling)
RH(DB): Relative Humidity (day before sampling)
DP(DB): Dew Point (day before sampling)
Reprinted by permission of the publisher.
109
-------
Table 6-2. DEPENDENCE OF CHEMICAL COMPOSITION ON
METEOROLOGICAL PARAMETERS (RURAL LOCATION)
ws
RH
DP
VIS
WS(DB)
RH(DB)
DP(DB)
MASS
.18(.18)
-.52(.00)
-.50(.00)
-.14(.31)
,20(.13)
-.40(.00)
-.46(.00)
Al
.20(.13)
-.70(.00)
-.56(.00)
-.14(.30)
.28(.04)
-.65(.00)
-.54(.00)
so4-
-.09(.51)
.20(.14)
-.08(.57)
-.16(.24)
-.12(.36)
.32 (.02)
-.02(.86)
NO.
.10(.46)
-.06(.64)
-.14(.32)
-.19(.17)
.17(.21)
-.17(.22)
-.14(.30)
WS
RH
DP
VIS
WS(DB)
RH(DB)
DP(DB)
Pb
-.13(.35)
.12(.37)
-.22(.ll)
-.06(.65)
.03(.81)
.25(.06)
-.19(.17)
Zn
-.06(.67)
-.12(.39)
-.08(.55)
.00(.99)
-.03(.81)
-.1K.42)
-.10(.45)
Cu
-.09(.51)
.28(.04)
.09(.51)
.14(.30)
-.07(.60)
.32(.02)
.09(.53)
Cd
-.12(.38)
.22(.10)
.08(.55)
.05(.69)
.00(.97)
.26(.05)
.07(.60)
Values in parentheses indicate the probability of true
correlation coefficient being zero (i.e., no correlation exists).
WS: Wind Speed
RH: Relative Humidity
DP: Dew Point
VIS: Visibility
WS(DB): Wind Speed (day before sampling)
RH(DB): Relative Humidity (day before
sampling)
DP(DB): Dew Point (day before sampling)
Source: Reference 34
Reprinted by permission of the publisher.
110
-------
The enrichment factor itself is calculated for the various
elements in the aerosol relative to the crust or sea or both,
usually normalized to an element considered to be the clearest
indicator of the naturally occurring source material. The gen-
eral formula for such an enrichment factor is:
E(x) = (x/Ref) aerosol
1 aerosol-source (X/Ref) source (crust or sea)
where E(X) , is the enrichment factor of the element
aerosol-source
X in the aerosol, relative to some source material and a refer-
ence element; X/Ref is the ratio of the concentrations of element
X and the reference element in the aerosol and the source mate-
rial.
Elements most frequently used as crustal indicators include
Si, Al, Fe, and Sc, whereas for sea salt Na is almost always
used. An enrichment factor of close to 1.0 indicates that the
crust/sea is the primary source of the measured aerosol. A
factor significantly greater than 1.0 can normally be interpreted
as the result of man-generated emissions.
A number of concepts inherent to the enrichment factor
technique are still under active investigation, among which are
the following. First, the element fractionation of the crust/sea
material as it interfaces with the atmosphere is largely unex-
plored. What little data are available relate primarily to sea
salt fractionation. Second, although average elemental concen-
trations of crustal rock are normally used for calculation pur-
poses, the true contribution of this natural source is probably
much more closely related to the elemental concentrations of soil
in the region of aerosol sampling. Few data are available con-
cerning the regional variation in elemental concentrations of
soil. Third, the most common crustal reference element, Al, is
used frequently in the vicinity of samplers and risks being
subjected to biased calculations. Si, a better reference mate-
rial, has only recently become amenable to reasonably available
measurement techniques.
Ill
-------
Some limitations to the technique have been emphasized in
the literature: first, man-induced fugitive dust (from traffic
on unpaved roads, for example) will cause significant amounts of
dust, but will not necessarily produce high enrichment factors
due to the natural (crustal) origin of the material disturbed by
the activity; second, moderate or high enrichment factors in
relatively remote areas may be the result of natural soils which
differ from crustal rock, or the action of such other material
sources as volcanoes, forest fires, or vegetation.
6.2.2 Applicability of Technique
The enrichment factor technique characterizes the aerosol i
terms of its elemental concentration relative to the reference
material. This is perhaps the second best attribute of the
technique. The best attribute is the fact that this technique
will provide evidence as to the degree to which anthropogenic
sources contribute to ambient concentrations measured at ruralo
remote sites. In other words, it will provide evidence concern-
ing the nature and impact of long-range transport of man-related
source emissions.
The enrichment factor method has not been shown to be
useful in assessing the impact of a single source or in assessin
the effect of changes in emissions from that source. Nor has it
been shown to be useful in identifying major source categories
impacting upon a sampler in an urban area.
6.2.3 Relationship to Other Techniques
No relationships between the enrichment factor technique an
those which analyze temporal or spatial patterns of air quality
have been reported in the literature, with one exception. The
variation between rural and urban sites in the vicinity of a
34
large city can be analyzed in terms of enrichment factors.
Developments in this area are possible.
The only relationships between this technique and those
which assess the effect of meteorological variables derives from
112
-------
the fact that enrichment factors calculated at rural sites are a
way of deducing the impact of long-range transport.
No relationships with previously discussed emission inven-
torying and diffusion modeling techniques have been reported.
However, enrichment factors can be used in conjunction with
element intercorrelations to provide evidence of soil dust impact.
In addition, enrichment factors can be stratified by particle
size ranges to investigate further the impact of windblown and
fugitive dust. Again, further developments in this area are
possible.
6.2.4 Resource Requirements
Resource requirements for the enrichment factor technique
are summarized below:
Manpower Low
Skill Moderate
Computer Not necessary
Data Chemical/element measure-
ments for aerosol and
crust or seab
Does not consider the laboratory skills required to utilize the
, instrumentation indicated.
Specific choice of instrumentation depends upon specific ele-
ments investigated.
The dependency of this technique upon sophisticated ana-
lytical instrumentation to produce element data is a fact which
makes this technique simultaneously possible yet costly to apply
on a routine basis.
6.2.5 Example Application
An example of this technique is drawn from a recent study of
34
Tucson, Arizona. Al was used as the reference element and soil
was used as the reference material. A comparison was made be-
tween average enrichment factors (E's) for urban alia background
particulates. Elements which the authors of the st&dy considered
113
-------
to be attributable to the airborne soil-derived material (Al, Mg
Sr, Fe, Na, Si, Mn, Ti, Rb, Ca, K, and Li) showed E's very close
to unity. Other species (Zn, Cu, Cd, Pb, In, Tb, Bi, N0_, SO,
3 4
and NH.) were observed to have E's of 50 to 2000 regardless
of their urban/rural location. Non-soil sources for these
species were deduced. Four other elements (Ce, Co, Ni, and Cs)
showed E's of close to 1.0, but their poor correlation with the
soil-derived elements led the authors to conclude that either the
local crustal composition of these elements was highly variable
or that non-soil sources were responsible for introducing a
significant fraction of those species into the atmosphere.
The fact that E's were higher for most non-soil species at
the rural location than they were at the urban location (see
Figure 6-2) suggested to the authors that nonurban sources for
those species contributed in large measure to the total atmo-
spheric burden measured in the urban area (as well as the rural
area).
Enrichment factors were also calculated for large (>2 urn)
and small (<2 urn) particles. In comparison with data for the
entire size range, non-soil species E's were found to be lower
for large particles and higher for small particles (see Figures
6-3 and 6-4), thereby adding further credence to the proposition
that windblown and fugitive dust were dominant source categories
6.2.6 References
26, 34, 59, 60, 61, 62, 63, 64 [103, 104]
6.3 CHEMICAL ELEMENT BALANCE (CEB)
6.3.1 Description of Technique
The purpose of the CEB technique is to permit the analyst t
use available data concerning typical elemental emission rates
from selected source categories and elemental concentrations of
the aerosol at a given sampling site to back-calculate the frac-
tions of the overall aerosol contributed by all major source
114
-------
10?-
• Tucson
A Background
• •
Mg Sr Fe Na Mn Cr Si K Co Ti Ni Rb Ca Li Cs Zn Cu Sq= Tl
Bi
Pb
Figure 6-2. Enrichment factors for species in desert background
and urban particulate matter.
Source: Reference 34
Reprinted by permission of the publisher.
115
-------
10 ^
10'-.
10
10'
10-
Small particles =
A -
A * *
l,!^i'_ A A • * ^
IX/AII A :
crxnf A
•
~
•
w1-
A E :
A A
• • A :
i ^ • urban
A A 1 1 • . • A * ArurG! ;
A A
. _ ,„«
• * • 10 :
:
Large particles
& •
A
r IX,'AI)oir A .
lX/A1!tru,,
A
;•
. A • Ourbon
e A * A I
*AA$ *A*A
• • * •
A
F. MS Ti Mn No K Sr Ni C« Cr Co U Kb Zn SQj Cu Cd NOj NK«
F. Mg Ti Mn Ma K Sr Ni Co Cr Co li lib Zn SQ^ Cu Cd NO^ NK4
Figure 6-3. Enrichment factors for species in
desert background and urban small particle
(<2 urn) particulate matter.
Source: Reference 34
Figure 6-4. Enrichment factors for species
in desert background and urban large
particle (>2 urn) particulate matter.
Reprinted. t>y pel-mission of tin
-------
categories. Certain preselected chemical elements are used as
tracers for primary particulate emissions from the selected source
categories. Secondary contributions, if deemed to be significant
enough to warrant inclusion, can be estimated from measured
concentrations of SO,, NO.,, and organic vapors.
The fraction (P.) of element i in an aerosol sample is given
by the following simple mass balance equation:
W
P. = Z a. . x. . S . (eq.3)
j = l vj ID D
where S. = fraction of the aerosol contributed by source
^ category j
a.. = fraction of source category j's emissions con-
^ tributed by element i
x. . = coefficient of fractionation of the element i
1-' between the source category j and the atmosphere
W = number of source categories emitting element i
If n elements are considered, then n such equations must
be satisfied.
From the definition of S., the continuity relation must
hold:
IS. =1 (eq.4)
j J
Values for P. are obtained from the elemental analysis of
atmospheric samples. Values for a.. are extracted from published
references or, where practicable, from recent source sampling.
Values for x.. representing the fraction of the emitted species
^O '
i which appears at a sampling site, are usually assumed to be
equal to 1.0. There is, however, some evidence of ion frac-
tionation in sea spray, elemental fractionation in blowing dust,
and fractionation resulting from diffusion and sedimentation if
particles of different sizes have different chemical composi-
tions. ' The fraction of the aerosol due to source category j
(S.) is initially unknown.
117
-------
Where an element comes entirely from one source category,
that element can be used as a tracer. In such cases, the mass
balance equation is very simple. Values for S. might be found by
solving the following equation:
Sj =
J
13
For example, Pb is predominantly emitted by the combustion of
leaded gasoline. The generally accepted value for the fraction
of lead in auto exhaust is 0.4. If the measured Pb fraction of
the aerosol is .025, then the mass balance for Pb is simply:
c _ -025
S
auto 0.4 (1.0)
This yields an: S . = .0625 or 6.25 percent. The resulting
SUtO
S.'s for the tracer elements are then used to generate predicted
ambient concentrations for non-tracer elements. The overall
accuracy of the chemical element balance can then be assessed by
comparing measured versus predicted ambient concentrations.
If there were as many tracer elements as source categories
(as in the above example), then the analyst could calculate a
unique solution for the source contributions. This is rarely the
case, however. If there are more tracer elements than source
categories, then one could find a linear least squares solution
to the mass balance equations of the tracers. Where there are
fewer tracers than sources, then linear programming solutions
\
56,6!
54
must be found. The least squares approach, which has been use<
most frequently in the literature, will be described herein.
The principle behind the least squares method is finding th<
combination of S.'s that minimizes the sum of the squares of the
deviations between measured and predicted elemental concentra-
tions. The quantity to be minimized, F, is given by:
118
-------
F = I —i * 2iJ iJ J (eq.6)
ai
where o. = experimental uncertainty in P.
Setting the partial derivatives of F with respect to the
S.'s to zero results in a system of equations of the form:
I S.. E aik Xik aij xij = I. Pi aik xik (eq>?)
ai ai
The subscript k denotes the kth equation. This yields one
equation for each source rather than one for each element. The
68 69
system of equations is solved by standard matrix methods. '
Source categories typically investigated with this tech-
nique,, tracers used for those sources, and the cities in which
those sources/tracers have been used are given below:
Source category Tracer
Soil dusta'b;c'd'e'f AI, (Si)e
Sea Salt 'D/r abcdef Na c
Automobile emissions v, A d e f pk' (Br)
Fuel oil combustion,' 'f' ' ' V
Portland cement0'c'Q'e/ c Ca
Coal burning and coke produc^ign ' AI (Fe,Ti)
Iron and steel manufacturing ' (Mn, Fe)c
a Pasadena, CA (1971)
*, Pasadena, CA (1973)
^ Chicago, IL (1974)
° St. Louis, MO (1974)
f St. Louis, MO (1975)
Miami, FL (1975)
°It will be noted that the list of source categories shown
above is not all inclusive. Other source categories, such as
incinerators and diesel powered motor vehicles, are not included,
The contribution, of such source categories is estimated by using
119
-------
a scaling factor derived from routinely available emission inven-
tories. This scaling factor is the ratio of the emissions of the
source category not included in the chemical element balance to
the emissions of a source category which is included in the
balance.
Most of the referenced applications of the CEB technique
terminate at this point. In other words, they discount the
effect of secondary particulates. In some cases, notably those
of Friedlander, the role of secondary particulates is explicity
accounted for. The reader is referred to Reference 65 for a
detailed explanation of this development in the CEB technique.
The technique clearly has some drawbacks that should be
noted by the potential user:
0 A detailed knowledge of the elemental composition of
each source—not routinely available—is necessary. As
a result, a large number of assumptions (e.g.,.a..
is the same for all sources in a given category)xand
approximations are required.
0 The technique has no predictive modeling capability.
0 Source categories, soil dust for example, are indis-
tinct. Within this category are such important sub-
categories as naturally occurring windblown dust,
anthropogenic fugitive dust sources, and wind erosion
fugitive dust sources.
6.3.2 Applicability of Technique
The chemical element balance technique uses elemental data
that partially characterize the aerosol in terms of its composi-
tion, but the primary utility of the technique lies in its unique
ability to identify the relative contribution of various source
categories to measured TSP concentrations. No other technique
has demonstrated an equal ability to quantify the impact of soil
dust or sea salt. This capability is limited to the extent that
suitable tracers have not yet been identified for many industrial
processes and that the technique has not yet shown an ability to
distinguish among sources within a given source category.
120
-------
Even though the CEB technique can identify major impacting
source categories, it has not yet been shown to be capable of
defining the impact of a specific source. For the same reason/
it is not yet possible to use this technique to determine the
effect of changes in emissions or the location of specific sources
6.3.3 Relationship to Other Techniques
The chemical element balance technique has not yet been used
in conjunction with techniques that assess temporal variations in
air quality- One reason is that the technique depends upon
costly analytical techniques such as neutron activation to pro-
duce the ambient elemental concentrations (P.'s). It is possible
that temporal variations could be assessed by using aggregate
sets of elemental data, but this possibility is dependent upon
the emergence of less expensive, real time, analytical instru-
mentation.
It is possible to use the technique to assess the spatial
variation in TSP values by performing CEB for different sites for
the same time period. By doing so, one can identify the relative
contributions of source categories at various sites.
No efforts to relate this technique to those which assess
the effect of meteorological variables have been reported. The
reasons are identical to those expressed for temporal variations.
Developments in this area are possible.
An emission inventory is a necessary ingredient of the CEB
technique at the present time because, in the absence of suitable
tracers for all source categories, some means of accounting for
"untraced" source categories must be used. That means has been
the emission inventory.
Diffusion modeling and CEB technique results can be used to
cross-check each other. In the case of diffusion modeling the
analyst begins with "known" emission rates and predicts TSP
concentrations. In the case of the CEB technique, he begins with
known ambient .concentrations and back-calculates diffused source
contributions. Further, in one recent study the Hanna-Gifford
121
-------
diffusion model was combined with the CEB and factor analysis
54
techniques to produce a new factor model of air pollution. The
pattern recognition, interspecies correlation, and microscopy
techniques could be used to support the findings of the CEB
technique.
6.3.4 Resource Requirements
In the following estimate of requirements, it is assumed
that the technique will be applied in a fashion similar to that
reported in the literature (i.e., applied to small sets of short-
term data):
Manpower Low
Skill High
Computer Necessary
Data Chemical/element measurements
for aerosol and major source
categories, coefficients of
fractionation for sources,
emission inventory, tracer,
elements for major sources
Does not consider the laboratory skills required to use the
, instruments indicated.
Specific choice of instrumentation depends upon specific ele-
ments investigated.
6.3.5 Example Application
An example of how the CEB technique is applied can be drawn
from a recent study in St. Louis. Major anthropogenic sources
of primary particulates in that city were drawn from a 1975
emission inventory (see Table 6-3).
122
-------
Table 6-3. SOURCES OF PRIMARY PARTICULATES IN ST. LOUIS
Emissions,
Source ton/yr
Coal combustion 19,817
Fuel oil combustion 862
Waste incineration 1,018
Automobile exhaust 1,276
Diesel exhaust 287
Industrial processes 1,328
Tire dusta 128
Soil dust
Cement dust
Assumed equal to one-tenth of automobile exhaust.
Source: Reference 67
Data on the chemical composition of soil dust, automobile exhaust,
fuel oil fly ash, and cement dust were extracted from previous
reports which applied the CEB method (see Table 6-4). Similar
data for emissions from coal-fired power plants were derived by
averaging the results of four previous studies. The resulting
averages are also presented in Table 6-4. Two 24-h periods were
studied: July 23 and 25, 1975. Hi vol samples for those two
periods were subjected to organic analyses for benzene, methanol,
and water extracts. Two-hour nucleopore filter samples collected
over the same time intervals were analyzed for their elemental
concentrations using X-ray fluorescence. Analysis for nitrates
and ammonium were not made. [It should be noted that other
applications have included these variables.] Measured concen-
trations for these variables are given in Table 6-5.
Al and Si were used as tracers for soil dust; Ca, Pb, and V
for cement dust, automobile exhaust, and fuel oil, respectively;
and Fe and Ti were used for coal combustion. Resulting balances
for elemental concentrations are also given in Table 6-5.
Calculated source contributions are given in Figure 6-5.
Industrial, diesel, and tire dust contributions were calculated
123
-------
Table 6-4.
SOURCE CONCENTRATIONS OF PARTICULATE MATTER
(percentages)
Al
Ca
Fe
K
Si
Pb
Ti
V
Cd
Co
Cr
Cu
Mg
Mn
Na
Ni
Zn
Br
Soil dust
5
0.8
3
2.
20b
0.005
0.3
0.007
-
0.002
0.005
0.003
0.7
0.03
0.6
0.005
0.01
-
Auto exhaust
U
U
0.4
U
U
40.0
U
U
U
U
U
U
U
U
U
U
0.14
7.9
Fuel oil
fly ashc
5.0
0.4
2.5
0.10
1*
0.18
0.03
2.5
—
0.15
0.12
0.16
0.3
0.03
1.5
6.0
0.05
-
Cement dust
2.4
46.0
1.09
0.53
10.7
—
0.14
-
—
—
-
-
0.48
—
0.4
-
-
-
Coal combustion
11.7
3.8
10.6
3.3
24
0.09
1.0
0.08
0.004
0.008
0.065
0.031
0.9
0.034
0.037
0.52
-
f Gatz (1975)
" Miller, et al. (1972)
JJ Winchester and Nifong (1971)
Average of results in a table provided in Reference 64
U = Unknown
Source: Reference 67
-------
Table 6-5. RESULTS OF CHEMICAL ELEMENT BALANCE FOR ELEMENTAL CONCENTRATIONS
(ug/m3)
July 23, 1975
Predicted Measured
a
Caa
31
Si?
a
Tia
V
Cl
K
Cr
_ Mn
K Ni
01 Cu
Zn
Br
Benzene
Organics
Methanol
Organics
Water
Organics
1.5
3.0
1.26
3.5
0.92
0.121
0.017
0.156
0.42
0.0098
0.0053
0.0068
0.0036
0.066
0.18
n.d.
n.d.
n.d.
2.5
3.0
1.23
3.3
0.93
0.132
0.017
0.108
0.33
0.021
0.048
0.0068
0.36
0.54
0.24
4.2
11.8
14.2
± 1.2
0.2
0.05
1.0
0.05
0.009
0.002
0.009
0.02
0.002
0X003
0.0007
0:03^
0.04
± 0.02
July 25, 1975
Predicted Measured
1.0
3.5
0.70
3.5
0.07
0.070
0.0082
0.12
0.33
0.0033
0.0043
0.0032
0.0014
0.021
0.14
n.d.
n.d.
n.d.
1.2
3.4
0.68
2.9
0.70
0.082
0.0083
0.19
0.41
0.023
0.043
0.0019
0.023
0.24
0.103
4.8
13.6
17.8
± 0.5
0.3
0.04
1.2
0.05
0.007
0.0023
0.02
0.03
0.003
0.003
0.0011
0.002
0.01
± 0.006
a Used in chemical element balance.
n.d. = no data
Source: Reference 67
-------
July 23, 1975
25, 1975
Calculated TSP 51 ±8 ug/m3 a'b'c
Measured TSP 109
54 ±8 ug/m3 a'b'c
41
Waste incineration, diesel exhaust, industrial processes
, scaled to automobile exhaust by using emission inventory
Tire dust assumed to equal one-tenth of automobile exhaust
Ammonium assumed to balance sulfate as (NH4)2SO4
Figure 6-5. Source contributions to St. Louis aerosol
Source: Reference 67
126
-------
by scaling the automobile exhaust contribution according to the
Table 6-3 emission inventory. Secondary sulfate contributions
were calculated by assuming that all sulfur was secondary sul-
fate. All water-extractable organics were assumed to be second-
ary.
6.3.6 References
34, 54, 56, 57, 58, 65, 66, 67, 68, 69, 70, 71, 72 [105]
6.4 INTERSPECIES CORRELATIONS
6.4.1 Description of Technique
An important prerequisite for applying this technique is a
knowledge of the concentrations of each trace constituent present
in a set of particulate samples. Once the trace constituents or
species present in a set of particulate samples are quantified,
it is possible to evaluate the relationships between species
using a correlation analysis. Correlation coefficients are
calculated for each possible species pair measured in the set of
data. Examples of this technique in the literature include a
tabulation of the coefficients in a matrix.22'34'54'73'74'75
There are several types of correlations that can be used.
One of these involves the correlation between species concentra-
tions as measured at two different sites. For example, a corre-
lation matrix can be generated for Al or TSP as a function of
location. Another type of correlation involves the correlation
between species measured at only one location. For instance, the
correlation between TSP and Pb or between Pb and Br can be exam-
ined at one sampling site.
An important limitation of interspecies correlations is that
the correlation coefficients indicate how well two variables vary
together, not necessarily that there is or is not a causal rela-
tionship between the two variables.
127
-------
6.4.2 Applicability of Technique
A matrix of correlation coefficients can be developed for
species quantified at one site or several sites. By doing so,
the aerosol is quantified in terms of its composition and spatial
distribution. A good correlation between TSP and another, species
implies that a large part of the variations in measured TSP can
be explained by the variation in the emissions from the source of
that species. For example, a high correlation between TSP and Al
at one site would indicate that the variation in TSP could be
explained by the variability of Al.
When this technique yields good correlations between a
species measured at different locations, the results indicate a
common source or type of source for the species. For instance, a
high correlation between Pb at two different locations would
indicate that the Pb measured at the sites has a common source.
The resulting correlations between species associated with
a specific type of source serve to identify general impacting
source categories. However, no quantitative assessment of source
impact is possible. For example, a good correlation between Pb
and Br at a particular site would suggest an impact from vehicle
exhaust, but it would not indicate how much of an impact (in
terms of TSP concentrations) that source category had.
No determination of the impact of a single source or the
impact of changes in emissions has been demonstrated with this
technique.
One aspect of these correlations should be noted. A low
correlation between species suggests that the species may not be
related to each other. However, a low correlation does not
preclude the possibility that there is a relationship between
species. It is possible that the true relationship is hidden by
sampling errors, the range of variation among the variables, or
other interferences.
128
-------
6.4.3 Relationship to Other Techniques
With a sufficient number of samples, temporal patterns in
interspecies correlations can be investigated for a single site
or several sites. If the correlation coefficient changes signifi-
cantly, it can be inferred that the source category associated
with the species varies over time or that the meteorological
conditions affecting the measured concentrations changes with
time.
Detailed elemental analyses of source emissions can be used
with interspecies correlations to tentatively identify major
impacting sources.
TSP intersite correlation data can also be compared to the
interspecies correlation data. Where good correlations are found
between sites for TSP and different species, it can be inferred
that the sites are measuring essentially the same air mass. For
example, if a good correlation between TSP and Al are found to
exist for two or more sites, the implication is that the same air
mass is being measured by both sites. Interspecies correlations
can also be compared when using the pollution rose and upwind/
downwind techniques.
Assuming a sufficient data set is available, interspecies
correlations can be determined based on a stratification of
meteorological parameters. The nature of the dominant sources
can be inferred from observed changes in the correlations. For
instance, if TSP data are separated into two categories of pre-
cipitation—days with rainfall greater than 0.1 inch and days
without rain—a comparison of the correlations between TSP and Al
could be made. The results of the analysis might indicate whether
the Al is of soil origin or is being emitted from a stationary
source. In addition, correlations among N0_, SO., O_ and visi-
bility measurements can be used with trajectory analysis to
assess the impact of long-range transport.
Interspecies correlations are direct inputs to the pattern
recognition and factor analysis techniques, and can be used with
the techniques described in Section 6.1.
129
-------
6.4.4 Resource Requirements
As the number of samples increases/ the resource require-
ments change. A small number of samples can be evaluated rather
easily. However, given a sufficiently large set of data, the
resource requirements are:
Manpower Lowa
Skill Moderate
Computer Necessary
Data Chemical/element measure-
ments for more than one
site (number of values
must be adequate for .
statistical significance)
Does not consider the laboratory skills required to use the
. instruments indicated.
Specific choice of instrumentation depends upon specific ele-
ments investigated.
6.4.5 Example Application
A recent study of urban air quality in Tucson, Arizona
34
serves as an example of this technique. Table 6-6 presents the
linear correlation coefficients calculated for the data collected
at an urban location. The matrix includes the coefficients
calculated for all possible species pairs. Based on the amount
of data used, coefficients of 0.3 and greater would indicate a
statistically significant relationship. The principal correla-
tions, observed by the authors, from the matrix are:
0 The elements Al, Fe, Si, Ti, Li, Tb, K, Ca, Mg, Na,
Mn, Sr, and Cr are all correlated with each other
0 TSP concentrations are significantly correlated only
with the above elements
0 The species pair SO4~NH4 is very strongly correlated to
each other
0 Zn and Cu appear to have some correlation with SO. and
NH4
130
-------
Table 6-6. LINEAR CORRELATION COEFFICIENTS FOR
URBAN SAMPLES, TUCSON, ARIZONA
At S04 NH4 Pb Hiss ft H03 Si Tl Cs LI Bb In t U Ng N* Cu Nn Sr Hi Co Cr Cd Bl Tl In
A! —
S04 .08 ™
NH4 .10 .81 —
fb .33 .33 .32 ---
Hast .72 .19 .2i .43 —
F« .75 .15 .14 .34 .60 —
NO, .31 .24 .30 .62 .35 .Z3 ---
Si .89-.01 .08 .24 .70 .72 .27 —
Ti .84 .04 .IS .19 .70 .83 .18 .85 —
Cl .44-11 -.03 .30.47 .25 .39 .42 .*A —
Li .90 .03 .10 .32 .63 .70 .36 .80 .75 .44 —
Do .74 .01 .11 .35 .64 .52 .39 .78 .68 .54 .76 —
IM ,01.54 .50 .25 ,fl8 .42 .09 .0 -.OB-.10 .12,02 ---
K .72 .19 .17 .43 .67 .98 .27 .67 .58 .25 .68 .50 .45 —
C* .78 .25 .23 .23 .66 .91 .25 .70 .66 .22 .73 .45 .33 .89 —
Hg .79 .22 .IS .25 .61 .» .25 .70 .60 .24 .81 .53 .35 .87 .93 —
Ml .69 .15 .08 .45 .58 .85 .34 .64 .52 .27 .66 .47 .37 .88 .79 .83 —
Cu -.05 .55 .41 .23-.10 -.05 .14 -.12 '.20 .0 .07-.01 .36 -.03 .0 .05 .0 —
Nn .92 .03 .08 .76 .63 .68 .37 .80 .76 .49 .95 .70 .06 .68 .75 .79 .88 .04 —
Sr .90 .13 .15 ,35 .67 .68 .35 .82 .77 .42 .94 .80 .09 .65 .71 .77 .67 .05 .90 —
Hi .30 .19 .25 .23 .35 .34 .43 .27 .15 .21 .42 .33 .30 .36 .34 .47 .41 .23 .40 .40 —
Co .25 .11 .05 .17 .23 .20 .31 .19 .20 .38 .35 .27 .15 .19 .15 .20 .25 .15 .37 .32 .24 —
Cr .50 .0 .01 .05 .25 .45 .19 .54 .40 .05 .61 .41 .25 .45 .45 .54 .49 .06 .55 .57 .57 .18 —
Cd .15 .07 .12 .15 .04 .12 .31 .15 .11 .30 .25 .23 .36 .14 .13 .16 .IS .13 .27 .20 .30 .29 .02 —
81 .08 .07 .02 -.18-. 10 -.07 -.12 .05 .05 .09 .18-.01 -.03 -»11 .0 .10 -.07 .52 .20 .09 .15 .27 .15 .05 —
Tl .0 .08 .12 -.02-.05 .03 .07 -.03 .04.10 .18-.06 .03 .03.11 .13 .09 .20.14 .14.14 .27 .18-.12.29 —
In .20-.13 -.20 -.Ij-Oj .0 -,lb .23 .05.0 .25.15 .0 -.04-.02.05 -.05 .13.23 .ZO-.06 .09 .54-.02 .35 -03 -
Source: Reference 34
131
-------
0 The nitrate ion is only weakly correlated with lead
0 The elements Cd, Bi, In, and Tl show no apparent
relationship to any measured chemical component
In the Tucson study, the elements Al, Fe, Si, Tn, Na, Mg,
Ca, Rb, Sr, K, Mn, and Si were all related to each other at all
of the locations sampled. This relationship suggested to the
authors that a possible common source for these species was the
airborne soil material generated by erosion processes.
6.4.6 References
22, 34, 54, 73, 74, 75
6.5 PATTERN RECOGNITION
6.5.1 Description of Technique
Pattern recognition is a statistical technique used for
evaluation of complex data sets. Pattern recognition can involvi
either supervised or unsupervised learning. In supervised learn
ing, a data set is characterized based on previously established
patterns. Unsupervised learning, referred to as cluster anal-
ysis, requires no preconceived patterns. Discrete elements of a
data set are grouped depending on the degree of similarity be-
tween elements. In this section, only unsupervised learning wil
be discussed. As the discussion is only a cursory examination o
cluster analysis, the reader interested in a more complete descr
tion of the technique is directed to a standard text on the
v.- ,.76
subject.
Cluster analysis lends itself to the study of air pollution
in cases where patterns are only vaguely defined. To perform
clustering, some measurement of the similarity between elements
of a data set is needed. As applied to TSP samples, this measur
is usually Euclidean distance. The measure of distance between
species is based on the correlation coefficient (r), where 1-r
represents the distance between two species.
132
-------
The first step in a clustering algorithm is to join the two
most similar clusters. Then, the distance between the new clus-
ter and each remaining cluster is computed and the process is
repeated. Reiterations of this process result in all of the
elements forming a single cluster. Results of this hierarchical
clustering are usually displayed in a dendrogram, (i.e., a dia-
34 54 77 78
gram of the clusters present at each stage of clustering) ' ' '
An alternative way to present the clustering results is to gen-
erate a nonlinear mapping of the clu
a dendrogram is given in Figure 6-6.
erate a nonlinear mapping of the clusters. ' ' An example of
6.5.2 Application of Technique
This technique can characterize the aerosol by providing a
measure of the way in which TSP concentrations and the chemical/
elemental constituents thereof vary over space.
The cluster analysis technique can be used to identify
general source categories present in the aerosol. In order to
apply the technique in this fashion, a knowledge of the concen-
trations of the species present in the aerosol is needed. By
analyzing the clusters formed from different species, general
emission categories can be identified. For example, if Al and Si
cluster together, one probable source for these elements is
windblown soil particles.
This technique does not lead to the quantification of the
impact of a specific source of emissions. The technique has no
predictive capabilities and it cannot be used to trace changes in
emissions over time.
6.5.3 Relationship to Other Techniques
The most useful relationship of this technique to other
techniques is in the interpretation of the results of correlation
analyses. The pattern recognition technique, cluster analysis,
condenses the information obtained from previously discussed
correlation techniques and then presents the information graph-
ically. As such, this technique allows general relationships to
133
-------
HIERARCHICAL CLUSTERING
t
n
n
£ S04 CU CO ZN PB NOj Nl CO CS CA K FE MG NA CR RB Tl AL SI SR LI MN MS
The vertical distance is proportional to the degree of dis-
similarity between clusters (1-r).
Figure 6-6. Dendrogram of feature clustering for desert
urban particulate matter.
Source: Reference 34
Reprinted by permission of the publisher.
134
-------
be viewed more efficiently. However, no new information is
obtained using the technique that could not have been observed
from a careful study of the correlation matrices. The same sets
of species with mutually high correlations and similar behavior
noted in the matrices will normally be observed in the cluster
analysis.
The cluster analysis technique can also be used to support
findings of chemical element balance, factor analysis, and the
techniques described in Section 6.1. In addition, a knowledge of
the existing emission inventory is necessary in order to inter-
pret the results of the pattern recognition technique properly-
6.5.4 Resource Requirements
Given a sufficiently large data set, the resource require-
ments are:
Manpower Low
Skill Moderate
Computer Necessary
Data Chemical/element measurements
for more than one site
(number of values must be
adequate for statistical
significance)
a Does not consider the laboratory skills required to use the
, instruments indicated.
Specific choice of instrumentation depends upon specific ele-
ments investigated.
6.5.5 Example Application
An example of the application of cluster analysis can be
""5 A
drawn from a recent analysis of air quality in Tucson. Based
on the correlation matrix presented previously in Table 6-6,
hierarchical clustering was performed on the species using 1-r as
a measure of the dissimilarity among species. The resulting
dendrogram is depicted in Figure 6-6.
As shown in the figure, there is a clustering of the species
Caf K, Fe, Mg, Na, Cr, Rb, Ti, Al, Si, Sr, Li, and Mn. These
135
-------
elements are probably of soil origin. The remaining features,
NH4, S04/ Cu, Cd, Zn, Pd, NO.J, Ni, Co and Cs, are well separated
from the soil cluster. This separation suggests different sourc<
and/or different chemical and physical behavior in the atmo-
sphere.
Ni, Co, and Cs exhibit a behavior intermediate between soil
and nonsoil indicating that a portion of these elemental concen-
trations may have a soil origin.
The high degree of correlation between NH. and SO. is
probably explained by the chemical combination of these species
resulting from the acid base processes after S02 is oxidized to
H2S04.
6.5.6 References
34, 54, 61, 76, 77, 78
6.6 FACTOR ANALYSIS
6.6.1 Description of Technique
Factor analysis is related to cluster analysis in its
intuitive approach to the data. It was originally developed to
explain intelligence test scores and has found wide application
in the social and biological sciences. It is a tool for explain*
ing observed relationships between large numbers of variables in
terms of fewer variables. The new variables are linear combina-
tions of the original ones. The reader should be aware that the
description presented in this section is not designed to convey i
complete understanding of the statistical principles and the
mathematical derivation of the technique. To obtain a complete
understanding of factor analysis and the computational procedures
involved in its application/ a standard text on the subject
79
should be consulted.
Factor analysis may be useful in treating atmospheric data
for the following reasons: first, it is not possible to control
136
-------
the values of the variables experimentally, and second, the
introduction of particulates into the atmosphere from a number of
different sources and areas can be described by a variable which
is a weighted sum of factor values.
Two models of factor analysis are described in the literature
*5 A C. A 1 1 T T *T O
the classical or common factor model, ' ' ' ' and the prin-
73 80
cipal components model. '
Classical factor analysis assumes that observed correlations
between variables are the result of regularity in the data. Each
variable is expressed as a linear combination of factors common
to all variables and a unique factor not shared with the other
variables. The values of the variables can be expressed as a set
of n linear equations:
m
Zi = S aii Fi * di Ui' ^m
-------
estimated values of the communalities (variances of each variable
that result from the common factors). The matrix is then called
a communality matrix. There are several methods available to
estimate the communalities. The usual methods involve the high-
est correlation coefficient for each variable or the squared
multiple correlation.
The method found most frequently in the literature is the
squared multiple correlation (SMC). The SMC is found from the
inverse of the correlation matrix (with unity in each diagonal
portion) by the formula:
SMC. = 1 - -ir (eq.9)
j rJJ
where jj = diagonal element for the jth variable in the
inverse of the correlation matrix
The next steps in factor analysis are to diagonalize the
communality matrix, determine the eigenvalues, and determine the
corresponding orthogonal eigenvectors. The factor loadings can
then be calculated from the following equation:
1/2
a.. = a.. X. ' (eq.10)
13 ij x ^
where X. = eigenvalues
a.. = components of the eigenvectors
a.. = factor loadings
The factor loadings represent the correlations of a variable
with the factor of which it is a part. In the factor matrix,
each row of the matrix corresponds to one variable and each
column represents one factor. There are as many factors as
there are variables. However, normally only those factors with
eigenvalues greater than or equal to 1.0 are used.
Using matrix algebra, factors are rotated to point a factor
at a cluster of variables so the factor may be easier to interpr*
138
-------
The object of rotation is to produce a factor matrix with high
loadings in some rows and near zero values in the other rows. A
number of methods of rotation are described in the literature,
including varimax, quartimax, and equimax. The factor matrix is
examined to determine which variables reveal high loadings (0.50
or greater).
An important aspect of factor analysis is that the solution
is not unique. It is possible to produce equally valid sets of
transformations through rotation from the same input data. The
high loadings cannot be defined objectively. They require that
other information be used to select appropriate descriptors for
the factors.
The analyst does not preselect these factors. Rather, they
are combinations of the original variables that explain the
observed variance in the data. The first factor explains more of
the variance than any other factor, the second more than any
other except the first, and so on.
The principal component model mentioned earlier in this
section ignores the unique factor of a variable. It does not
require"communality estimates, since it is interested in total
variation among all variables instead of the common variance.
The calculations of the component solution and the subsequent
rotation are essentially the same as in the classical factor
model.
The final step in factor analysis is the extraction of
factor scores (the measure of each variable on each factor).
These factor scores can be used as independent variables in
regression analysis or as a new data set for further analysis.
6. 6 ."2 Application of Technique
The application of this technique to the quantified species
present in TSP samples leads to a characterization of the aero-
sol. The loadings on the factors give an indication of which
species are-linearly related. For example, one factor may con-
tain high loadings on Pb and Br, indicating the species are
related in the aerosol.
139
-------
The factor analysis technique can be used to identify
general source categories. By analyzing the high loadings in
each factor, general emission categories can be inferred from th
species having the high loadings. In an industrial area, for
example, high loadings on Fe, Mn, and Ti for one factor could bel
expected to be associated with heavy industrial processes, such
as steel making.
The technique does not lead to a quantification of the
impact of a specific source of emissions, nor does it have any
predictive capabilities. It cannot be used to trace changes in
emissions over time.
6.6.3 Relationship to Other Techniques
The primary relationship of other techniques to factor
analysis is in the interpretation of the results of the analysis
The factor loadings obtained for a particular factor need to be
related to some aspect of the physical world. The cluster anal-
ysis, enrichment factor, microscopy, and interspecies correlatio
techniques can be used to aid in the interpretation of the resul
of factor analysis. In addition, knowledge of the existing
emission inventory is necessary to attach meaning to the results
Meteorological variables, such as wind speed, precipitation
and wind direction, can be included as variables in a factor
analysis.
A recent study combined the techniques of factor analysis
and chemical element balance with a simplified diffusion model,
54
Hanna-Gifford, into a factor model. This model accounted for
the observed means, standard deviations, and correlations of the
primary chemical elements in the aerosol. Another study used thi
factor analysis technique to determine which types of sources
8 0
were affecting sampling sites in St. Louis.
140
-------
6.6.4 Resource Requirements
The mathematical formulation of factor analysis is most
simply expressed in matrix notation. Therefore, because of the
extensive computations necessary to reduce large matrices,
factor analysis has become practical only with the ready avail-
ability of large digital computers. Programs from the Statis-
tical Package for the Social Sciences (SPSS) or the BMDP Bio-
medical Computer Programs can be used to carry out the factor
analysis computations described above. The resource requirements
for factor analysis are:
Manpower Moderate3
Skill Higha
Computer Necessary
Data Chemical/element concentra-
tions; meteorological data
optional
Does not consider the laboratory skills required to use the
b indicated instruments.
Specific choice of instrumentation depends upon the specific
elements investigated.
6.6.5 Example Application
A good example of the application of factor analysis can be
drawn from a recent analysis of air quality in Tucson. Using
the correlation matrix presented previously in Table 6-6, a
factor analysis was performed. Table 6-7, presents the results
of the factor analysis for an urban site. This six factor solu-
tion accounted for about 88 percent of the total variance of the
species (see Table 6-8). Examination of the loadings of the
corresponding factors resulted in the interpretation discussed
below and summarized in Table 6-9. Recall that a factor loading
is a correlation coefficient between a species and a factor, and
that the results presented below are only one possible explana-
tion for the factors calculated from the analysis. Different
results could be obtained if data from another city were used or
if a different rotation were performed.
141
-------
Table 6-7. FACTOR LOADINGS FOR AN URBAN SITE
(VARIMAX ROTATION OF SIX FACTORS)
Factor
Species
Zn
Pb
Cu
Cd
mf
so4"
NO/
Nl
Cr
Li
Mn
Sr
Al
SI
Rb
Ti
Mass
Ma
Mg
Fe
Ca
K
1
.08
.04
-.03
.02
-.01
-.03
.17
.52
.85
.94
.94
.94
.97
.98
.87
.95
.84
.82
.93
.91
.87
.89
2
.27
.41
.69
.14
.86
.93
.19
.19
.02
-.03
-.06
-.06
-.05
-.03
.03
-.06
.11
-.05
.01
-.03
.06
-.00
3
.74
.10
-.01
.20
.14
.10
.02
.12
.08
-.09
-.10
-.05
-.07
-.06
-.19
-.11
.01
.36
.24
.35
.37
.39
4
.27
.04
.17
.60
.07
.02
.30
.35
.24
.13
.13
.09
.03
.07
.21
.05
-.04
-.07
-.03
-.09
-.09
-.10
5
-.09
-.74
-.04
-.20
-.20
-.16
-.62
-.15
.07
-.03
-.01
-.12
-.05
.01
-.14
.07
-.03
-.33
.01
-.10
-.08
-.17
6
.02
.05
.35
.02
-.19
-.06
-.01
.36
.29
.18
.20
.03
-.02
-.07
-.05
-.18
-.27
.18
.10
-.05
-.10
.02
Source: Reference 34
Reprinted by permission of the publisher.
142
-------
Table 6-8.
EIGENVALUES OF CORRELATION MATRIX
FOR AN URBAN SITE
EIGENVALUE
13.2
3.46
1.43
1.21
1.06
0.719
0.607
0.517
0.375
0.338
0.239
0.168
0.136
0.102
0.094
0.085
0.080
0.040
0.035
0.026
0.022
0.011
0.005
0.004
% OF VARIANCE
55.17
14.43
5.94
5.03
4.41
3.00
2.53
2.15
1.56
1.41
1.00
0.70
0.57
0.43
0.39
0.36
0.33
0.16
0.15
0.11
0.09
0.04
0.02
0.01
CUMUI.ATIVK 7. np VARIANCE
55.17
69.60
75.54
80.57
84.98
87.98
90.51
92.66
94.23
95.64
96.63
97.33
97.90
98.33
98.72
99.08
99.41
99.57
99.72
99.83
99.92
99.97
99.99
100.00
Source: Reference 34
Reprinted by permission of the publisher,
143
-------
Table 6-9.
ELEMENT LOADINGS ON INDIVIDUAL FACTORS AND POSSIBLE
EXPLANATIONS FOR FACTOR SIGNIFICANCE
Llements
Major Loading
(Minor Loading)
Physical or
Chemical
Significance
1
Mass, Al
Ti, Si
Sr, Rb,
Li, Mn,
Ca, K, Fe,
Mg, Na,
Cr
(Ni, Zn)
Soil
Large
particles
2
+ a
NH ,S04,
Cu
(Ni, Cd,
Zn, Pb,
NO *)
Distant
and/or
diffuse
source
Small
particles
Gas-
particle
conversion
3 4 | 5
Zn Pb
_
NO-
'
i
> i
i *
(Mg, Na, i
Fe, Ca, '
K)
Common Automo-
aliquot
dilution
variance
tive
Gas
particle
conversion
Elements
Major Loading
(Minor Loading)
Physical or
Chemical
Significance
6
Cd
(Zn, Ni,
Cu, Pb,
NO 3")
Unknown
May have
combus-
tion and/
or trans-
port sig-
nificance
. 7
(Li, Mn,
Co, Cu)
Urban
only
May sug-
gest the
signature
of one or
more urban
sources
8
(Ni, Cr,
Cu)
May sug-
gest local
. mining or
other
anthropo-
genic
activitv
9
(Rb)
Unknown
10
(Na)
Background
only
Marine
source
Source: Reference 34
Reprinted by permission of the publisher.
144
-------
The first factor had high loadings on Na, K, Fe, Ca, Mg, Rb,
Sr, Al, total mass, Ti, Si, Li, Mn, and Cr. The variance common
to these species was probably due to their common source, soil.
Factor two has high loadings from SO4, NH., and Cu while Pb,
Zn, and a few other species had some variance in this factor.
From these loadings, it was inferred that the factor corresponds
to a background aerosol from nonlocal sources.
Factor three had a high loading from Zn and medium loadings
from K, Ca, Na, Fe, and Mg. The medium loadings were accounted
for by the chemical analysis procedures used to analyze these
species. Zn does not cluster with these species as shown above
in the dendrogram in Figure 6-6. Therefore, it appears that Zn
originates from other sources.
Cd was the major contributor to factor four. Intermediate
loadings were observed on Ni, NO_, and Zn. It was suggested that
this factor is due to a high temperature or combustion process
since these species have been identified as being enriched in fly
ash.
Factor five is loaded from Pb and NO- and seems to reflect
an automotive source.
Factor six has only intermediate loadings on Cr, Ni, and Al.
One possible explanation for these factors is mining activity or
local soil variations.
As indicated in the above analysis, the factor analysis took
the results of correlation analysis a step further. A number of
variables which contained portions of the variance common to
several species were noted. However, interpretation of the
factors by a researcher is the most critical part of the anal-
ysis.
6.6.6 References
34, 54, 71, 73, 77, 78, 79, 80
145
-------
6.7 INTERPRETATION OF MORPHOLOGICAL DATA
Filter analysis by optical or electron microscopy to iden-
tify the probable generic origins of individual particles is
becoming more frequently used in the assessment of TSP sources.
The microscopy techniques are primarily laboratory analyses and
are therefore not directly comparable with the data analysis
techniques described in this digest. Data analysis following
microscopy has generally been limited to a comparison of micros-
copy results (mass percent of particle types indicative of
different source categories) with estimates of source contribu-
tion from other analyses such as emission inventory/ X-ray fluo-
rescence, or diffusion modeling.
The use of microscopy results to support or extend the
findings of TSP data analyses depends on the confidence placed i;
quantitative results of microscopy. There are a number of
divergent opinions on this topic. Therefore, the laboratory
microscopy technique is discussed here in some detail to provide
a basis for utilizing its output as part of TSP analyses.
6.7.1 Description of Technique
Filter analysis by microscopy is viewed by EPA as a semi-
quantitative technique for identifying generic types of particu-
late in the >l-2 urn particle size mode. Particles are either
viewed in situ through the use of immersion oil or removed and
remounted using probes pr an adhesive. The principal instrument
used for routine analysis of the filters is the polarizing light
microscope. Particles are examined with both transmitting and
reflected light and characterized as to their generic type by
observing morphology (structure and form), transparency, color,
and other physical properties. Other analytical techniques such
as electron microscopy are often used in conjunction with optical
microscopy as a supplement to the morphological identification.
The particles are usually grouped into categories such as min-
erals, combustion products, biological material, and miscel-
laneous, with further subcategorization where possible. A
146
-------
detailed breakdown of particle categories and the average com-
position of 300 filter analyses performed by one laboratory are
81
shown in Table 6-10.
Particle counts and categorization must be made by size
range in order to convert to mass percents. Also, the average
density and shape factor for each particle category must be
determined or assumed. In some cases, complex mathematical
calculations have been made to derive values other than the size
range midpoints to use in estimating the percent by mass for each
particle category. However, estimates of relative contributions
by source category are often limited to such classifiers as
major, moderate, and trace. If percentages are assigned, the
estimate should be rounded to 10 percentile values.
Several studies have been done to assess the reproducibility
of microscopic analysis results. Such studies used reanalyses or
blind replicate samples and results were compared. One study
involved inter- and intra-laboratory analysis of blind replicate
samples. The results showed a wide disparity (as much as a
severalfold difference) among analysts, which suggests a poten-
tially serious shortcoming of the technique. Factors such as
analyst fatigue, misidentification, and misassignment of par-
ticles to source categories were suggested as causes for this
lack of reproducibility.
Particles smaller than 1-2 urn are very difficult to identify
microscopically. It has been shown that TSP is bimodally dis-
tributed with a minimum between the modes occurring in the 1 to 2
um range. These modes are referred to as "fine" and "coarse,"
and the fine mode may include as much as 25 to 50 percent of the
mass on the filter. Thus, any analysis which relies solely on
microscopy may miss a large portion of the fine particle mode.
The typical makeup of the fine and coarse modes have been char-
acterized generically with the coarse mode consisting mostly of
natural or'fugitive particulates and the fine mode consisting
mostly of stack emissions and secondary particles.
147
-------
Table 6-10. COMPOSITE SUMMARY OF FILTER ANALYSES
Components
Quantity, percent
Average Range
Minerals
Quartz
Calcite
Feldspars
Hematite
Mica
Other
Combustion Products
Soot: Oil
Coal
Misc soot
Glassy fly ash
Incinerator fly ash
Burned wood
Burned paper
Magnetite
Carbon black
Other
Biological Material
Pollen
Spores
Paper
Starch
Misc plant tissue
Leaf trichomer
Miscellaneous
Iron or steel
Rubber
Other
(65)
29
21
5
10
<1
<1
(25)
7
5
5
6
2
<1
<1
<1
<1
<1
( 3)
1
<1
<1
1
1
<:L
( 7)
<1
7
<:
3-99
-------
Particle sizing by microscopic analysis can give some clues
as to the size distribution in the large mode, but is hampered by
agglomeration on the filter and also by the randomness of occur-
rence of very large particles within the field of view of the
microscopist. Also, a heterogeneous distribution of particle
sizes may be present through the filter such that the large
particles and agglomerates may not penetrate as far into the
filter mat as the fine particles. Unless fields at several
focusing depths are analyzed in situ, this particle stratifica-
tion may bias sizing and identification studies.
Microscopy has certain limitations in the accuracy by which
particles can be identified. Crystalline compounds such as
quartz or calcite can be easily identified but the inability to
designate whether these constituents were entrained by the wind
or resuspended by man's activity is a severe limitation. Some
particles, such as rubber, are sometimes mistaken for combustion
products. Many fine opaque particles, such as automotive, diesel
fuel oil, and coal combustion products, cannot be easily distin-
guished from each other. Using optical and electron techniques
in concert can provide more accurate and reliable particle
identification, but some of the problems associated with particle
sizing and sample handling may still exist.
Finally, sample preparation procedures may bias results.
Removal of particles from the filter by manual manipulation
(particle picking) or transfer (using an adhesive substance such
as Aroclor) only samples the uppermost layers of particles. If
any heterogeneous size stratification does exist, as suggested
above, the removal procedure exacerbates this problem. Removal
by ultrasonics is sometimes used, but this may disaggregate the
agglomerates and solubilize some particles.
6.7.2 Applicability of Technique
Microscopy has had limited success in identifying specific
sources of particulate matter from routinely collected ambient
samples. It seems to have had the most success in either: (1)
149
-------
very general qualification of the problem (e.g., mineral matter
versus combustion as generic categories) or (2) very specific
studies around a suspected source where the morphology of emitted
particles is unique to the area. It is limited in capability by
problems associated with particle sizing, sample preparation, the
identification and assignment of particles to generic categories,
and reproducibility of results, as discussed above.
Microscopy definitely characterizes the aerosol; it provides
information on both the size distribution and physical properties
of the particles. In conjunction with simultaneous analyses for
elemental composition of the particles, these techniques provide
more data on characteristics of the aerosol than any other avail-
able TSP analyses.
The technique has been employed most frequently to determine
impacting source categories, as described in Section 6.7.1.
However, its results have been overextended in some highly quan-
titative analyses of the contributions from single sources or
source categories.
6.7.3 Relationship to Other Techniques
The most important relationships are with techniques that
simultaneously analyze filter samples for elemental composition.
These include X-ray diffraction, energy dispersive X-ray analyses
(X-ray fluorescence), electron microprobe (electron spectroscopy
for chemical analysis), and neutron activation. Most of these
techniques give the elemental composition of the entire sample on
the filter, but some (at least energy dispersive X-ray analysis)
can determine the composition of individual particles for even
greater resolution of morphological/elemental groupings.
In addition to being a particle sizing technique, microscopy
is closely related to the cascade impactor sizing technique.
Microscopic analyses have been performed on the individual filter
stages of impactor samples to determine the morphology and prob-
able generic origin of particles on each stage. This analysis
also provides a direct comparison of the two sizing techniques.
150
-------
Microscopy can be used as a supportive or confirmatory
analysis with any other technique that determines impacting
source categories.
6.7.4 Resource Requirements
Microscopic analysis of filter samples is quite time con-
suming and requires a high level of skill. Analysis by optical
microscopy alone usually takes two to four hours per filter plus
sample handling and preparation time. In contrast, requirements
for data analysis of microscopy results are minimal. The re-
source requirements summarized below are for the microscopy work:
Manpower High
Skill High
Computer Optional (for calculating mass
percentages)
Instrumentation Microscope and accessories
6.7.5 Example Application
In order to determine the feasibility and usefulness of
routine microscopic examination of hi vol filters, the Texas Air
Control Board (TACB) submitted 126 samples for optical microscopy
82
analysis. These filters had previously been analyzed by X-ray
fluorescence spectroscopy (XRF) to identify and quantify source
contributions, so the microscopy data optimally would confirm the
XRF findings and aid in source identification where XRF data were
inconclusive or where particle types were not identifiable through
XRF.
Particles were identified under the polarizing microscope by
noting their color, shape, optical characteristics, crystal
system, and other distinguishing properties. They were assigned
to the same particle categories as listed in Table 6-10. Volume
percentages were determined for each particle category, then an
assumed density factor was applied to determine the weight per-
centages .'
151
-------
Additional sections of five filters were resubmitted for
analysis. Three of the duplicate pairs had overall comparisons
within the ±8 percent precision claimed by the laboratory per-
forming the work. However, the other two samples had errors in
the range of 30 to 60 percent. Because of this discrepancy, TACI
considered the microscopy results only as supportive evidence to
the XRF and drew no conclusions on source contributions based on
microscopic evidence alone.
In general, microscopic analysis proved to be supportive of
the XRF results. In cases where industrial particles were iden-
tified microscopically, elemental analysis had shown the presenc*
of significant amounts of the same material. For traffic-relate<
emissions, filters with a high Pb content and a good Pb/Br corre
lation also had a significant percentage of tire rubber. Iden-
tification of local soil types and quantitative determination of
fugitive dust impact was accomplished by XRF. Samples with high
Si and Ca values had a large proportion of silicates and lime-
stone identified by microscopy. Samples with high Al values had
significant amounts of feldspars and clays microscopically.
However, correlations between XRF values and particle mass per-
centages were occasionally difficult to show.
The results of microscopic analysis of filters from one hi
vol site near the Houston ship channel are shown in Table 6-11.
Table 6-11
MICROSCOPIC ANALYSIS RESULTS. SITIL115004
Date
12/09/71
1/22/72
3/11/73
12/30/73
3/12/74
8/27/74
1/30/75
2/23/75
7/17/75
12/08/75
TSP
(,us/m5j
268
317
334
473
291
276
138
204
217
332
Minerals
47°,
27
10
65
35
14
52
4
63
13
Combustion Iron
51°,
0
90
0
5
24
21
90
1
87
2*
8
0
3
60
2
26
0
11
0
Other
0°.
65
0
52
0
60
0
0
25
0
Wind
Direction
SE
S
NE
SE
SE
SE
SK
NW
SE
N
Comments
Ship*
Coke
Ship
Ship
Ship*
Coke
Ship
Coke
Coke - Fugitive dust from coke storage pile.
Ship Fugitive dust resulting from ship unloading activities.
Ship*- Unloading activities one or two days prior to actual sampling date.
Source: Reference 82
Reprinted by permission of the publisher.
152
-------
6.7.6 References
59, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93
153
-------
7-0 INTERPRETING PARTICLE SIZE DATA
Several different techniques for analyzing total suspended
particulate and chemical species data have been presented in the
preceding chapters. Used individually or interactively, they can
tell the analyst much about the nature of the aerosol and its
sources. But regardless of how well they are used, these tech-
niques will always be limited by their blindness to the physical
property of the aerosol that is most fundamentally determinative
of the impact of that aerosol upon the health of the human orga-
nism: particle size. Particles in excess of ^15 um do not
normally become trapped in the respiratory system, hence they will
not cause any adverse health consequences. Particles in excess
of ^3.5 um cannot normally intrude into the terminal bronchial
and alveolar portions of the human lung. As a result, it is
desirable for the analyst to have access to techniques which
assess the particle size distribution of the measured aerosol.
There are several different methods currently being used to
measure an aerosol's particle size distribution, among which are
filter analysis by microscopy, cascade impactors, dichotomous
samplers, diffusion batteries, optical counters, and electrical
analyzers. It is not the purpose of this chapter to discuss the
theory and construction of each of these methods, but some brief
comments are warranted, given the methods potential use in
providing data to which various analytical techniques can be
applied.
Table 7-1 briefly summarizes the size ranges that are
94 95 96
reportedly sampled by the above-mentioned methods: ' '
154
-------
Table 7-1. SIZE RANGES SAMPLED BY CURRENT METHODS
Method Size range, urn
Diffusion battery 0.002 - 0.2
Electrical analyzer 0.0032 ~a1.0
Cascade impactor ^*^b"~ ^
Dichotomous sampler <16
Optical counter >1.0
Microscopy >l-2
Final filters collect a composite sample below some cutoff size;
. e.g., 1.1 um for typical five-stage device.
One intermediate stage cutoff occurs at 3.5 um.
Of these methods, only cascade impactors and microscopy are
commonly used in ambient aerosol investigations. The other three
19 97
have been used only in advanced research. ' Thus, currently
available equipment indicates that only particles in excess of
0.5-1.0 um can be successfully stratified into size classes.
This is a severe limitation. It has been shown that ambient
aerosol size distributions are almost always bimodal in surface
or volume, are frequently trimodal in surface area near sources
of fresh combustion aerosols, and that the minimum between the
two most common modes occurs in the 1 to 2 um range (see Figure
97
7-1). As a result, the most commonly used methods provide
meaningful distribution data only for larger particles. Such
distribution data are of limited utility, since it has also been
shown that many sources emit particulate matter in the submicron
range and that most secondary particulates also can be found in
98
that range.
Given these limitations, the remainder of this chapter
concentrates upon ways of analyzing particle sizing data that are
reasonably available to air pollution control agencies. Two
general techniques for analyzing such data have been commonly
used:
Frequency distribution comparisons
Species-specific size distribution comparisons
155
-------
Figure 7-1. Normalized frequency plots of number, surface,
and volume distributions for the grand average October
1977 measurements at Denver's City Maintenance Yard.
Note the bimodal distribution of volume and the fact
that each weighing shows features of the distribution
not shown by the other plots.
Source: Reference 97
Reprinted by permission of the publisher.
156
-------
7.1 FREQUENCY DISTRIBUTION COMPARISONS
7.1.1 Description of Technique
Particles are most often found to be lognormally distributed
in aerosols sampled with cascade impactor type devices. As a
result, the size distribution of such data can be completely
described in terms of the geometric mean particle size (d ) and
the geometric standard deviation (a ). Two samples with the same
d and a values can be interpreted as being drawn from the same
air mass. Therefore, this technique consists of comparing values
obtained from two or more different sites, or from the same site
for two different time periods or sets of meteorological con-
ditions. Such a comparison can be performed numerically or
graphically.
In those cases where smaller particle sizes can be sampled,
their distribution is often found not to be lognormal. This
results either from the fact that emissions from a dominant
impacting source are bimodal or because more than one source
(each with a different size distribution) is contributing to the
aerosol. Size distributions can be investigated for evidence of
such polymodality.
7.1.2 Applicability of Technique
Characterization of the aerosol is the primary use of this
technique. Without it, the analyst cannot judge what portion may
be respirable, hence potentially hazardous to the public health.
Similarly, such data are necessary for investigating the rela-
tionship between visibility and suspended particulate concentra-
tions .
The technique is useful as an indicator of the general types
of sources that are impacting upon a sampler. Where coarse
(i.e., >3.5 urn) particles dominate, abrasive sources such as soil
dust, fugitive dust, or sea spray can be inferred. When fine
particulates dominate, fuel combustion sources and/or secondary
particulates may be inferred.
157
-------
The quantitative impact of specific sources cannot be
deduced with this technique. Use of refined instrumentation
(diffusion battery, electrical analysis, etc.) can permit the
analyst to deduce the quantitative impact of nearby fresh com-
97
bustion sources.
7.1.3 Relationship to Other Techniques
Given an adequate number of samples, the temporal patterns
in particle size distributions could be assessed. Similarly,
spatial pattern techniques described in Chapter 3 and techniques
assessing the effect of meteorological conditions could also be
used in conjunction with particle size distribution data. A
statistically adequate number of samples is rarely available to
the analyst, however, and conclusions derivable from analysis of
the data may be limited to the relative contribution of fugitive
and windblown dust.
Source particle size distribution data are quite important
in adapting diffusion models to account for deposition and fall-
out effects. Certain recently developed models attempt to take
28
these effects into account. Furthermore, particle size dis-
tributions of emission data can be used to implicate certain
source categories (e.g., windblown dust, automotive emissions).
Techniques for analyzing elemental data can also be adapted
for use with this technique. Perhaps the most obvious inter-
relationship involves the combination of interelement correlation
data with particle size distribution data to investigate the
relative contribution of soil and fugitive dust sources. Where
the soil group (i.e., Al, Si, K, Ti, et al.) intercorrelation is
high and the particle size distribution is greatly skewed toward
the larger particle size ranges, then the predominance of such
sources may be inferred. This is especially true if analysis of
meteorological effects shows the expected impact of precipi-
a
tation. Other chemical/elemental techniques such as chemical
element balance and enrichment factor can be used in conjunction
with this technique. No clear relationships with the microinven-
tory, trajectory analysis, pattern recognition, and factor anal-
ysis techniques can be seen.
158
-------
7.1.4 Resource Requirements
The resources the analyst needs in order to apply this
technique are not severe unless advanced equipment is utilized.
The following summary is based upon the use of cascade impactor
data. Given in parentheses are estimates of resource require-
ments when more sophisticated equipment is used.
Manpower Low (moderate)
Skill Moderate (high)
Computer Not required (necessary)
Data TSP data stratified by size
ranges down to 0.5 urn (number,
surface, and volume distribu-
tions down to 0.01 um)
7.1.5 Example Application
Particle size distribution data were collected in Alaska
using Andersen sizing head modifications (cascade impactor) to
99
standard high volume samplers. Totals of 23, 37, and 13 24-h
samples were collected from Fairbanks, Anchorage, and Juneau
monitoring sites, respectively. The resulting size distributions
were then compared to similar data that had been collected for
seven large urban areas (see Figure 7-2). Mean particle sizes
were much larger for the Alaskan areas than for the cities ad-
dressed, thus leading the analysts to conclude that much of the
Alaskan particulate matter was not of industrial origin.
7.1.6 References
19, 28, 85, 94, 95, 96, 97, 98, 99, 100
7.2 SPECIES-SPECIFIC SIZE DISTRIBUTION COMPARISONS
7.2.1 Description of Technique
Size distributions for elements and chemicals can be devel-
oped with cascade impaction-type devices and used in a number of
ways. Urban/rural comparisons can be made, size distributions by
wind direction can be produced, enrichment factors can be cal-
culated, and the effect of meteorological conditions can be
159
-------
100-0
90.0
80.0
70-0
60.0
50.0
40.0-
30.0
20.0
10.0-
9.0-
8.0-
g 7.0-
3 6.0
^ 5.0-
(1) 4.0-
0)
-H
(U
•H
O
•H
4J
3.0|-
2.0
.2
/:/ »v
-------
assessed. Other combinations are possible as well. All that is
required is for chemical/element measurements to be made for each
particle size range of interest, and for resulting data to be
stratified in a manner that appears useful to the analyst.
7.2.2 Applicability of Technique
This technique provides the most refined characterization of
the aerosol of any discussed in this digest. Given sufficient
data and taking meteorological conditions into account, the
spatial and temporal variation in species concentrations by
particle size can be described. Limitations imposed by currently
available sampling techniques must be noted, however.
Major source categories impacting upon a sampler can be
identified with this technique. For example, the impact of
soil dust can be assessed by observing variations in Al, Si, and
K concentrations in the coarse particulate mode under different
precipitation conditions. Likewise, stratifying elemental size
distribution by wind directions will permit one to judge whether
fuel combustion or abrasive sources are prevalent in each wind
direction. The elemental mix can then be related to with likely
source types.
It is not yet possible to use this technique to quantify the
TSP impact of a specific source or to assess the effect of changes
in emissions or the location of a source.
7.2.3 Relationship to Other Techniques
It has already been noted that this technique can be used
interactively with most of the others presented in this digest.
Exceptions include emission inventorying, microinventorying,
trajectory analysis, factor analysis, pattern recognition, and
diffusion modeling techniques.
7.2.4 Resource Requirements
It is necessary to have access to a considerable amount of
12- to 24-h cascade impactor data, or to use more sophisticated
161
-------
size distribution-related equipment. Further, the resources
needed to produce species data can be burdensome. The following
estimate assumes a 10-site network being analyzed for 20 elements
and stratified into size ranges by cascade impactors:
Manpower Moderate
Skill Moderate
Computer Optional
Data Same as for Section 7.1, plus
chemical/element analysis;
meteorological data optional
a Manpower commitment is dependent upon whether the data are
subsequently computerized and the number of interrelationships
with other techniques that are developed. Laboratory
manpower is not included.
7.2.5 Example Application
The example presented herein is drawn from a study of urban
aerosols in Toronto, Ontario. Concentrations of 25 elements at
four different locations were determined with neutron- and photon-
activation analysis. Size distributions were produced for most
of those elements with a five stage Andersen sampler. Approxi-
mately 25 sampling periods of 24-h each were obtained at each
site.
Most atmospheric concentrations did not vary from site to
site. Exceptions noted were Sb, As, Cr, Cu, Pb, and Zn. The
increase in Sb, As, and Pb at one site was attributed to a nearby
battery recycling plant. The Cu increase was attributed to brush
wear on the high volume sampler. Mass median diameters of each
element were estimated by plotting the logarithms of the effec-
tive cutoff diameters for each stage versus the cumulative percent
mass <_ each cutoff diameter on log probability paper. The ele-
ments Al, Ca, Co, La, Mg, Fe, Sm, Sc, Na, and Ti were found to be
concentrated on larger particles, to have relatively low enrich-
ment factors (see Table 7-2), and to be well correlated with each
other. Other elements, particularly V, Pb, Cl, and Br, were
found to be concentrated on smaller particles and, with the
exception of V, to have relatively high EF's.
162
-------
Table 7-2.
CLASSIFICATION OF AIRBORNE TRACE ELEMENTS
ACCORDING TO SIZE
Element
Aluminum
Magnesium
Calcium
Samarium
Iron
Scandium
Titanium
Lanthanum
Sodium
Cobalt
Potassium
Manganese
Copper**
Arsenic .
Chromium
Iodine
Mercury
Nickel
Zinc
Antimony
Vanadium
Lead
Chlorine
Bromine
Average
cone
in Toronto,
ng/m3a
2100
1400
5300
0.33
2200
0.27
170
2.4
650
1.0
870
74
<330
12
26
<4
<5
21
320
6.9
14
970
1200
290
Enrichment,
factor, EF
1.0
3.6
6.8
1.9
2.3
0.7
1.3
2.0
1.0
3.2
1.1
4.0
<420
260
14
<320
<6300
18
200
1300
5.7
2500
150
3700
Approx
mass
median
diam, um
8
7
7
6.5
6
6
6
4.7
4
4
2.5
2.4
1.5
1.5
1.3
1.3
1.3
1.2
1.2
1.2
0.9
0.7
0.6
0.3
Percent
of mass
<1.1 um
in diam
12
11
8
21
15
12
15
15
30
19
27
36
42
37
32
41
45
45
47
40
54
51
55
71
Average of all Andersen and high volume samples, excluding con-
centrations of antimony, arsenic, and lead near industrial
sources.
EF = atmos cone normalized to Al/crustal abundance normalized
to Al.
Size data are for a typical summer aerosol, excluding samples
influenced by local sources.
Cu concentration represents an upper limit because of con-
tamination.
Source: Reference 63
Reprinted by permission of the publisher.
163
-------
No specific conclusions were drawn as to the sources causing
elevated element concentrations in the smaller particle size
range/ but combustion sources were implied.
7.2.6 References
11, 58, 62, 63, 101, 102
164
-------
8.0 DESIGNING STUDY AND INTERPRETING RESULTS
A wide variety of individual particulate analysis techniques
have been presented in the previous chapters. Each was discussed
in terms of its general applicability, resource requirements, and
interrelationships with other techniques. That information is
necessary in order to permit the analyst to decide whether or not
a certain technique applies to the problem he faces and whether
or not he can afford to use it. However, more information is
necessary for the analyst to integrate these techniques and
synthesize the various findings that emerge, in this chapter,
the analyst will be presented some guidance as to how to select,
apply/ and interpret the results of these techniques in light of
the specific problem he faces.
8.1 SELECTING TECHNIQUES
Three general questions confront the analyst preparing to
study a TSP problem:
0 What is the general nature of the particulate problem
that needs to be investigated?
0 What techniques can be used to define this problem
better and then to resolve it?
0 How much effort can be expended?
The general nature of the problem at hand can vary con-
siderably. The analyst might be interested in learning how the
chemical composition of the aerosol changes over space and time.
He might want to define the Pb emission rate from automobile
exhaust in terms of g/VMT. He might want to develop a regional
control'strategy which will ensure the attainment and maintenance
165
-------
of TSP NAAQS throughout a metropolitan area. Or he might want to
investigate a host of other problems.
For the purposes of this section, it will be assumed that
the analyst is primarily interested in developing a regional TSP
control strategy. This is not meant to denigrate the importance
of the other types of problems he faces. Rather, it is simply to
focus attention upon the type of problem which will probably be
uppermost in the mind of the typical user of this digest.
Once he has generally identified his problem, the analyst
will need to define that problem as well as possible and to
select those techniques which are most likely to help solve it.
This initial decision as to which techniques to use can be aided
by the early application of a simple screening procedure. This
screening procedure consists of assessing the applicability and
task effectiveness of the various techniques, and then assessing
their applicability in terms of resource requirements and the
sponsoring organization's resources.
The general applicability of each technique, as well as
resource-related information which will be explained in the next
few paragraphs, is summarized in Table 8-1. Each technique is
ranked on a scale of 1 through 5 (where 5 indicates the greatest
potential) for each of the general problem areas most pertinent
to TSP control strategy development. As that table and Table 8-2
below indicate, different techniques are more effective at dif-
ferent tasks and none of the techniques is effective* at all of
the tasks.
166
-------
Table 8-1. SUMMARY OF TECHNIQUES
Technique (Section)
Long term trends (2.1)
Seasonal patterns (2.2)
Daily patterns (2.3)
Diurnal variation (2.4)
Emission patterns (2.5)
Site classification (3.1)
Inter- site correlation (3.2)
Pollution rose (3.3)
Upwind/downwind (3.4)
Correlation and regression (4.1)
Decision-tree (4.2)
Precipitation (4.3)
Wind speed (4.4)
Trajectory analysis (4.5)
Emission inventory (5.1)
Microinventory (5.2)
Diffusion modeling (5.3)
Temporal, spatial, and meteoro-
logical (6.1)
Enrichment factor (6.2)
Chemical element balance (6.3)
Interspecies correlations (6.4)
Pattern recognition (6.5)
Factor analysis (6.6)
Microscopy (6.7)
Frequency distribution (7.1)
Species-specific (7.2)
Task rankings
Charac- Identify Quantify
terize source specific
aerosol categor. src.imp.
2
2
2
2
1
3
4
2
2
1
1
1
1
2
1
1
4
4
3
2
4
4
4
4
3 (4)
5
1
1
1
2
2
2
1
2
1
2
2
2
2
2
3
4
4
2
2
4
2
3
3
3
2
4
1
1
1
1
2
1
1
2
3
1
1
1
1
1
1
1
5
2
1
1
1
1
1
2
1 (3)
1
Resource requirements
Man- Compu-
power Skill ter Data
2
2
2
4
4
4
2
2
2
6
6
2
2
6
10
6
4
2
2
4
4
4
6
8
2 (4)
4
1
1
1
1
1
1
2
1
1
3
3
1
1
3
2
2
4
2
2
5
3
4
5
5
2
2
2
2
2
1
1
2
2
2
2
3
3
2
2
3
2
1
3
2
1
3
2
3
3
2
(3) 1 (3)
2
1
1
1
2
2
1
1
2
2
2
2
2
2
2
1
1
1
5
5
5
5
5
5
2
4
5
Cost-effectiveness
Charac- Identify Quantify
terize source specific
aerosol categor. src.imp.
.3
.3
.3
.3
.1
.4
.6
.3
.3
.1
.1
.1
.1
.1
.1
.1
.3
.4
.3
.1
.3
.3
.2
.2
.3
.4
.2
.2
.2
.3
.3
.3
.1
.3
.2
.1
.1
.3
.3
.1
.2
.4
.3
.2
.2
.2
.1
.2
.2
.2
.2 (.1)
.3
.2
.2
.2
.1
.3
.1
.1
.3
.4
.1
.1
.1
.1
.1
.1
.1
.4
.2
.1
.1
.1
.1
.1
.1
.1 (.2)
.1
-------
Table 8-2. EFFECTIVENESS RANKINGS
Aerosol
characterization
Source category
identification
Specific
source impact
quantification
Species-specific size
distribution
Intersite correlation
Diffusion modeling
Temporal, spatial,
and meteorological
variations of
chemicals
Interspecies correla-
tions
Pattern recognition
Factor analysis
Microscopy
Diffusion modeling
Microinventorying
Chemical element
balance
Species-specific
size
distributions
Emission
inventorying
Factor analysis
Pattern recognition
Diffusion modeling
Upwind/Downwind
Emission patterns
Temporal, spatial,
and meteorological
variations of
chemicals
Pollution rose
Microscopy
It should be noted that these rankings are necessarily
somewhat subjective due to the fact that there are no absolute
measures of the relative accuracy of these techniques. One can
assess the accuracy of each technique in its own terms. For
example, the true correlation among TSP measurements at different
sites can be assigned confidence limits which are dependent upon
the sample correlation and the number of samples. Likewise, the
accuracy of microscopy results can be assessed by analyzing
replicate samples. But there is no accepted way of comparing the
results of one technique with those of another.
Once he has conducted this initial screening, the analyst
will want to know whether or not he can afford to implement those!
techniques which he considers most promising. To answer this
question he can use the information presented in Table 8-1.
Table 8-1 summarizes resource requirements in terms of
skill, computer, and data needs. It would be best to present
168
-------
these data in common units (e.g., dollars) so that the overall
resource needs could be calculated in a direct manner. Unfor-
tunately, there can be so much variation in the cost of applying
any one technique that this is not really possible. A diffusion
model run may, for example, cost anywhere from $20 to $1,000+
depending on the complexity of the model and the situation to
which it is being applied. As a surrogate for dollar cost,
skill, computer, and data requirements are rated at 1 through 5
(5 being most resource intensive). Due to the greater cost that
is normally associated with manpower activities, that resource is
given twice the weight of the other three and is rated at 1
through 10. It has been assumed that the cumulative resources
required to apply a given technique are represented by the sum of
four individual resource areas (e.g., the combined resource
requirement for the" chemical element balance technique is 17) .
The analyst must inventory his organization's resources to
determine whether or not they are sufficient to apply the set of
analytical techniques which he previously identified as being
most promising. Those techniques which exceed his agency's
resources must be excluded, unless the analyst is able to obtain
financial or in-kind assistance from another organization.
In addition to excluding resource-excessive techniques, the
analyst must also exclude techniques which are dependent upon
those excluded techniques. Diffusion modeling is, for example,
dependent upon emission inventorying. if an organization cannot
afford to conduct an inventory, then it will not be able to apply
a diffusion model.
The analyst will also be concerned about getting the most
for his money. Thus, he will be concerned about cost-effective-
ness as well as cost or effectiveness alone. By dividing indi-
vidual task rankings by combined resource requirements, a general
assessment of each technique's cost-effectiveness can be derived.
The chemical element balance technique, for example, has been
assigned task rankings of 2, 4, and 1 for aerosol characteriza-
tion, source category identification, and source impact quantifica-
tion, respectively, and a combined resource requirement of 17.
169
-------
This results in cost-effectiveness ratings of 0.1, 0.2, and 0.1.
As shown in Table 8-1, the techniques listed in Table 8-3 below
are assessed as being most cost-effective:
Table 8-3. COST-EFFECTIVENESS RANKINGS
Aerosol
characterization
Source category
identification
Specific
source impact
quantification
Intersite correlation Microinventory
Species-specific size
distribution
Site classification
Species-specific
size
distributions
Precipitation
Wind speed
Pollution rose
Emission patterns
Diffusion modeling
Upwind/Downwind
Diffusion modeling
Pollution rose
Emission patterns
At this point, the analyst will be left with a general set of
cost-effective techniques that can provide answers to his spe-
cific questions without imposing excessive strain on his orga-
nization's resources. He will now want to apply those techniques
to the specific problem at hand.
8.2 APPLYING TECHNIQUES AND INTERPRETING RESULTS
Two more specific questions confront the analyst at this
point:
In what sequence should the available techniques be
applied?
How should conflicting results be handled?
170
-------
In line with the reasoning previously presented, it is
recommended that the available techniques first be used to
define the existing problem as explicitly as possible. The
resulting definition should be in terms of the regional/local,
short-term/long-term, improving/worsening/no change nature of the
problem and should be discussed in terms of statistical signifi-
cance.
Whether or not the problem is long-term and/or short-term
should be assessed for each site and described in terms of prob-
ability. Hence, the analyst should calculate the probability
that the primary annual NAAQS (75 ug/m ) is exceeded for the
entire year and the number of days that the secondary 24-h stan-
dard (150 ug/m ) would be expected to be exceeded during the
entire year.
The analyst should then assess whether or not the problem is
region-wide or local. This can be accomplished by plotting the
measured concentrations geographically, by determining intersite
correlations, by applying a diffusion model, by using the site
classification process described in Section 3.1, and/or by apply-
ing a trajectory analysis. Correlations should be qualified by
also determining their true confidence limits (see Figure 1-1).
Thirdly, the trend of past data should be determined. The
analyst should determine whether or not there has been a statis-
tically significant trend in annual averages.
At this point, the analyst should have a pretty good defini-
tion of the specific problem he faces. For example, he should be
able to say that there is a region-wide primary annual NAAQS
problem which is worsening or a local short-term secondary 24-h
NAAQS problem which is remaining generally stable. And he should
be able to attach statistical significance to his definition.
However, it may well be that there are not enough sampling sites
in the region or number of samples at each site to do so. In
this case, the analyst would conclude that he cannot define his
problem clearly. If this occurs, he must weigh the consequences
of developing a TSP control strategy with an inadequate problem
171
-------
definition versus the consequences of delaying major control
strategy implementation while he conducts a special study which
has been designed to fill the gaps in available data.
Once he has defined the problem to his satisfaction, the
analyst should determine whether that problem is primarily
caused by a clearly identifiable source. To do so, he can
compare the temporal patterns of the TSP data with the activity
patterns of a suspected source, perform upwind/downwind analyses,
apply a diffusion model, calculate a pollution rose, or—if
agency resources are sufficient—calculate interspecies correla-
tions, analyze filters microscopically, perform factor and clus-
ter analyses, apply the chemical element balance technique,
and/or evaluate the source contributions indicated by the pre-
viously applied diffusion model. If there is a major source
significantly responsible for measured concentrations, each of
these techniques should point out that source.
In some cases, however, results will conflict. When this
occurs, the analyst should evaluate the techniques in terms of
the statistical significance of their results. Of the techniques
identified above, only the microscopy and chemical element bal-
ance technique cannot be assessed in terms of confidence limits
about the mean. If, for example, the pollution rose fails to
indicate a source identified by all of the other techniques, a
statistical analysis may reveal that the shape of the rose may
not identify that source simply because of the small number of
samples upon which each directional mean is based. If the results
of the various techniques are each statistically significant but
still conflict, then the analyst should see if a clear majority
of the techniques indicate one source. If that is the case, he
should accept the results as sufficient evidence of that source's
role. If there is no clear majority, he should attach more
weight to the techniques ranked highest in task effectiveness
(see Table 8-1).
If no single source can be clearly identified, the analyst
should attempt to identify the source categories which have the
172
-------
greatest effect upon TSP concentrations. Where the first stage
of the study indicated a localized problem, a microinventory
should be performed in the vicinity of the violating sampler.
Where the problem was regional in nature or where additional
evidence in support of the microinventory is desired, it is
suggested that species-specific size distributions, the effects
of precipitation and wind speed, and the results of pollution
rose and emission pattern calculations should be analyzed. Where
resources are sufficient, other techniques can be applied as
well.
If no source category is shown to be the primary cause of
violations, then a special study which takes the techniques
discussed in this digest and the resources of the agency into
consideration should be initiated.
The analyst may also consider applying some of the other
meteorology-related techniques discussed in Chapter 4. Spe-
cifically, stepwise multiple linear regression or the AID deci-
sion tree technique can be used to determine the percent of
variance in TSP concentrations that can be explained by meteo-
rological variables.
In some cases, the analyst will conclude that he has not
defined the problem clearly, found a culpable source or source
category, or found that meteorological conditions are having an
effect. In these cases, he should determine the reasons for this
inability and initiate their correction if at all possible.
8.3 EXAMPLE APPLICATIONS
8.3.1 Small City/Simple Aerosol
The first example is a Rocky Mountain town of approximately
50,000 people, located at the juncture of two fairly broad river
valleys. Two mountain ranges tightly surround the valley within
which the city itself is located. Precipitation is normally
rather slight (annual average of 14 in.), but the cool temper-
atures and low average wind speeds result in a climatic factor of
173
-------
only about 10 (compared to 100 at Garden City, Kansas). There
are large numbers of unpaved roads and lots in the city and
surrounding countryside and three large wood products industry
point sources located generally upwind of the city. Growth in
population is expected to average 2-3 percent per year, but the
growth in basic industry is uncertain.
There are two governmental agencies responsible for air
pollution control activities in the city: a 2-man local agency,
and a 24-man state agency. Given their sizes, these agencies are
staffed with a broad range of skills and expertise. The state
agency has ready access to computers required for sophisticated
analyses. The data base for the area is somewhat limited, con-
sisting primarily of the following:
0 Four high volume sampler sites, each with more than
three years of standard TSP data, and three other sites
with less than one year of data
0 Meteorological data from one site 5-10 miles upvalley
of the city
0 A 1974 area source emission inventory plus updated
point source data
0 No chemical, elemental, or particle size data are
available; the agency does have a morphology capability
The two agencies seek to define their TSP problem clearly,
to identify major contributing source categories, and to quantify
the impact of specific sources. Hence, it is assumed that they
would be generally interested in applying all the techniques
discussed in this digest. As indicated previously, though,
neither agency has sufficient resources to apply each individual
technique without access to outside resources. It is difficult
to say precisely whether any one technique exceeds the specific
resources available to the specific agency in question. For the
purpose of this discussion, it will be assumed that manpower
resources are exceeded by those techniques ranked 8 or higher in
that portion of Table 8-1, and that data resources are exceeded
174
-------
by those techniques rated 4 or higher unless it has been explic-
itly noted that the agency has the resources necessary to imple-
ment that specific technique. Techniques which are cost-inef-
fective may also be excluded if the analyst so desires. For
illustration purposes/ it is assumed herein that cost-ineffective
techniques are those which are rated no higher than 0.2 for any
one task area and which are rated at 0.1 for the other two areas.
Figure 8-1 illustrates the process of screening out tech-
niques which cannot be used in this example city. According to
that figure, 13 techniques survived the exclusion process for the
example city. At this point, the analyst must apply these tech-
niques and interpret their results. Table 8-4 suggests how these
techniques could be applied to the example city and suggests some
qualifications that would probably be attached to the results of
each of these techniques.
Creating a hypothetical situation, let it be assumed that
the following conditions are found after applying the techniques
presented in Table 8-4:
0 Only one of the sites exceeds either NAAQS. It is
located in the downtown area 10 miles down-valley from
the major point sources.
0 A good correlation is found between the two TSP sites
which share a common wind direction from the wood
products point sources and poor correlations are found
for the other sites.
0 There is a statistically significant slight decrease in
annual TSP concentrations at two sites (those which
correlated well).
0 The activity patterns of the major point sources do not
correlate well with temporal TSP patterns.
0 A pollution rose at the downtown site shows no statis-
tically significant pattern.
0 TSP concentrations at the downtown site are slightly
higher than those at the site lying between it and the
point sources. The difference is not statistically
significant.
175
-------
Q)
§, Identify
•H techniques
.£ desirable
^ to solve
E-" problem
Exclude
techniques
that are Exclude Exclude cost-
resource- dependent ineffective
excessive techniques techniques
Select
remaining
techniques
"> 0
9 R
-> 1
-> o
3T -I..
41 „ , . ,
.1
40
41
. J
4 A
4f ,
. --
51
. X
5->
. J
6 A
. o
. ± —
Figure 8-1. Screening techniques for example city No. 1.
176
-------
Table 8-4. APPLYING TECHNIQUES TO EXAMPLE CITY NO. 1
Task
Applicable
technique
Comments
Assess regionality 3.1
3.2
Determine direction 2.1
of past data
Determine whether a
specific source is
clearly culpable
Determine whether a
source category is
culpable
2.5
3.3
3.4
2.5
3.3
4.3
4.4
Investigate meteoro- 4.3
logical effects 4.4
Probably not very useful; terrain
is essentially the same at all
sites
Current data will be very precise
Trends will be difficult to
see due to relatively short
time period available
Dependent upon analysis of
temporal patterns of TSP and
some measure of emission patterns
Wind direction data may not
be representative
Wind direction data may not
be representative
Same as above
Same as above
None
None
Same as above
Same as above
177
-------
0 A three year old Air Quality Display Model (AQDM) run
predicts concentrations fairly well with a correlation
coefficient of +0.65. Based upon a four-site network,
Figure 1-1 indicates that this represents a true cor-
relation of -0.65 to +0.96.
0 Since the above results are not conclusive, the effect
of meteorological conditions are investigated. TSP
concentrations at the downtown site are found to vary
with days since precipitation. Wind speed is not shown
to have a significant effect.
The cumulative impact of these techniques is to suggest that
either fugitive dust or the wood products point sources are
dominant causes of TSP violations but that it is not clear
which. The initiation of a special study appears warranted.
8.3.2 Large City/Complex Aerosol
The second example is a Southeastern United States bistate
metropolitan area of approximately 700,000 people, located along
a major river valley characterized by gently rolling hills.
Precipitation is somewhat greater than that for the first city,
resulting in an annual average rainfall of 43 in. Most roads in
the city are paved, but there is a large aggregation of chemical
and power generation industrial facilities in the area. Over-
all, there are more than 100 point sources in the area which emit
25 ton/year or more. Growth in population is expected to average
^2 percent per year.- and a large industrial park is planned for
an area to the southwest of the city.
There are three governmental agencies responsible for air
pollution control activities in the metropolitan area: a 37-man
local agency, a 91-man state agency with heavy responsibilities
in other portions of its state, and a 104-man state agency.
Given their sizes, these agencies are staffed with a broad range
/
of skills and expertise. All three agencies have ready access to
computers required for sophisticated analyses. The data base for
the area is considerably greater than that indicated for the
first example city, consisting primarily of the following:
178
-------
0 Eighteen high volume sampler sites, most of which have
more than five years of standard TSP data
0 Meteorological data for at least three sites
0 An emission inventory which has been updated annually
0 Particle sizing data from a number of sites covering
periods of six weeks each; monthly composites for SO ,
NC>3, and 13 metals at most sites, available from 1971
on; microscopy capabilities are also readily available
Using the same process described in subsection 8.3.1, the
best techniques reasonably available for use in this city can be
identified. The only exception from the process described in
subsection 8.3.1 is that manpower constraints are not assigned
due to the relatively large staffs available.
These reasonably available techniques are identified in
Figure 8-2. Twenty-two techniques survived this screening
process. Table 8-5 suggests how these techniques could be
applied to the example city and suggests some qualifications that
would probably be applicable to the specific situation described
for this example city.
Creating a hypothetical situation, let it be assumed that
the following conditions are found after applying the techniques
identified in Table 8-5:
Classifying sites by environment does not produce
meaningful results.
Of the five sites which are exceeding the primary
annual NAAQS, two in the industrial area correlate very
well together (based upon >50 samples). Correlations
among the other sites are <+0.50.
An AQDM run for the current year predicts TSP concen-
trations well and correlates with observed concentra-
tions at +0.72, based upon 18 sites. This implies a
true r of +0.35 to +0.87 and is considered to be good.
A steady decrease in annual concentrations is shown at
all sites until 1976, at which time a slight increase
is observed.
179
-------
CD
D
o
Identify
techniques
desirable
to solve
problem
Exclude
techniques
that are Exclude Exclude cost
resource- dependent ineffective
excessive techniques techniques
Select
remaining
techniques
o -l
0 0
2")
. J
2jl
2r
. 1
30
3T I.
. J
3,1
4T
. 1
40
4~>
. J
4 A
4r-
. _)
5T
. 1
50
. <.
. J
60
. ^
6-> .
. J '
6 A
. 4
6r
. ->
6->
. 1
. J.
. /
^
^~-
^ —
^
^
^
Figure 8-2. Screening techniques for example city No. 2
180
-------
Table 8-5. APPLYING TECHNIQUES TO EXAMPLE CITY NO. 2
Task
Applicable
technique
Comments
Assess regionality 3.1
3.2
5.3
Determine direction
of past data
Determine whether a
specific source is
clearly culpable
Determine whether a
source category is
culpable
2.1
2.5
3.3
3.4
5.3
6.4
6.5
6.6
7.2
2.5
3.3
4.3
4.4
5.2
Investigate meteoro- 4.3
logical effects 4.4
May be useful due to river valley
conditions
None
Relatively large number of sampling
sites will permit tight calibration
assessment
Large time period will permit
good assessment
Dependent upon analysis of temporal
TSP patterns and some measure of
emission patterns
Wind direction data may not be
representative
Wind direction data may not be
representative
Same as above
Limited number of samples increase
confidence limits
Broad confidence limits; monthly
composite data may create dif-
ficulties
Same as for 6.5 above
Limited number of samples increase
confidence limits
Same as above
Same as above
None
None
None
Same as above
Same as above
181
-------
0 Pollution roses for the two industrial area sites point
in the general direction of those industrial facilities,
but no other significant relationships are seen.
0 One of the violating sites, in a semiurban, largely
unpaved area, shows higher concentrations than sur-
rounding upwind sites. Other relationships are not
significant.
0 Another of the violating sites, located downtown, shows
a strong TSP versus daily traffic correlation (+0.87).
No other significant patterns are revealed.
0 Diffusion modeling indicates that area sources are the
largest contributor to violating sites, but that the
point source impact is greatest in the industrial area.
0 Certain species are shown to co-vary strongly with TSP
at various violating sites, but there is a strong
probability that the correlation is still zero due to
the small number of samples. The downtown site varies
with Pb, the industrial sites with V, and a fourth site
(located near a cement factory) varies strongly with
Ca. No other sites show high correlations.
0 Limited particle size/chemical data reveal no meaning-
ful relationships.
Based upon these results, the analyst concludes that indus-
trial sources probably cause observed violations at two sites.
He feels that more detailed analyses are required for the remain-
ing three violating sites. Subsequent analyses show that:
0 Concentrations at the downtown and suburban site co-
vary strongly with days since rainfall, but not with
wind speed. The relationship is much weaker at the
site near the cement factory.
0 Microinventories of those three sites show that there
are large amounts of fugitive dust emitted near the
downtown and suburban sites, but no such results are
shown for the cement plant site.
Based upon these results, the analyst concludes that fugi-
tive dust sources are most probably causing violations at two of
the sites. The third may be caused by emissions from the cement
plant but any such conclusion would be based upon inadequate
data. A special study of that site is recommended.
182
-------
9.0 REFERENCES
1. Nelson, Jr., A. C. and F. Smith. Guidelines for Development
of a Quality Assurance Program: Reference Method for the
Determination of Suspended Particulates in the Atmosphere
(High Volume Method). Research Triangle Institute, Research
Triangle Park, North Carolina. Publication Number EPA-R4-
73-0286. June 1973. 115 p.
2. Guidance for Air Quality Monitoring Network Design and
Instrument Siting (Revised). U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina. Publication
Number OAQPS 1.2-012. September 1975.
3. Hicks, C. R. Fundamental Concepts in the Design of Experi-
ments (Second Edition). New York, New York, Holt, Rinehart
and Winston, 1973.
4. Crow, E. L., F. A. Davis, and M. W. Maxfield. Statistics
Manual. New York, New York, Dover Publications, Inc., 1960.
5. Spiegel, M. R. Schaum's Outline of Theory and Problems of
Statistics. New York, New York, McGraw-Hill Book Company,
1961.
6. Conover, W. J. Practical Non-Parametric Statistics. New
York, New York, John Wiley and Sons, Inc., 1971.
7. Torrie, J. H. and R. G. Steel. Principles and Procedures of
Statistics. New York, New York, McGraw Hill Publishing
Company, Inc., 1960.
8. Spertas, R. and H. J. Levin. Patterns and Trends in Levels
of Suspended Particulate Matter. J. Air Poll. Control
Assoc. 2^:329-333, June 1971.
9. Guidelines for the Evaluation of Air Quality Trends. U.S.
Environmental Protection Agency, Research Triangle Park,
North Carolina. Publication Number OAQPS 1.2-014. February
1974.
10. 1975 Air Quality Data Summary. Puget Sound Air Pollution
Control Agency, Seattle, Washington. 1976.
183
-------
11. Philadelphia Particulate Study (Draft). GCA Corporation.
Prepared for U.S. Environmental Protection Agency, Phila-
delphia, Pennsylvania. February 1978.
12. National Assessment of the Urban Particulate Problem.
Volume VI: Philadelphia. U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina. Publication
Number EPA-450/3-76-026d. July 1976.
13. Analysis of Probable Particulate Non-attainment in the
Kansas City AQCR. PEDCo-Environmental Specialists, Inc.
Prepared for U.S. Environmental Protection Agency, Kansas
City, Missouri. February 1976.
14. Kneip, J., et al. Studies of Trace Substances in an Urban
Atmosphere. New York University Medical Center, New York,
New York. Unpublished.
15. Development of Emission Factors for Fugitive Dust Sources.
U.S. Environmental Protection Agency, Research Triangle
Park, North Carolina. Publication Number EPA-450/3-74-037.
June 1974.
16. An Implementation Plan for Suspended Particulate Matter in
the Phoenix Area. Volume I: Air Quality Analysis. U.S.
Environmental Protection Agency, Research Triangle Park,
North Carolina. Publication Number EPA-450/3-77-0212.
November 1977.
17. Draft Final Report on Suspended Particulate Study. Volume
VI: Particulate Resuspension Study. Prepared by the Inter-
state Sanitation Commission pursuant to Federal Demonstra-
tion Grant Number 5802496-01-0, New York, New York. Unpub-
lished.
18. A Study of the Suspended Particulate Problem in the Duwamish
Basin. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. Publication Number EPA-68-
02-1499. May 1975.
19. Leaderer, B. P., et al. Summary of the New York Summer
Aerosol Study (NYSAS). J. Air Poll. Control Assoc. 28:321-
327, April 1978.
20. Control of Reentrained Dust from Paved Streets. U.S. Envi-
ronmental Protection Agency, Kansas City, Missouri. Pub-
lication Number EPA 907/9-77-007. August 1977.
21. National Assessment of the Urban Particulate Problem.
Volume I: National Assessment. U.S. Environmental Protec-
tion Agency, Research Triangle Park, North Carolina. Pub-
lication Number EPA-450/3-76-024. July 1976.
184
-------
22. Attainment Status of the Total Suspended Particulate Air
Quality Standards in the Houston/Galveston and Dallas/Fort
Worth Areas. Texas Air Control Board, Austin, Texas. May
1976.
23. Warner, P- O., J. O. Jackson, S. Schaldenbrand, and B.
Raphael. Effects of Street Salting on Ambient Air Moni-
toring of Particulate Pollutants in Detriot. (Presented at
Sixth Central Regional Meeting, American Chemical Society-
Detroit, Michigan. April 21-24, 1974.) 15 p.
24, Technical Support Document on the Phase Down of Oregon Open
Field Burning. U.S. Environmental Protection Agency,
Region X, Seattle, Washington. March 1977.
25. Cahill, T. A. Results of Highway Particulate Investigations
in Califormia. Department of Physics and the Crocker Nu-
clear Laboratory, University of California, Davis, Cali-
fornia. Unpublished.
26. Lioy, P. J., et al. Characterization of Aerosols Upwind of
New York City: I. Aerosol Composition. (Presented at
American Industrial Hygiene Association Conference. New
Orleans, Louisiana. May 22-27, 1977.)
27. Memo for J. F. Anderson to I. A. Hoekstra. Subject: High
Sulfur Dioxide, Lackawanna STP, First Quarter 1975. County
of Erie, New York, Department of Environmental Quality.
April 30, 1975.
28. Guideline for Development of Control Strategies in Areas
with Fugitive Dust Problems. U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina. Publication
Number EPA-450/2-77-029. October 1977.
29. Ohlendorf, M. W. and S. P. Pitt. Air Pollution in Texas:
Approaches to the Analysis and Modeling of Aerometric Data.
Prepared for Texas Air Control Board, Austin, Texas. October
1975.
30. Menczel, J. H., et al. Monitoring for Compliance with
Particulate Matter Standards. (Presented at 70th Annual
Meeting of the Air Pollution Control Association. Toronto,
Ontario, Canada. June 20-24, 1977.)
31. Technical Manual for Measurement of Fugitive Emissions:
Upwind/Downwind Sampling Method for Industrial Pollutants.
U.S. Environmental Protection Agency, Research Triangle
Park, North Carolina. Publication Number EPA-600/2-76-0892.
April 1976.
185
-------
32. Development of Emission Factors for Fugitive Dust Sources.
U.S. Environmental Protection Agency, Research Triangle
Park, North Carolina. Publication Number EPA-450/3-74-037.
June 1974.
33. Bruckman, L. Suspended Particulate Transport. (Presented
at 70th Annual Meeting of Air Pollution Control Association.
Toronto, Ontario, Canada. June 20-24, 1977.)
34. Identification and Analysis of Urban Air Quality Patterns.
University Analytical Center, University of Arizona. Pre-
pared for Electric Power Research Institute, Palo Alto,
California. December 1977.
35. Findley, C. E. and D. C. Bray. Attainment of Ambient Partic-
ulate Matter Standards in Idaho. (Presented at the 1973
Air Pollution Control Association Meeting. Seattle, Wash-
ington. November 1973.)
36. Sonquist, J. A., E. L. Baker, and J. N. Morgan. Searching
for Structure. Institute for Social Research, University of
Michigan, Ann Arbor, Michigan. 1973.
37. Trijonis, J., Y. Horie, and D. Bicker. Statistical Analysis
of TSP and Meteorological Data in EPA Region VI. Technology
Service Corporation. Prepared for U.S. Environmental
Protection Agency, Dallas, Texas. May 1978.
38. Wolff, G. T., et al. Characterization of Aerosols Upwind of
New York City. Volume II: Transport. (Presented at
American Industrial Hygiene Association Conference. New
Orleans, Louisiana. May 22-27, 1977.)
39. Samson, P- J., G. Neighmond, and A. J. Yencha. The Trans-
port of Suspended Particulate as a Function of Wind Direc-
tion and Atmospheric Conditions. J. Air Poll. Control
Assoc. 2j[:1232-1237, December 1975.
40. Long-Range Transport of Particulates: Case Study, October
15, 1976. (Presented at 70th Annual Meeting of the Air
Pollution Control Association. Toronto, Ontario, Canada.
June 20-24, 1977O
41 Guide for Compiling a Comprehensive Emission Inventory
(Second Edition). U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina. Publication Number
APTD-1135. December 1974.
42. Guidelines for Air Quality Maintenance Planning and Anal-
ysis. Volume 7: Projecting County Emissions (Second Edition)
U.S. Environmental Protection Agency, Research Triangle
Park, North Carolina. Publication Number EPA-450/4-74-008.
January 1975.
186
-------
43. Guidelines for Air Quality Maintenance Planning and Analysis.
Volume 13: Allocating Projected Emissions to Sub-County
Areas. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. Publication Number EPA-450/
4-74-014. November 1974.
44. Air Quality Analysis Workshop. Volume I: Manual. U.S.
Environmental Protection Agency, Research Triangle Park,
North Carolina. Publication Number EPA-450/3-75-0802.
November 1975.
45. Pace, T., K. Axetell, and R. Zimmer. Microinventories for
TSP (Draft). U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. April 1978.
46. TSP Source Inventory around Monitoring Sites in Selected
Urban Areas, Kansas City (Draft). PEDCo Environmental, Inc.
Prepared for U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. March 1978.
47. Fugitive Dust in Kansas and Nebraska. PEDCo-Environmental
Specialists, Inc., Cincinnati, Ohio. Prepared for U.S.
Environmental Protection Agency, Kansas City, Missouri.
February 1974.
48. Characterization of Particulate Sources influencing Monitor-
ing Sites in Region VIII Non-Attainment Areas. U.S. Envi-
ronmental Protection Agency, Denver, Colorado. Publication
Number EPA 908 1-76-007. June 1976.
49. TSP Source Inventory around Monitoring Sites in Selected
Urban Areas, Portland (Draft). PEDCo Environmental, Inc.
Prepared for U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. April 1978.
50. Pace, T. G. An Approach for the Preliminary Assessment of
TSP Concentrations. U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina. Publication Number
EPA 450/2-78-016. May 1978.
51. Interim Guideline on Air Quality Models. U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina.
Publication Number OAQPS 1.2-080. October 1977.
52. Guidelines for Air Quality Maintenance Planning and Anal-
ysis. Volume 10: Procedures for Evaluating Air Quality
Impact of New Stationary Sources (Revised). U.S. Environ-
mental Protection Agency, Research Triangle Park, North
Carolina. Publication Number EPA-450/4-77-001. October
1977.
187
-------
53. Guidelines for Air Quality Maintenance Planning and Anal-
ysis. Volume 12: Applying Atmospheric Simulation Models to
Air Quality Maintenance Areas. U.S. Environmental Protec-
tion Agency, Research Triangle Park, North Carolina. Pub-
lication Number EPA-450/4-74-013. September 1974.
54. Henry, R. C. A Factor Model of Urban Aerosol Pollution: A
New Method of Source Identification. A dissertation sub-
mitted to the faculty of the Oregon Graduate Center in
partial fulfillment of the requirements for the degree
Doctor of Philosophy in Applied Physics. 1977.
55. Lead Analysis for Kansas City and Cincinnati. PEDCo Envi-
ronmental, Inc. Prepared for U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina. June 1977.
56. Gatz, D. F. St. Louis Air Pollution: Estimates of Aerosol
Source Coefficients and Elemental Emission Rates. (Symposium
on Atmospheric Diffusion and Air Pollution. Santa Barbara,
California. September 9-13, 1974.) p. 109-114.
57. Gatz, D. F. Relative Contributions of Different Sources of
Urban Aerosols: Application of a New Estimation Method to
Multiple Sites in Chicago. Atm. Environ. 8^:1-18. Sep-
tember 1974.
58. Hardy, K. A Characterization of the Atmospheric Aerosol in
Miami, Florida. Department of Physical Sciences, Florida
International University. Unpublished.
59. Aerosol Sampling and Analysis: Phoenix, Arizona. U.S.
Environmental Protection Agency, Research Triangle Park,
North Carolina. Publication Number EPA 600/3-77-015.
February 1977.
60. Rahn, K. A. The Chemcial Composition of the Atmospheric
Aerosol. University of Rhode Island, Kingston, Rhode Island.
July 1976. 265 p.
61. Fordyce, J. S., R. B. King, and H. E. Neustadter. Elemental
Composition of Airborne Particulates and Source Identifica-
tion: Data Analysis Techniques. J. Air Poll. Control
Assoc. 26^:1079-1084, November 1976.
62. Martens, C. S., et al. Sources of Vanadium in Puerto Rican
and San Francisco Bay Area Aerosols. Environ. Sci. & Techn.
7^:817-820, September 1973.
63. Paciga, J. J. and R. E. Jervis. Multielement Size Character-
ization of Urban Aerosols. Environ. Sci. & Techn. 10(12);
1124-1128, November 1976.
188
-------
64. Shum, Y. Atmospheric Trace Elements and their Application
in Tracing Air Pollution. A dissertation submitted to the
faculty of Oregon State University in partial fulfillment of
the degree of Doctor of Philosophy. June 1974.
65. Friedlander, S. K. Chemical Element Balances and Identifica-
tion of Air Pollution Sources. W. M. Keck Engineering
Laboratories, California Institute of Technology, Pasadena,
California. 1973. p. 235-240.
66. Friedlander, S. K., G. M. Hidy, and M. S. Miller. A Chemical
Element Balance for the Pasadena Aerosol. Journal of Colloid
and Interface Science. 3£(1):165-176, April 1972.
67. Heisler, S. L. Source Breakdown of the St. Louis Aerosol by
the Method of Chemical and Material Balances. California
Institute of Technology, Pasadena, California. November
1975. 16 p.
68. Investigation of Ozone and Ozone Precursor Concentrations at
Nonurban Locations in the Eastern United States. Phase II:
Meteorological Analysis. U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina. Publication
Number EPA-450/3-74-0342. February 1975.
69. Noble, B. Applied Linear Algebra. Englewood Cliffs, New
Jersey, Prentice-Hall, Inc., 1969.
70. Friedlander, S. K., et al. The Relationship of Smog Aerosol
Size and Chemical Element Distributions to Source Charac-
teristics. Atm. Environ. 7^:633-649, February 1973.
71. Henry, R. C. The Application of Factor Analysis to Urban
Aerosol Source Identification. Environmental Research and
Technology, Inc., Westlake Village, California. Unpublished.
p. 134-138.
72. Majewski, R. and N. Kolak. Fugitive Dust and Proportional
Contribution of Sources of Suspended Particulates in the
Niagara Frontier Air Quality Control Region. New York State
Department of Environmental Conservation, Albany/ New York.
September 1976.
73. Blifford, I. H. and G. O. Meeker. A Factor Analysis Model
of Large-Scale Pollution. Atm. Environ. 1:147-157, January
_1967.
74. John, W., et al. Trace Element Concentrations in Aerosols
from the San Francisco Bay Area. Atm. Environ. 7:107-118.
June 1972. ~
189
-------
75. Ohlendorf, M. W. and S. P. Pitt. Empirical Analyses toward
Total Suspended Particulate Source Classification in Texas.
Texas Air Control Board—University of Texas at Austin.
Publication Number ARL-TR-76-48. October 1976. 48 p.
76. Everitt, B. Cluster Analysis. London, Great Britain,
Heinemann Eductional Books Ltd., 1974.
77. Gaarenstroom, P. D., et al. Pattern Recognition and Factor
Analysis Applied to Characterization of Particulate Com-
position from Southwestern Desert Atmosphere. Purdue Uni-
versity, West Lafayette, Indiana. Publication Number AD-
A023 235. March 1976. 26 p.
78. Gladney, E. S., et al. The Use of Multivariate Analysis to
Identify Sources of Selected Elements in Boston Urban
Aerosol. Atm. Environ. 10^:1015-1025, May 1976.
79. Harmon, H. H. Modern Factor Analysis (Second Edition).
Chicago, Illinois, University of Chicago Press, 1967.
80. Gatz, D. F. Identification of Aerosol Sources in the St.
Louis Area Using Factor Analysis. J. Appl. Meteor. 17:600-
608, May 1978.
81. National Assessment of the Urban Particulate Problem.
Volume II: Particle Characterization. U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina.
Publication Number EPA-450/3-76-025. July 1976.
82. Gise, J. P. Particulate Characterization and Source Iden-
tification Using Optical Microscopy and X-ray Fluorescence
Spectroscopy. Texas Air Control Board. (Presented at 70th
Annual Meeting of the Air Pollution Control Association.
Toronto, Ontario, Canada. June 20-24, 1977.)
83. Denver Air Pollution Study—1973. Volume I: Proceedings of
a Symposium. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. Publication Number EPA-
600/9-76-007a. June 1976.
84. Bradway, R. M. and F. A. Record. Application of Polarizing
Microscopy to the Characterization of Ambient Suspended
Particulates. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. Unpublished.
85. Nevada Particulate Control Study for Air Quality Maintenance
Areas. PEDCo Environmental, Inc., Cincinnati, Ohio. March
1977.
190
-------
86. Bray, D. C. Analysis of Airborne Particulate Collected on
Glass Fiber Filters. U.S. Environmental Protection Agency,
Seattle, Washington. November 1977.
87. Barton, S. C., H. G. McAdie, and J. A. McCoubrey. The Char-
acterization of Suspended Participates in an Industrial
Area. (Presented at 70th Annual Meeting of Air Pollution
Control Association. Toronto, Ontario. Canada. June 20-
June 24, 1974.) p. 1-13.
88. Draft Final Report on Suspended Particulates Study. Volume
VII: Particle Analysis Task Force. Prepared by the Inter-
state Sanitation Commission pursuant to Federal Demonstra-
tion Grant Number 5802496-01-0, New York, New York. Unpub-
lished.
89. Draftz, R. G. Microscopical Analysis of Aerosols Collected
in Miami, Florida. IIT Research Institute, Chicago, Illinois
May 1977.
90. Warner, P. O., et al. Identification and Quantitative Anal-
ysis of Particulate Air Contaminants by X-ray Diffraction
Spectrometry. (Presented at 64th Annual Meeting of the Air
Pollution Control Association. Atlantic City, New Jersey.
June 27-July 1, 1971.)
91. Cramer, H. E., H. V. Geary, S. G. Saterlie, and J. F. Bovers.
Assessment and Updating of Particulate Emissions Data for
the Southwest Pennsylvania Intrastate Air Quality Control
Region. H. E. Cramer Company, Inc., Salt Lake City, Utah.
December 1977.
92. Draftz, R. G. Types and Sources of Suspended Particles in
Chicago. IIT Research Institute. Prepared for City of
Chicago, Department of Environmental Control, Chicago,
Illinois. May 1975.
93. Mukherji, S., et al. Rural Fugitive Dust Impact on an Urban
Area. Indiana Air Pollution Control Division, Indianapolis,
Indiana. Unpublished.
94. Discussion of the Advantages and Limitations of Filter
Analysis by Microscopy as a TSP Analysis Technique. U.S.
Environmental Protection Agency, Research Triangle Park,
North Carolina. Unpublished.
95. Fochtman, E. G. and J. D. Stockham (ed.). Particle Size
Analysis. Ann Arbor, Michigan, Ann Arbor Science Publishers,
Inc., 1977. 140 p.
96. Heskelth, H. E. Fine Particles in Gaseous Media. Ann Arbor,
Michigan, Ann Arbor Science Publishers, Inc., 1977. 214 p.
191
-------
97. Willeke, K. and K. T. Whitby. Atmospheric Aerosols: Size
Distribution Interpretation. J. Air Poll. Control Assoc.
25^:529-534, May 1975.
98. Harris, Jr., F. S. Atmospheric Aerosols: A Literature
Summary of their Physical Characteristics and Chemical
Composition. Old Dominion University—NASA. Publication
Number CR-2626. January 1976. 44 p.
99. Gilmore, T. M. and T. R. Hanna. Applicability of the Mass
Concentration Standards for Particulate Matter in Alaskan
Areas. J. Air Poll. Control Assoc. 25^535-539, May 1975.
100. Hidy, G. M., et al. Characterization of Aerosols in Cali-
fornia (ACHEX). Volume IV: Analysis and Interpretation of
Data. Rockwell International, Newbury Park, California.
September 1974.
101. Hardy, K. A., et al. Elemental Constituents of Miami
Aerosol as Function of Particle Size. Environ. Sci. &
Techn. 1£: 176-182, February 1976.
102. Baum, E. J. Aerosol and Particulates Evaluated in Portland,
Oregon. Department of Environmental Technology, Beaverton,
Oregon. Unpublished.
[The following three references were obtained in September 1978,
too late for inclusion in the main body of the report.]
103. Gordon, G. E., W. H. Zoller, and E. S. Gladney. Abnormally
Enriched Trace Elements in the Atmosphere. Reprint from
Trace Substances in Environmental Health, VII. A Symposium.
D. D. Hemphill, Editor. University of Missouri, Columbia,
Missouri. 1974.
104. Zoller, W. H., E. S. Gladney, G. E. Gordon, and J. J. Bors.
Emissions of Trace Elements From Coal Fired Power Plants.
Reprinted from Trace Substances in Environmental Health,
VIII. A Symposium. D. D. Hemphill, Editor. University of
Missouri, Columbia, Missouri. 1974.
105. Kowalczyk, G. S., C. E. Choquette, and G. E. Gordon. Chem-
ical Element Balances and Identification of Air Pollution
Sources in Washington, D.C. Atm. Environ. 12:1143-1153.
1978.
192
-------
APPENDIX A
TECHNIQUES
FOR
ELEMENTAL ANALYSIS
193
-------
Table A-l. TECHNIQUES FOR ELEMENTAL ANALYSIS*
Element
H
He
Li
Be
B
C
N
O
F
Ne
Na
Mg
Al
Si
P
S
Cl
Ar
K
Ca
Sc
Ti
V
Cr
Mn
Fe
Co
Ni
Cu
Zn
Ga
Ge
As
Se
Br
Kr
Kb
Sr
Y
Zr
Nb
Mo
Tc
Ru
Rh
Pd
Ag
Cd
In
Sn
Sb
Te
I
NA
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
PA
X
X
X
X
X
X
X
X
X
X
X
X
X
X
XRF
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
XRFS
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
XPA
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
PS
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
AA
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
ES
X
X
X
X
X
X
X
X
X
X
X
X
194
-------
Table A-l (Continued). TECHNIQUES FOR ELEMENTAL ANALYSIS3
Element NA PA XRF XRFS XPA PS AA ES
Xe
Cs
Ba
La
Ce
Pr
Nd
Pro
Sm
Eu
Gd
Tb
Dy
Ho
Er
Tm
Yb
Lu
Hf-
Ta
W
Re
Os
Ir
Pt
Au
Hg
Tl
Pb
Bi
Po
At
Rn
Fr
Ra
Ac
Th
Pa
U
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X X
X X XX
X X
X
X
X
X
X X
X
X
X
X
XXX X
XXX XX
X
X
Abbreviations:
NA = Neutron activation
PA = Proton activation
XRF = X-ray fluorescence
XRFS = X-ray radioactive source
XPA = X-ray photon activation
PS = Proton scattering
AA = Atomic absorption
ES = Emission spectroscopy
Source: Reference 22
195
-------
11.0 INDEX
Air Quality Display Model, 93
Analysis of variance, 4
Cascade impactors, 154,157
Chemical element balance, 114
Chi-square analysis, 8
Climatological Dispersion Model, 93
Cluster analysis, 132
Communality, 138
Correlation coefficients, 34
Correlation matrix, 38,49,51-53,127,
141
CRSTER Model, 93
Days since rain, 33,50,57,61
Decision tree analysis, 54
Dendrogram, 133
Deposition, 66,94
Dichotomous samplers, 154
Diffusion batteries, 154
Diffusion modeling, 87
Directionally-actuated samplers, 46,
48
Diurnal variation, 6,20,22-24
Dosage rose, 38,102
Electrical analyzers, 154
Electron microscopy, 146
Element rose, 105
Elemental fractionation, 117
Emission inventory, 72
Emission pattern, 6,22,25
Enrichment factor, 108
Factor analysis, 136
F-distribution, 8,51
Fugitive dust, 14,16,19,31,33,52,
61-62,77,80,104,158
Gradient rose, 40,102
Intersite TSP correlation, 34
Interspecies correlation, 127
Inversion height, 50
Land use classification, 83
Linear regression, 4,23,26,107
Long-range transport, 45,69,94,
112,129
Long-term trends, 6,9-12
Meteorologically-affected patterns,
102
Microinventory, 77
Microscopy, 22,146,154
Mixing height, 33
Moving average, 7,11
Multiple regression, 4
Neutron activation, 121,150,162
Nonlinear mapping, 133
Optical counters, 154
Parametric correlation, 8
Particle size distribution, 154
Particle sizing, 149
Pattern recognition, 132
Pollution roses, 15,19,26,36,52,66,
76,102
Precipitation, 58
Principal components model, 137
Principal factor method, 137
Rainfall, 61
Rainfall frequency, 16
RAM Model, 94
Reentrainment, 23,26
Regression analysis, 8,23,26
Residual sum of squares (RSS), 54
Seasonal and monthly patterns, 6,13
15-16
Secondary particulates, 45,69,75,
120,155,157
Sequence for applying techniques,
170
Soil group, 158
Source emission patterns, 62
Spatial patterns, 102
Spearman rank correlation, 7
Squared multiple correlation, 138
Stepwise multiple regression, 4,
51-52
Temporal patterns, 102
Texas Climatological Model, 93
Texas Episodic Model (TEM), 94
Tracer, 15,117
Traffic patterns, 18-19,21,24,27,29
Trajectory analysis, 68,103
Trajectory vectors, 68
Unconventional sources of particu-
lates (fugitive dust), 73
Upwind/downwind, 45,62,76,103,107
Visibility, 50,53,107,129,157
Weekday/weekend and daily patterns,
6,18,20-21,29,62
Whittaker-Henderson formula, 7,9,13
Wind persistance, 65
Wind speed, 63
X-ray diffraction, 22,150
X-ray fluorescence, 105,107,150-151
196
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA-450/3-78-113
2.
4TlTGESAfD0SFUAMBLIEENT PARTICULATE ANALYSIS AND ASSESSMENT
METHODS
J
T.AUTHORIS) jarnes y\< Throgmorton and Kenneth Axetell
9. PERFORMING ORGANIZATION NAME A IV
PEDCo Environmental, Inc.
2420 Pershing Road
Kansas City, Missouri 641 C
12. SPONSORING AGENCY NAME AND ADC
U.S. ENVIRONMENTAL PROTECT
Office of Air, Noise, and
Office of Air Quality Plar
Research Triangle Park, NC
D ADDRESS
)8
RESS
ION AGENCY
Radiation
ining and Standards
) 27711
3. RECIPIENT'S ACCESSION NO.
5. REPORT DATE
September 1
978
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO.
10. PROGRAM ELEMENT NO.
2AA635
11. CONTRACT/GRANT NO.
68-02-2603
13. TYPE OF REPORT AND PERIOD COVERED
Final Report
14. SPONSORING AGENCY CODE
200/4
15. SUPPLEMENTARY NOTES
16. ABSTRACT
A compendium of techniques is provided which describes approximately 25
techniques for analysis and interpretation of ambient particulate data. The tech-
niques can be grouped categorically as follows: Temporal Patterns; Spatial Patterns;
Meteorological Effects; Emissions Assessment; Interpretinq Chemical, Elemental
and Morphological Data; and Interpreting Particle Size Data. The techniques are
described briefly, and references for a more thorough treatise of the subject are
provided. The techniques span a range of complexity, cost and effectiveness.
They are evaluated in terms of cost effectiveness and resource requirements.
The digest provides the user with guidelines on the use of the techniques in terms
of their problem applicability, resource requirements and interrelationships.
It provides a framework for designing studies or analyses to interpret data from
ambient particulate or any similar pollutant.
17.
a. DESCRIPTORS
Particulate Matter
Fugitive Dust
Monitor Siting
Data Analysis
Field Study
18. DISTRIBUTION STATEMENT
Unlimited
KEY WORDS AND DOCUMENT ANALYSIS
b.lDENTIFIERS/OPEN ENDED TERMS
19. SECURITY CLASS (This Report)'
Unclassified
20. SECURITY CLASS (This page)
Unclassified
e. COSATI Field/Group
21. NO. OF PAGES
206
22. PRICE
EPA Form 2220-1 (Rev. 4-77) PREVIOUS EDITION is OBSOLETE
197
------- |