Tennessee
Valley
Authority
United States
Environmental Protection
Agency
Research and Development
Office of Natural
Resources
Chattanooga TN 37401
TVA ONR-79/03
Office of Energy, Minerals, and
Industry
Washington DC 20460
EPA-600 7-79-084
March 1979
The Analysis of
Suspended
Particulates and
Sulfates
A Way to Begin
Interagency
Energy/Environment
R&D Program
Report
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are:
1. Environmental Health Effects Research
2. Environmental Protection Technology
3. Ecological Research
4. Environmental Monitoring
5. Socioeconomic Environmental Studies
6. Scientific and Technical Assessment Reports (STAR)
7. Interagency Energy-Environment Research and Development
8. "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the INTERAGENCY ENERGY-ENVIRONMENT
RESEARCH AND DEVELOPMENT series. Reports in this series result from the
effort funded under the 17-agency Federal Energy/Environment Research and
Development Program. These studies relate to EPA's mission to protect the public
health and welfare from adverse effects of pollutants associated with energy sys-
tems. The goal of the Program is to assure the rapid development of domestic
energy supplies in an environmentally-compatible manner by providing the nec-
essary environmental data and control technology. Investigations include analy-
ses of the transport of energy-related pollutants and their health and ecological
effects; assessments of, and development of, control technologies for energy
systems; and integrated assessments of a wide range of energy-related environ-
mental issues.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/7-79-084
TVA/ONR-79/03
THE ANALYSIS OF SUSPENDED PARTICULATES AND SULFATES:
A WAY TO BEGIN
by
Walter Liggett and William Parkhurst
Office of Natural Resources
Tennessee Valley Authority
Chattanooga, Tennessee 37401
Interagency Agreement No. EPA-IAG-D5-E721
Project No. 80 BDM
Program Element No. INE-625B
Project Officer
James T. Stemmle
401 M Street RD-681
Washington, D.C. 20460
Prepared for
OFFICE OF ENERGY, MINERALS, AND INDUSTRY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
-------
DISCLAIMER
This report was prepared by the Tennessee Valley Authority and has
been reviewed by the Office of Energy, Minerals, and Industry, U.S.
Environmental Protection Agency, and approved for publication. Approval
does not signify that the contents necessarily reflect the views and
policies of the Tennessee Valley Authority or the U.S. Environmental
Protection Agency, nor does mention of trade names or commercial products
constitute endorsement or recommendation for use.
11
-------
ABSTRACT
Total suspended particulate (TSP) and suspended sulfate (SS) levels
have been sampled since November 1973 at five isolated sites across the
Tennessee Valley. A method for beginning to analyze such data is demon-
strated. This beginning is intended to lead finally to information on
pollution sources, an objective that may require modeling meteorological
influences and resolving sources. Analysis with this objective,
which can be very complex, is effectively begun by using the method demon-
strated in this paper. Applied to the TSP and SS data, this method suggests
agricultural contributions to TSP levels, distant-source contributions
to SS levels, and various influences of the meteorology. This method
also shows deficiencies in the data collection that prevent the building
of better, more quantitative models. One deficiency in this data set is
the sixth-day sampling, which is not frequent enough to allow monthly
variations in pollution levels to be distinguished from more rapid
variations. Thus, data analysis would be more effective if the sampling
frequency were increased and, further, if particle size and chemical
composition were better resolved.
This report was submitted by the Tennessee Valley Authority,
Office of Natural Resources, in partial fulfillment of Energy
Accomplishment Plan 80 BDM under terms of Interagency Agreement EPA-
IAG-D5-E721 with the Environmental Protection Agency. Work was com-
pleted as of January 12, 1979.
111
-------
CONTENTS
Abstract iii
List of Figures v
List of Tables v
1. Introduction 1
2. Conclusions and Recommendations 3
3. The Method 4
Overview of the method 4
The TSP and SS components 5
The algorithm 7
4. Interpretation of the Data 14
Seasonal component 14
Valley-wide and local smooths 15
Valley-wide and individual roughs 17
5. Design of Monitoring 20
References 22
-------
LIST OF FIGURES
Number
1
2
3
4
5
Seasonal patterns for TSP and SS
Smoothed levels of total suspended particulates
Smoothed levels of suspended sulfates
Data decomposition showing flow of calculations
Daily sulfate data with sixth-day sampling
smoothed
Page
6
8
9
11
16
LIST OF TABLES
Number
Robust Correlations and Number of Nonmissing
Observations for the Roughs
10
VI
-------
SECTION 1
INTRODUCTION
Since November 1973, the Tennessee Valley Authority (TVA) has
operated high-volume samplers at five sites to obtain background concen-
trations and trends for total suspended particulates (TSP) and water-
soluble suspended sulfates (SS). These sites, which are intended to
represent large subregions of the Tennessee Valley, are remote from
power plants and other large sources of industrial pollution. From east
to west, these sites are in Washington County, Virginia (at Loves Mill);
Monroe County, Tennessee (at Loudon); Jackson County, Alabama (at Hytop);
Giles County, Tennessee; and Trigg County, Kentucky [at Land Between The
Lakes (LBL)]. Samples have been collected for a 24-h period every sixth
day and analyzed by standard methods.1'4
This paper demonstrates a method that helps investigators explain
data like these. The explanations answer questions such as how much
each source contributed to the observed levels, an important question in
the application of the 1977 Clean Air Act Amendments. The method may
suggest explanations with clear implications. However, the method may
be only the first step in developing a more complete model of what
influences the measurements. In this case, the method is intended to
show the potential benefits of a more complete model and the require-
ments for its development so that the considerable expense and expertise
possibly needed can be justified and planned. Important benefits may
not be available from a particular data set because meteorological
influences or something else cannot be adequately modeled. The method
is intended to indicate such a possibility.
Models that explain air quality measurements are needed for many
purposes, for example, to obtain information relevant to control strate-
gies or to interpret the trends that monitoring is meant to detect.6
Such models involve several factors including the sources of pollution
and the transport and transformation of pollutants. Such models are
needed because they differentiate among these factors. Thus, they allow
the effects of control strategies and other changes to be predicted and
the causes of a trend to be understood.
The method demonstrated in this paper decomposes the data into com-
ponents that represent data variations of different temporal and spatial
extents.7'8 A guide to the method is given by the equation,
log-transformed data = seasonal component +
Valley-wide smooth + local smooth + individual rough.
This equation shows that log transforms of the data rather than the
original data are decomposed into components. One component represents
the seasonal (i.e., annually recurring) variation for all sites.
Smooths, which show variations that persist in time, are computed for
all sites (Valley-wide) and for the variations unique to each site
(local). The individual roughs are the irregular variations not
accounted for by the other components. A Valley-wide rough has been
computed, but not incorporated in the decomposition.
-------
-2-
The method is useful because the data are easier to interpret
component by component than all at once. The data are determined by
many factors. The influence of these factors on each component is
easier to understand than their influence on the undivided data. For
example, the influence of seasonal factors can be seen in the seasonal
component, but generally not in the other components. Thus, the method
is much more revealing, yet no more complicated, than the histograms
often used to summarize air quality data.
The TSP and SS data from remote sites on which the method is demon-
strated are interesting because of questions about pollutant origins.
These origins are both distant sources and local, nonindustrial sources
such as agriculture. The questions involve the methods and benefits of
controlling such sources and the interference of such sources with the
monitoring of a specific industrial source.
Decomposing these data into components reveals several features
observed in other regions. One feature is the patterns shown by the
seasonal component. Nationwide, seasonal patterns in TSP are not con-
sistent, indicating the importance of local sources, which differ for
urban and rural monitoring.9 In the east, the seasonal patterns in SS
have a single peak in the summer.9 Another feature is the relations
among the series observed at different sites. For SS data, similarities
in time behavior at widely separated sites have been observed in New
York State.10"12 Some intersite differences observed in the data ana-
lyzed here (unusually high SS levels at LBL) have been explained by
Reisinger and Crawford.13
To describe the method, we discuss the components it produces from
the TSP and SS data before we specify the computational details. This
discussion, which is in Section 3, is thus more data-oriented than the
usual description of a method. In Sections 4 and 5, we interpret the
data presented in Section 3. Section 4 discusses the physical mechanisms
responsible for the observations. Section 5 discusses consequences for
the design of monitoring.
-------
-3-
SECTION 2
CONCLUSIONS AND RECOMMENDATIONS
Many factors other than emission levels influence air quality moni-
toring data; most obviously, the weather influences transport. Further,
many sources other than those usually controlled contribute to pollution
levels. We recommend that, when possible, the influences of these factors
be modeled rather than treated as random.
The influences on pollution levels most obvious in monitoring data
are often of little interest in decision making. When this is true, we
recommend that the data collection and analysis needed to adjust for
these influences be undertaken. For example, adjustment of the data
for meteorological influences should allow emission trends to be detected
more easily.
The effort finally needed to model the influences on the data might
require expertise and data collection, which make monitoring much more
expensive. We recommend that the importance of the information to be
obtained determine the degree of monitoring to be done.
The analysis method demonstrated helps guide future data collection
and analysis by showing what is needed to meet objectives. We recommend
that all monitoring data be subjected to such preliminary analysis.
-------
-4-
SECTION 3
THE METHOD
OVERVIEW OF THE METHOD
Both the TSP and the SS data are composed of time series from each
site. These series each have 258 observations that have been transformed
by:
y = log1Q(x + 1), (1)
where
3
x = an original observation, |Jg/m ,
3
y = the corresponding log-transformed observation, (Jg/rn .
Data transformations are discussed by Tukey.7
The first part of the decomposition computes a smooth trace through
each series. This trace follows the slowly changing variations in the
data, the variations that persist from sample to sample. It is not
affected by the irregular sample-to-sample changes. It represents the
data variations that monthly averages are intended to portray. Sub-
tracting the smooth trace from the series that generated it gives a
component that represents irregular sample-to-sample changes in the
data. Thus, each series is decomposed into two components, a smooth
trace and an irregular component called an individual rough. The smooth
trace represents data fluctuations caused, for example, by seasonal
changes in the weather, and the rough represents fluctuations caused,
for example, by frontal passages.
The smooth traces are computed by the use of running medians.
Consider, for example, a running median that spans five observations.
It is computed by finding the middle value (the third largest value)
of every group of five successive values of a series. An alternative,
a running month-long average, is computed by finding the average of
every group of five successive values. A running median is less sensi-
tive to isolated values that are very large or very small. Thus, running
medians give a smooth trace that is less influenced by such values and,
consequently, a rough that better represents such values. The actual
algorithm for computing the smooth traces, which is described below,
involves repeated computation of running medians, a method for obtaining
the smooth trace at the ends of the series, and an approach to missing
values.
The second part of the decomposition extracts the Valley-wide com-
ponent from the smooth traces for each site. Our choice for the Valley-
wide component is the sample-by-sample average of the five smooth traces.
This choice was made despite one- and two-day differences in sampling
day that occur before May 1976 because smooth traces rather than the
original data are averaged. Subtracting the Valley-wide component from
the smooth traces gives the local smooths.
-------
-5-
The third part of the decomposition extracts the seasonal component
from the Valley-wide component. The seasonal component is computed with-
out the first and last seven values of the Valley-wide component so that
exactly four years of data are used. Since each year has sixty-one
values, the seasonal component has sixty-one values. Each of these values
is the midmean of the corresponding four yearly values. (The midmean of
four values is the average of the second and third largest.) Subtracting
the seasonal component from the Valley-wide component gives the Valley-
wide smooth. The Valley-wide smooth shows unusual years more clearly
because the midmean instead of the average is used to compute the seasonal
component.
THE TSP AND SS COMPONENTS
Further understanding of the method can be gained by considering
the components produced from the TSP and SS data. However, before pre-
senting these components, we present annual and 24-h summaries of these
data to help the reader relate them to other data.
For the calendar years 1974 through 1977, the annual geometric
means of the TSP for these sites ranged from 28 to 43 (Jg/m3. These TSP
levels are well below the primary and secondary National Ambient Air
Quality Standards of 75 and 60 (Jg/m3, respectively. The 24-h TSP levels
found in these data also do not exceed the primary and secondary stan-
dards of 150 and 260 [Jg/m3, respectively. However, the 24-h levels for
February 24, 1977, a day during a severe dust storm, are recorded as
lost records. These levels are actually 88, 767, 699, 654, and 138 (Jg/m3
for Loves Mill, Loudon, Hytop, Giles County, and LBL, respectively, as
shown by TVA laboratory files. This dust storm caused 24-h levels to
exceed standards throughout the Southeast.14
For the same periods and sites, the annual arithmetic means of the
SS ranged from 5.9 to 10.0 |Jg/m3. These levels are within the range
expected in rural areas east of the Mississippi River.15 Some states
have standards for SS, and the EPA is considering national standards.
Suggestions for the annual standard16 lie between 5 and 15 |Jg/m3, and
suggestions for the 24-h standard16 lie between 10 and 25 (Jg/m3. Four-
teen instances of 24-h levels above 25 (Jg/m3 are contained in the data
from these sites.
Consider the components discussed above, starting with the seasonal
component. The TSP and SS seasonal components are the most pronounced
feature of the data. They are shown in Figure 1 after retransformation
to compensate for the log transform. They are plotted on a horizontal
axis that starts on the first day of winter and is divided seasonally.
The estimate of the TSP pattern has peaks in mid-April and mid-July that
reach 54 (Jg/m3- It has levels as low as 22 (Jg/m3. The estimate of the
SS pattern has a peak in mid-July that reaches 13.0 (Jg/m3. It has levels
as low as 3.7 (Jg/m3. The April and July TSP peaks invite comparison because
the SS is a much larger fraction of the July TSP peak than of the April TSP
peak.
-------
-6-
80
NJ
£
03
ct:
in
C_J
z
CD
C_J
0
TOTAL SUSPENDED PARTICIPATES
WINTER SPRING SUMMER FALL
20
O5
D.
s 10
^—
ex
o
o
0
SUSPENDED SULFATES
WINTER SPRING SUMMER FALL
SEASON
Figure 1. Seasonal patterns for TSP and SS.
-------
-7-
The Valley-wide and local smooths in Figure 2 show any annual
trends and persistent local conditions contained in the TSP data. The
Valley-wide smooth shows that 1974 and 1977 are worse than 1975 and
1976, but it does not seem to provide convincing evidence of an increas-
ing trend. Further, the Valley-wide smooth shows peaks in fall 1974 and
in 1977 that invite explanation. The local smooths show that Hytop and
Loudon have generally higher levels than the other sites. They also
show some interesting peaks.
The corresponding smooths for SS are shown in Figure 3. The Valley-
wide smooth seems to show a decreasing trend. As part of this trend,
the Valley-wide smooth shows that the winter, spring, and summer of 1975
had unusually high levels. The local smooths show that Hytop has generally
higher levels than the other sites and that, except for 1974, Loudon has
higher winter levels. Like the TSP smooths, these smooths have many
peaks that suggest further investigation.
The roughs are better summarized by the correlations shown in
Table 1 than depicted by graphs because the roughs appear nearly random.
This table requires four explanations. First, before May 17, 1976,
Loves Mill was sampled one day and Loudon was sampled two days before
the other three sites. Starting May 17, 1976, all sites were sampled on
the same day. Thus, the table has two entries for Loves Mill, Loudon,
and the Valley-wide rough, the first for the earlier period and the
second for the later period. Second, the Valley-wide rough summarizes
the three, then five, roughs from the sites sampled on the same day. The
table contains correlations of the individual roughs and the Valley-wide
rough to show the similarity of these roughs. Third, the table contains
in the lower triangle the numbers of observations not missing and there-
fore included in the correlations. These numbers are helpful in making
inferences. Fourth, the correlations are computed by a robust method
that prevents a few observations from dominating the results. This
method is the standardized sum and difference method with 5 percent
Winsorized variances centered at 10 percent trimmed means.17
The roughs have two striking features: (1) Roughs from sites
sampled the same day are closely related; and (2) in most cases, for the
days on which the Valley-wide rough is unusually high or low, all sites
have unusually high or low levels.
THE ALGORITHM
Having described the data, we now show how the decomposition is
computed. Before the decomposition is started, missing values in the
data are replaced by linear interpolation between nearby values from the
same site. Each data point is then transformed as described by Equa-
tion (1). These steps produce a 5 x 258 array of values that should be
thought of as being in block 1 of Figure 4 at the start of the decomposi-
tion. These values represent the period from November 1973 through
January 1978. From each smooth, seven values are dropped from each end
to reduce the smooths to exactly four years.
-------
-8-
.00
.00
CD
.00
Ld
CJ
LJ
LD
O
.00
.00
.25
.00
-.25 -
1 1 1 1 T
"1 1 1 1 1 T
LOVES MILL
GILES COUNTY
J I I I I L
J L
WSSFWSSFWSSFWSSF
1974 1975 1976 1977
Figure 2. Smoothed levels of total suspended particulates,
-------
-9-
.00
.00
.00
LU
C_J
I •«
CD
O
.00
.25
.00
-.25
- -
1 - 1 - 1 - 1 - 1 - 1 - 1 - 1 - r
LOVES MILL
v\
HYTQP
MvA/^/^/yyA,
GILES COUNTY
i i i i i i i i
WSSFWSSFWSSFWSSF
1974 1975 1976 1977
Figure 3. Smoothed levels of suspended sulfates.
-------
-10-
TABLE 1. ROBUST CORRELATIONS AND NUMBER OF NONMISSING
OBSERVATIONS FOR THE ROUGHS
Loves Mill
Loudon
Hytop
Giles County
LBL
Valley-wide
Loves
Mill
•u
129/ 84°
132/ 90
131/ 79
137/ 91
143/ 95
Loudon
Total suspended
0.09/0.58a 0.
0.
128/ 88
126/ 77
133/ 87
138/ 92
Hytop
Giles
County
LBL
Valley-
wide
particulates
32/0.48
12/0.61
--
212
228
142/ 97
0.31/0.57
-0.01/0.60
0.62
--
217
141/ 85
0.35/0.52
0.09/0.51
0.53
0.60
--
148/ 98
0.34/0.76
0.05/0.76
0.84/0.77
0.83/0.91
0.84/0.76
— —
Suspended sulfates
Loves Mill
Loudon
Hytop
Giles County
LBL
Valley-wide
V,
125/ 86°
130/ 92
132/ 82
133/ 93
141/ 97
0.30/0.58a 0.
0.
126/ 89
126/ 80
129/ 89
136/ 93
43/0.55
09/0.69
--
217
227
142/ 98
0.21/0.54
0.07/0.58
0.66
—
222
144/ 88
0.33/0.47
0.21/0.45
0.51
0.46
—
145/100
0.40/0.73
0.11/0.83
0.90/0.83
0.82/0.83
0.73/0.67
^Correlation before May 17, 1976/correlation after May 17, 1976.
^Number before May 17, 1976/number after May 17, 1976.
-------
-11-
i LOVES MILL
1 LOUDON
HTTOP
GILES COUNTY
LBL
\
INDIVIDUAL
ROUGHS
i
i
\
3 VALLEY-yiDE ROUGH
1 LOCAL
SMOOTHS
\
7
* VALLEY-yiDE SMOOTH
i
5 SEASONAL
COMPONENT
Figure 4. Data decomposition showing flow of calculations.
-------
-12-
The first step in the decomposition computes a smooth trace through
the data for each site. The particular algorithm we chose for this
purpose is called 4253H and is specified below.8 The smooth traces are
subtracted from the data originally in block 1 and stored in block 2;
the differences are left in block 1.
The second step computes the Valley-wide component from the values
remaining in block 1 by finding the median value for each sampling day.
Before May 17, 1976, these medians are determined by Hytop, Giles County,
and Land Between the Lakes only. Thereafter, they are determined by all
sites. The resulting medians are stored in block 3.
The third step replaces the values stored in block 1 that were
initially missing. These values are replaced by the corresponding value
of the Valley-wide component in block 3, except for the missing values
from Loves Mill and Loudon before May 17, 1976. Missing values from
these two sites before May 17, 1976, are replaced by zero.
The fourth step ensures that in the end the values in block 1 have
no smooth trace. It repeats computations like those in steps 1 through
3, using the values left in block 1 as inputs. The smooth traces of the
values in block 1 are computed, subtracted from the values in block 1,
and added to the values in block 2. Next, the Valley-wide component is
recomputed as in step 2 and stored in block 3. Then, missing values are
replaced as in step 3. Finally, these analogs of steps 1 through 3 are
repeated yet another two times. What then remains in block 1 are the
individual roughs, and what then remains in block 3 is the Valley-wide
rough.
The fifth step removes the Valley-wide component from the values
stored in block 2. It averages the values for each sampling day, ignor-
ing the one- and two-day differences in the schedule. These averages
are subtracted from block 2, leaving the local smooths, and are stored
in block 4.
The sixth step obtains the seasonal component from these averages.
The seasonal component is computed for each of the 61 sampling days in
a year by finding the midmean of the yearly values for that sampling
day. It is subtracted from block 4, leaving the Valley-wide smooth, and
is placed in block 5. To obtain the values in Figure 1, we retransformed
this seasonal component.
The smoothing algorithm 4253H is the following sequence of computa-
tions.8 First, running medians of length 4 and 2 are applied to give
yt(1) = (1/2) median [y^, y^,
+ (1/2) median [y, y,
Second, a running median of length 5 is applied to give
y^'", 7t(1), yt+1(I), »„,<"]. (3,
-------
-13-
Third, a running median of length 3 is applied to give
yt(3)= median [yt_/2),yt(2),yt+1(2)]. (*>
Fourth, a running weighted average called banning is applied to give
yt(4) = [Vl(3) * ' yt(3) * yttl(3)]/*. (5)
The above formulas show that y *• ' is obtained from 13 original data
points, Yt_g, . . . , Yt+6- To make the output the same length as the
input, six points are joined to each end of the series. The points at
the beginning are obtained by applying the sequence 4253H to the first
14 data points to give y?(- and Vg . The six new points, which are
denoted by y_5> y_4> . . . , yQ, are obtained by linear extrapolation:
The points for the other end are obtained similarly.
-------
-14-
SECTION 4
INTERPRETATION OF THE DATA
The decomposition of TSP and SS data allows comparison of the
various components to possible causal factors. Some causal factors are
regional and some are local; some vary rapidly and some vary slowly.
Thus, the decomposition is useful because the causal factors relate to
some components, but not to others.
SEASONAL COMPONENT
The seasonal component is not only a prominent feature of most
environmental data, but often the component that is most difficult to
explain unambiguously. This difficulty is due to the seasonal nature of
most possible causes.
The TSP spring peak is interesting in that it seems to be related
to annually recurring events in late March and early April. The most
plausible explanation for this peak is regional agricultural and bio-
logical activity. This period is the planting season in the Tennessee
Valley and also the season for release of pine pollen. Both of these
particulate sources should be important at rural sites and quite possibly
at industrial-urban monitoring sites as well.
The TSP and SS summer peaks coincide with many interrelated factors.
These peaks result from the increased frequency of meteorological condi-
tions conducive to the transformation, transport, and buildup of both
primary and secondary pollutants. Among these factors are
• High incidence of stagnating anticyclonic (high-pressure) airmasses;
• High absolute atmospheric water vapor content;
• High insolation;
• High temperature;
• Higher convective and less advective mixing;
• Low frequency of regional rainfall; and
• Low wind speed.
Although anthropogenic emissions may be the origin of a significant
portion of summer air pollution, the variation in emissions alone does
not seem to account for these peaks since the power demand on the TVA
system is as high in winter as in summer.
-------
-15-
VALLEY-WIDE AND LOCAL SMOOTHS
Smooths are, by definition, representative of persistent behavior
and, as such, are useful in determining the trend of the data. The
Valley-wide smooth is indicative of persistent behavior common to all
sites. The many features of the smooths shown in Figures 2 and 3 have
not been analyzed, but two examples taken from the Valley-wide smooths
and two examples from the local smooths will be discussed.
Examining the Valley-wide smooth for TSP in Figure 2, note that the
fall of 1974 is a period with high levels. We attribute these unusually
high levels to a prolonged dry spell. This dry spell shows the effect
of meteorology on pollutant levels. The mechanisms for the pollutant
increase are the dry conditions and the presence of stagnating high-
pressure systems, which allow a greater amount of wind-borne soil and
pollutant buildup.
Turning to the Valley-wide SS smooth in Figure 3, consider the
general downward trend of the data. It appears that 1974 and 1975
experienced higher sulfate levels than did 1976 and 1977. What does
this indicate? It could represent an actual decline in regional sulfate
concentrations, which as mentioned previously, could be a function of
year-to-year meteorological fluctuations. It also could be the result
of the change in sampling techniques in July of 1976--the switch from
Mine Safety Appliance Co. to Gelman Spectrograde high-volume filters.
Subsequent experimentation with sulfate extraction suggests that the SS
data obtained from the Gelman filters are on the average too low.
The local smooths for TSP and SS at Giles County in the fall of
1974 are unusually low. Examination of the TSP and SS data during this
period indicates either extremely low pollutant concentrations or lost
records. An examination of corresponding data collected from the nearby
Cumberland Steam Plant indicated no unusual data. This suggests that
this negative peak is due to a defective high-volume sampler.
The local SS smooth for LBL in August of 1976 is another inter-
esting example. In this instance, the sixth-day sampling resulted in a
smooth not typical of the entire month. Three of five sampling days
during the month had high levels of SS. These levels, which were
peculiar to LBL, were caused by transport from the Ohio Valley, a meteo-
rological circumstance that occurs infrequently.13 This particular
peak, therefore, is not representative of the entire period.
This problem is an example of failure of the smoothing to separate
the slowly varying and rapidly varying components. With sixth-day
sampling, we are unable to separate these components. This problem,
which is called aliasing18 and is a form of confounding, makes explana-
tion of the smooths more difficult. The effect of aliasing is further
demonstrated in the following example.
The data used in this example are 192 days of daily SS values. In
Figure 5, the six lines superimposed over the actual data are sixth-day
smooths generated by using different starting days. Each smooth shows a
-------
/-A
K)
£
\ 40.0
0)
Z
o
f—(
h
(T
h-
Z
UJ
U
Z
O
U
20.0
0
DAY
180
Figure 5. Daily sulfate data with sixth-day sampling smoothed.
-------
-17-
single peak in late August or early September, although the actual data
have three major peaks in this period. Thus, the six smooths do not
describe the actual data very well. Further, they are not similar to
each other. The problems of aliasing could be eliminated through daily
sampling or at least reduced through more frequent sampling.
VALLEY-WIDE AND INDIVIDUAL ROUGHS
The roughs contain that part of each day's value unsupported by
adjacent values. In other words, a rough is the irregular part of a
series, the high-frequency component. The roughs serve as a means for
detecting unusual episodes. Also, they can be used for comparisons
among the high-frequency variations at the various sites.
Examination of the Valley-wide rough shows episodes that exhibit
extreme levels throughout the region. Seventeen such episodes are
considered in detail. Five episodes were chosen because they had the
highest values of the SS Valley-wide rough. Of these, three are typical
and two are unusual. The other twelve episodes have the four lowest
values of the SS Valley-wide rough and the four highest and four lowest
values of the TSP Valley-wide rough.
The episodes occurring on January 4, 1974, August 26, 1974, and
July 4, 1975, are typical high-SS episodes. The common factor in these
cases is the presence of a stagnating anticyclonic airmass. The winter
episode of January 4, 1974, coincides with the presence of a cold polar
continental anticyclone (PcK), which had been stagnating over middle-
America since January 1. The summer episodes coincide with the presence
of warm maritime anticyclones (TmW), which had stagnated over the south-
eastern United States. The presence of fog, smoke, haze, and low visi-
bility are typically associated with such episodes.
The episodes occurring on January 28, 1974, May 22, 1974, May 7,
1976, and April 7, 1977, are typical low-SS episodes. The common factor
associated with these cases is the presence of regional precipitation in
substantial amounts on the days preceding and/or during the sampling
day. This precipitation is associated with airmass convergence.
Episodes that do not fit into a similar mode require further inves-
tigation. The episodes on February 12, 1977, and on September 10, 1977,
represent two such nontypical cases.
The episode occurring on February 12, 1977 is quite unusual. SS
concentrations at the trend stations were 9.9, 6.5, 8.4, 8.4, and
7.6 (Jg/m3, consistently above the winter seasonal mean of 4.5 pg/m3.
The meteorology on the sampling day was dominated by a cyclonic center,
moving in a northeasterly direction across southeastern Missouri. The
associated cold front resulted in measurable precipitation across the
Valley on the day of sampling. The elevated SS values conceivably
result from two factors: (1) static sampling during the five days
before sampling, when the presence of a stagnating anticyclone could
-------
-18-
have resulted in SS buildup, and (2) partial sampling of this stagnating
airmass until the rain began. It is logical to assume that, had the
cyclone not developed, SS concentrations would have been much greater.
The episode occurring on September 10, 1977, is also quite unusual.
SS concentrations on this day were 16.9, 15.5, 30.4, 27.0, and 0.5 (jg/m3,
generally above the summer seasonal mean of 10.4 |jg/m3. Indeed, as can
be seen in the variation of values, this situation is unusual. On the
day of sampling, the Valley meteorology was dominated by an approaching
cold front from the northwest, followed by a maritime polar anticyclone
(PmK). The 0.5-(Jg/m3 concentration recorded at LBL is so unusually low
that SS concentrations at nearby TVA steam plants were also checked.
This check confirmed low SS concentrations in the northwest section of
the Tennessee Valley—undoubtedly associated with the PmK airmass. The
airmass to the southeast of the front is associated with much higher SS
concentrations. Because of the complex meteorology on the days before
sampling, resulting from the passage of a tropical depression, the
origin of this prefrontal airmass is uncertain. The three-dimensional
trajectory model of the National Weather Service Techniques Development
Laboratory shows that on September 7, 8, and 9, the trajectories into
the Valley were from the north to northeast. These trajectories crossed
the large sulfur dioxide emissions sources in the Ohio Valley.
The episodes occurring on October 25, 1974, and July 28, 1975, are
examples of one type of high-TSP episode. In this case, the common
factor is a stagnating anticyclone, a PcK in the former episode and a
TmW in the latter. The stagnating conditions are associated with fog,
haze, smoke, low wind speed, and reduced visibility. We believe that,
in cases such as these, fine particulates from natural and anthropogenic
sources build up in the atmosphere and result in the elevated TSP levels.
The episodes occurring on January 4, 1974, and April 4, 1974, are
also examples of high-TSP episodes. The mechanisms are, however, much
different from the ones discussed above. In these cases, the episodes
are associated with frontal activity, rain, and high wind speeds. We
believe that in these cases, coarse particulates, primarily from natural
sources, are carried aloft by the high winds associated with the frontal
activity and result in the elevated TSP levels.
The episodes occurring on February 4, 1975, May 25, 1977, September 16,
1977, and November 3, 1977, are examples of low-TSP episodes. The
common factor associated with these episodes is the presence of regional
precipitation before and on the day of sampling. In all these episodes,
the precipitation is associated with frontal activity. There is no
readily apparent explanation for differentiating between meteorological
conditions occurring during the second type of elevated TSP episodes and
these low episodes. The differences are most likely related to the
sources.
The individual roughs isolate unusual data points and describe for
each site the rapidly varying part of sample-to-sample variation.
Unusual data points may reflect an unusual set of environmental circum-
stances or an error in sampling, laboratory work, or recording. As
such, the individual roughs may be used in quality assurance.
-------
-19-
As seen in Table 1, the individual roughs are well correlated.
This, indeed, is a manifestation of the common regional behavior. When
compared with the Valley-wide rough, these correlations provide a quanti-
tative measure of regional "representativeness." The Giles County site
appears to be most representative of regional TSP behavior, whereas the
Loudon, Hytop, and Giles County sites appear to be most representative
of regional SS behavior.
-------
-20-
SECTION 5
DESIGN OF MONITORING
Ambient monitoring can be used to estimate exposure as part of a
study of pollution effects or to evaluate sources as part of a study of
control strategies. To achieve this latter objective, power plant
sources, other industrial sources, agricultural sources, other local
sources, and distant sources must be resolved. Considerations important
to accomplishing this are shown by the data analyzed in this paper.
One consideration is how to resolve the agricultural and biological
contribution to the TSP. Compared with some other contributions, this
contribution is believed to contain mostly larger-size particles and to
be less dangerous to human health.19 Whatever the relative health
effects of various types of particles, this contribution must be distin-
guished as effectively as possible in studies of control strategies.
Thus, studies of control strategies are an important basis for the
frequently repeated recommendation that particulates be measured by size
and chemical composition.
Another consideration is how the meteorological influence can be
removed. This influence is important in the study of long-term varia-
tions, which is one of TVA's purposes for monitoring at these isolated
sites. How such variations apparent in the data are interpreted depends
on their cause: Variations caused by the weather have different implica-
tions for control than variations caused by other factors. Thus, long-
term variations must be analyzed by removing the influence of year-to-year
differences in the weather to obtain the series that would have occurred
had each year's weather been the same. This series should show the part
of the variation caused by changes in emissions.
The meteorological influence is also likely to be important in
analyzing data from sites surrounding a power plant. This analysis
could start with the same decomposition used above. Because all the
sites would be sampled on the same day, the common rough, which is the
analog of the Valley-wide rough, would be subtracted from the individual
roughs to obtain a local rough for each site. The dependence of the
local roughs and smooths on plume behavior would contain the evidence of
pollution from the power plant. However, this dependence might exist
even with no power plant contribution because of other sources. Thus,
the resolution of sources also arises in this context, showing that
analysis involving the weather will also be important for power plant
data.
Analysis involving the meteorological influence, although it is
never easy, is made harder by the aliasing problem. In the analysis of
these data, the aliasing problem prevents separation of slowly varying
components from rapidly varying components. Such separation is important
in an observational study such as this, where the objective is to explain
as much of the variation as possible. If the sampling were daily, the
data would be separated into more than just rough and smooth components.
The most irregular component would contain rare meteorological events as
-------
-21-
well as the results of measurement blunders. Another component would
reflect mostly the passage of weather systems, thus tracking the day-to-
day variations in transport. A third component would be compared with
monthly summaries of causal factors in the same way that we would like
to compare Figures 2 and 3 with such summaries. This more extensive
decomposition should allow monitoring to provide, under some circum-
stances, better information than modeling.
The recommendation that the sampling frequency for particulates be
increased has been made previously on the basis that particulate measure-
ments are a random sample.20 Although this basis for thinking about air
quality data is widespread,21 it fails to acknowledge the possibility of
modeling and adjusting for meteorological and other influences. When
adjustment for these influences is considered, the major problem with
sixth-day sampling is seen to be aliasing rather than accuracy.
The features of these data, revealed by the analysis demonstrated
in this paper, suggest various changes to be made in the data collection.
These changes include more resolution in the sampling itself and collec-
tion of more ancillary data. If these changes were made, adequate data
for more detailed and quantitative model building would become available.
The analysis demonstrated here has thus been shown to be important in
ensuring that all the data necessary to satisfy the purposes of the
monitoring are collected. Analysis with this purpose should be a part
of all ongoing monitoring.
-------
-22-
REFERENCES
1. Jutze, G. A., and K. E. Foster. Recommended Standard Method for
Atmospheric Sampling of Fine Particulate Matter by Filter Media—
High Volume Sampler. J. Air Pollut. Control Assoc., 17:17-25, 1967.
2. U.S. Public Health Service. Determination of Sulfate in Atmospheric
Suspended Particulates. 999-AP-ll, 1965.
3. Appendix B--Reference Method for the Determination of Suspended
Particulates in the Atmosphere (High Volume Method). Fed Regist
36(84):8191-8194, 1971. "'
4. U.S. Environmental Protection Agency. Tentative Method for the
Determination of Sulfates in the Atmosphere (Automated Technicon II
Methylthymol Blue Procedure), 1977.
5. Goldsmith, B. J., and J. R. Mahoney. Implications of the 1977 Clean
Air Act Amendments for Stationary Sources. Environ. Sci Technol
12:144-149, 1978.
6. Pratt, J. W., et al. Environmental Monitoring. National Academy of
Sciences, Washington, D.C., 1977.
7. Tukey, J. W. Exploratory Data Analysis. Addison-Wesley, Reading, Mass.
8. Velleman, P. F. Robust Nonlinear Data Smoothers: Definitions and
Recommendations. Proc. Natl. Acad. Sci. USA, 74:434-436, 1977.
9. Hidy, G. M., E. Y. long, and P. K. Mueller. Design of the Sulfate
Regional Experiment (SURE), vol. 1. EPRI EC-125, Electric Power
Research Institute, 1976.
10. Lioy, P. J., G. T. Wolff, J. S. Czachor, P. E. Coffey, W. N. Stasiuk,
and D. Romano. Evidence of High Atmospheric Concentrations of
Sulfates Detected at Rural Sites in the Northeast. J. Environ. Sci.
Health, A12:l-14, 1977.
11. Galvin, P. J., P. J. Samson, P. E. Coffey, and D. Romano. Transport
of Sulfate to New York State. Environ. Sci. Technol., 12:580-584,
1978.
12. Tong, E. Y., and R. B. Batchelder. Compilation and Analysis of Data
Sets for the Evaluation of Regional Sulfate Models. Teknekron, Inc.,
Berkeley, California, 1978.
13. Reisinger, L. M., and T. L. Crawford. August 1976 Sulfate Episodes
in the Tennessee Valley Region. TVA/EP-79/04, Tennessee Valley
Authority, Chattanooga, Tennessee.
14. U.S. Environmental Protection Agency. National Air Quality and
Emissions Trend Report, 1976. EPA-450/1-77-002, 1977.
-------
-23-
15. Altshuller, A. P. Atmospheric Sulfur Dioxide and Sulfate—Distribution
Of Concentration in Urban and Nonurban Sites in the United States.
Environ. Sci. Technol., 7:709-712, 1973.
16. Rowe, M. D., S. C. Morris, and L. 0. Hamilton. Potential Ambient
Standards for Atmospheric Sulfates: An Account of a Workshop.
J. Air Pollut. Control Assoc., 28:772-775, 1978.
17. Gnanadesikan, R. Methods for Statistical Data Analysis of Multi-
variate Observations. John Wiley and Sons, Inc., New York, 1977.
p. 132.
18. Bloomfield, P. Fourier Analysis of Time Series: An Introduction.
John Wiley and Sons, Inc., New York, 1976.
19. Hidy, G. M., et al. Summary of the California Aerosol Characteri-
zation Experiment. J. Air Pollut. Control Assoc., 25:1106-1114, 1975.
20. Tong, E. Y., and S. A. DePietro. Sampling Frequencies for Determining
Long-Term Average Concentrations of Atmospheric Particulate Sulfates.
J. Air Pollut. Control Assoc., 27:1008-1011, 1977.
21. Mage, D. T., and W. R. Ott. Refinements of the Lognormal Probability
Model for Analysis of Aerometric Data. J. Air Pollut. Control Assoc.,
28:796-798, 1978.
-------
TECHNICAL REPORT DATA
(Please read Intlnictions on the reverse before completing)
1. REPORT NO.
EPA/600/7-79-084
4. TITLE AND SUBTITLE
THE ANALYSIS OF SUSPENDED PARTICULATES AND SULFATES:
A WAY TO BEGIN
6. PERFORMING ORGANIZATION CODE
3. RECIPIENT'S ACCESSI OfV NO.
5. REPORT DATE
March 1979
7. AUTHOR(S)
Walter Liggett and William Parkhurst
8. PERFORMING ORGANIZATION REPORT NO.
TVA/ONR-79/03
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Office of Natural Resources
Tennessee Valley Authority
Chattanooga, TN 37401
10. PROGRAM ELEMENT NO.
INE - 625 B
11. CONTRACT/GRANT NO.
80 BDM
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Research & Development
Office of Energy, Minerals & Industry
Washington, D.C. 20460
13. TYPE OF REPORT AND PERIOD COVERED
Milestone
14. SPONSORING AGENCY CODE
EPA/600/7
15. SUPPLEMENTARY NOTES
This project is part of the EPA-planned and coordinated Federal Interagency
Energy/Environment R&D Program.
16. ABSTRACT
Total suspended particulate (TSP) and suspended sulfate (SS) levels have been
sampled since November 1973 at five isolated sites across the Tennessee Valley.
A method for beginning to analyze such data is demonstrated. This beginning is
intended to lead finally to information on pollution sources, an objective that
may require modeling meteorological influences and resolving sources. Analysis
with this objective, which can be very complex, is effectively begun by using the
method demonstrated in this paper. Applied to the TSP and SS data, this method
suggests agricultural contributions to TSP levels, distant-source contributions
to SS levels, and various influences of the meteorology. This method also shows
deficiencies in the data collection that prevent the building of better, more
quantitative models. One deficiency in this data set is the sixth-day sampling,
which is not frequent enough to allow monthly variations in pollution levels to
be distinguished from more rapid variations. Thus, data analysis would be more
effective if the sampling frequency were increased and, further, if particle
size and chemical composition were better resolved.
(Circle One or More)
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.IDENTIFIERS/OPEN ENDED TERMS
c. COSATI Field/Group
Inorganic Chemistry
Charac. Meas. & Monit.
7B
3. DISTRIBUTION STATEMENT
Release to public
19. SECURITY CLASS (This Report)
CURITY CLASS (Ihi.
Unclassified
21. NO. OF PAGES
23
20. SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)
------- |