United States Office of Air Quality EPA-450/4-80-006a
Environmental Protection Planning and Standards February 1980
Agency Research Triangle Park NC 27711 I/A / /
Air
Analysis of the St. Louis
RAMS Ambient
Particulate Data
Volume I:
Final Report
-------
EPA-450/4-80-006a
Analysis of the St. Louis
RAMS Ambient Particulate Data
Volume I:
Final Report
by
John Trijonis, John Eldon,
John Gins, and George Berglund
Technology Service Corporation
2811 Wilshire Boulevard
Santa Monica, California 90403
Contract No. 68-02-2931
EPA Project Officers:
Thompson Pace
EPA Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
James Reagan
EPA Environmental Sciences Research Laboratory
Research Triangle Park, North Carolina 27711
U.S. Environmental Probation Agency
Re.Son V, Library Prepared for
US. ENVIRONMENTAL PROTECTION AGENCY
Office of Air, Noise, and Radiation
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
February 1980
-------
This report is issued by the Environmental Protection Agency to report
technical data of interest to a limited number or 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 (MD-35), U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina 27711; or for a nominal fee,
from the National Technical Information Service, 5285 Port Royal Road,
Springfield, Virginia 22161.
This report was furnished to the Environmental Protection Agency by
Technology Service Corporation, 2811 Wilshire Blvd., Santa Monica,
Ca 90403, in fulfillment of Contract No. 68-02-2931. The contents of
this report are reproduced herein as received from Technology
Service Corporation. 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.
Publication No. EPA-450/4-80-006a
U,S. Environmental Protection Agency
-------
ABSTRACT
In this report, a variety of data analysis methods are used to study
the 1976 particulate data from the Regional Air Monitoring System (RAMS)
in St. Louis. The aerosol data, collected at ten sites, include Hi-Vol
measurements of total suspended particulate mass (TSP) as well as dichoto-
mous sampler measurements of inhalable particulate mass (IP); IP is sub-
divided into fine particles (less than 2.4 ym in diameter) and coarse
particles (between 2.4 and 20 ym in diameter). This study also includes
dichotomous sampler data for eight trace elements (S, Si, Al, Ca, Pb, V,
Ti, and Fe) and data for eleven meteorological parameters.
The analyses characterize the spatial pattern of particulate matter
in-and-near St. Louis; background aerosol concentrations and particulate
transport; temporal patterns of particulate concentrations; the dependence
of aerosol concentrations on meteorology; and the relationship between Hi-
Vol data and dichotomous data. These analyses, as well as chemical element
balance calculations, shed considerable light on the sources of particulate
matter in St. Louis.
Averaged over the RAMS network, IP mass consists of 50% fine particles
and 50% coarse particles. Sulfates, secondary aerosols occurring on a
large (air basin and synoptic) scale, constitute 53% of fine mass and 29%
of IP. Crustal material, mostly consisting of man-made fugitive dust, com-
prises 83% of coarse mass and 47% of IP. That sulfate is the major source
of fine mass and that crustal material dominates coarse mass are the two
major themes apparent in the spatial, temporal, and meteorological patterns
of the dichotomous data.
m
-------
CONTENTS
ABSTRACT
vi i
FIGURES
TABLES X
EXECUTIVE SUMMARY l
1.0 INTRODUCTION U
1.1 The RAMS Particulate Data u
1.2 Data Quality Analysis 13
1.3 Summary of Annual Means 16
1.4 Chemical Element Balance 18
1.5 Organization of the Report 20
2.0 SPATIAL PATTERNS OF ST. LOUIS PARTICULATE DATA 21
2.1 Geographical Distribution of Ambient Particulate
Matter
-------
4.2 Weekly (Hebdomadal) Patterns 71
4.2.1 Weekly Patterns at Two Selected Sites 71
4.2.2 Weekend-Weekday Differences 80
4.3 Diurnal Patterns 83
5.0 DECISION-TREE ANALYSIS OF PARTICULATE AND
METEOROLOGICAL DATA 85
5.1 The CART Decision-Tree Program 85
5.2 Decision-Trees Relating IP and FINE to
Elemental Concentrations 89
5.3 Decision-Trees Relating Participate Variables
to Meteorology 92
6.0 RELATIONSHIP BETWEEN TSP AND IP 109
6.1 Scatterplots of Dichotomous Data Versus Hi-Vol Data . . 109
6.2 Ratio of IP to TSP 114
6.2.1 Spatial Patterns of IP/TSP 114
6.2.2 Temporal Patterns of IP/TSP 119
6.2.3 Decision-Tree Analysis of IP/TSP 119
6.2.4 Conclusions Regarding the IP/TSP Ratio 121
7.0 SOURCES OF PARTICULATE MATTER IN ST. LOUIS 123
7.1 Sources of FINE, COARSE, IP, and TSP 123
7.1.1 Fine Fraction of IP 123
7.1.2 Coarse Fraction of IP 124
7.1.3 Total IP 126
7.1.4 TSP 127
7.2 Urban Versus Rural Sources of IP 127
8.0 REFERENCES 129
VI
-------
FIGURES
Number
Page
1.1 The RAMS monitoring network in St. Louis ......... 12
1 2 Source contributions averaged over the ten RAMS
sites for July and August 1976, as determined
by chemical element balance (Dzubay, 1979) ........ 19
2.1 Geographical distirbution of TSP mass concentration
in the St. Louis area (yg/m3) .............. "
2.2 Geographical distribution of IP mass concentration
in the St. Louis area (yg/m3) .............. 23
2.3 Geographical distribution of FINE mass concentration
in the St. Louis area (yg/m3) .............. 24
2.4 Plots of normalized particulate concentrations (with
respect to station 124) versus distance from station
103 ........................... 25
3.1 Regional upwind/downwind analyses for TSP ........ 35
3.2 Regional upwind/downwind analyses for IP ......... 36
3.3 Regional upwind/downwind analyses for FINE ........ 37
3.4 Regional upwind/downwind analyses for COARSE ....... 38
3.5 Regional upwind/downwind analyses for sulfur ....... 41
3.6 Regional upwind/downwind analyses for lead ........ 42
3.7 Regional upwind/downwind analyses for silicon ...... 43
3.8 Wind roses for the ten RAMS particulate sites ...... 45
3.9 Pollution roses for IP at the ten RAMS particulate
sites .......................... 46
3 10 Pollution roses for FINE at the ten RAMS particulate
sites .......................... 47
3.11 Dosage roses for IP at the ten RAMS particulate sites
49
vn
-------
Number Page
3.12 Dosage roses for FINE at the ten RAMS particulate
sites 50
3.13 Wind and pollution roses for site 103 51
3.14 Wind and pollution roses for site 105 54
3.15 Wind and pollution roses for site 124 57
3.16 Emission sources in the vicinity of St. Louis
(Dzubay, 1979) 60
4.1 Seasonal patterns of TSP, IP, COARSE, and FINE 64
4.2 Seasonal patterns of elemental concentrations 65
4.3 Monthly averages of daily (vector) average wind speed . . 68
4.4 Monthly averages of number of days persistent calm
(#DY CALM) (Lambert Field data) 68
4.5 Monthly averages of morning and afternoon mixing
heights 69
4.6 Monthly averages of daily minimum and maximum
temperatures 70
4.7 Monthly averages of relative humidity 70
4.8 Number of days of precipitation per month 72
4.9 Monthly averages of number of days since last
precipitation 73
4.10 Hebdomadal and diurnal patterns of IP (yg/m ) 74
o
4.11 Hebdomadal and diurnal patterns of FINE (yg/m ) 74
4.12 Hebdomadal and diurnal patterns of COARSE (yg/m3) .... 75
o
4.13 Hebdomadal and diurnal patterns of sulfur (ng/m ) . . . . 75
4.14 Hebdomadal and diurnal patterns of silicon (ng/m ). . . . 76
3
4.15 Hebdomadal and diurnal patterns of aluminum (ng/m ) . . . 76
o
4.16 Hebdomadal and diurnal patterns of calcium (ng/m ). . . . 77
vm
-------
Page
Number
4.17 Hebdomadal and diurnal patterns of lead (ng/m ) ??
4.18 Hebdomadal and diurnal patterns of vanadium (nrj/m ) . . . 78
o
4.19 Hebdomadal and diurnal patterns of titanium (ng/m ) . . . 78
5
4.20 Hegdomadal and diurnal patterns of iron (ng/m ) 79
5.1 Example decision-tree, sulfur at site 103 versus
meteorological variables
6 1 Scatterplot of dichotomous variables versus TSP
at site 108
6.2 Scatterplot of dichotomous variables versus TSP
at site 115
6 3 Scatterplot of dichotomous variables versus TSP
at site 124
6.4 Relationship between annual mean IP and annual
mean TSP among the RAMS sites
6.5 Geographical distribution of IP/TSP ratio in the
St. Louis area
6.6 IP/TSP ratio versus distance to site 103
6.7 Seasonal pattern of the IP/TSP ratio
6.8 Weekly pattern of the IP/TSP ratio 12°
IX
-------
TABLES
Number Page
1.1 Trace Elements Included in This Study 1
1.2 Meteorological Parameters Included in This Study .... 15
1.3 Annual Mean Particulate Concentrations 17
2.1 Average Particulate Concentration as a Function
of Site Type 28
2.2 Rural Vs. Suburban Vs. Urban Concentrations of
St. Louis Particulate Data 30
2.3 Interconnections for St. Louis Particulate Data .... 31
3.1 Summary of Upwind/Downwind Analyses for TSP, IP,
FINE, and COARSE 39
4.1 Weekend-Weekday Differences in St. Louis
Particulate Concentrations 81
5.1 Percent of Variance in the Decision-Tree for Sulfur
at Site 103 Accounted for by Individual Meteorological
Variables 88
5.2 Percent Variance Explained by Decision-Trees
Relating IP and FINE to Elemental Variables 89
5.3 Relationship of FINE to Elemental Concnetrations
as Found in the Decision-Tree Analysis 90
5.4 Relationship of IP to Elemental Concentrations
as Found in the Decision-Tree Analysis 91
5.5 Percent Variance Explained by Decision-Trees Relating
Particulate Concentrations to Meteorological Parameters. 94
5.6 Relationship of TSP to Meteorological Conditions
as Found in the Decision-Tree Analysis 95
5.7 Relationship of IP to Meteorological Conditions
as Found in the Decision-Tree Analysis 96
5.8 Relationship of FINE to Meteorological Conditions
as Found in the Decision-Tree Analysis 97
x
-------
Number
Page
5.9 Relationship of COARSE to Meteorological Conditions
as Found in the Decision-Tree Analysis 98
5.10 Relationship of Sulfur to Meteorological Conditions
as Found in the Decision-Tree Analysis 99
5.11 Relationship of Silicon to Meteorological Conditions
as Found in the Decision-Tree Analysis 100
5.12 Relationship of Aluminum to Meteorological Conditions
as Found in the Decision-Tree Analysis 101
5.13 Relationship of Calcium to Meteorological Conditions
as Found in the Decision-Tree Analysis 102
5.14 Relationship of Lead to Meteorological Conditions
as Found in the Decision-Tree Analysis 103
5.15 Relationship of Vanadium to Meteorological Conditions
as Found in the Decision-Tree Analysis 104
5.16 Relationship of Titanium to Meteorological Conditions
as Found in the Decision-Tree Analysis 105
5.17 Relationship of Iron to Meteorological Conditions
as Found in the Decision-Tree Analysis 106
6.1 Correlation/Regression Analysis Between Daily Values
of Dichotomous and Hi-Vol Data 110
6.2 Annual and Monthly Statistics for IP/TSP Ratio 115
6.3 IP/TSP Ratio Vs. Distance to Site 103 118
6.4 Percent Variance Explained in the IP/TSP Ratio by
Decision-Tree Analysis 121
XI
-------
EXECUTIVE SUMMARY
The Regional Air Monitoring System (RAMS) operated by EPA in St. Louis
has supplied a rich data set for studying ambient particulate concentrations
in-and-near a major metropolitan area. In order to provide an overview of
the data and to illustrate some of the major implications of the data, this
report presents a series of statistical analyses conducted with the RAMS
particulate measurements. The statistical methods applied range from simple
graphical or tabular displays to complex techniques such as decision-tree/
pattern-recognition analysis. The investigation addresses not only aerosol
mass concentrations, but also the size distribution and chemical composition
of the particulate matter.
The data in this report include aerosol measurements for the year 1976
taken at the RAMS sites plus 1976 meteorological data collected at both the
RAMS sites and Lambert International Airport. Ten of the RAMS stations mon-
itored particulate matter; based on distance to the center-city, we have
classified five of these sites as urban, three as suburban, and two as rural.
Each of the RAMS particulate stations was equipped with two aerosol mass
monitors: a Hi-Vol sampler measuring total suspended particulate mass (TSP),
and a dichotomous sampler measuring inhalable particulate mass (IP). The
TSP data essentially represent the mass concentration of particles less than
approximately 50 ym in diameter. The IP measurements, in total representing
*
the mass of particles less than approximately 20 ym in diameter, consist of
two parts: mass of fine particles (less than 2.4 ym in size) which we denote
as FINE, and mass of coarse particles (between 2.4 and 20 ym in size) which
we denote as COARSE. In addition to TSP, IP, COARSE, and FINE, this study
analyzes data for eight of the chemical elements measured as part of the
dichotomous sampler program. Although some of these elements stem from
Actually, our use of the term "inhalable particulate mass" involves a
slight misnomer. EPA (Miller et al., 1979) has recently recommended that
particles less than 15 ym in diameter be defined as inhalable particulate
matter. Although the type of dichotomous sampler used in St. Louis had a
20 ym rather than a 15 ym upper size cut-off, for convenience we will
nevertheless refer to these data as IP.
-------
several source types, we have found that, in St. Louis, these elements serve
as key tracers for the following source types: sulfate aerosol (sulfur),
shale-type crustal material (silicon and aluminun), limestone-type crustal
material (calcium), automotive exhaust (lead), fuel oil combustion (vanadium),
paint pigment plant (titanium), and iron/steel industry (iron). As des-
cribed in a companion report (Eldon et a!., 1979), all the particulate data
were subjected to a data quality analysis before being used in this study.
The specific purposes of this study are to investigate the spatial
patterns of particulate matter in-and-near St. Louis (Chapter 2); to estimate
background aerosol concentrations and examine particulate transport (Chapter
3); to characterize the temporal patterns of particulate concentrations
(Chapter 4); to analyze the dependence of particulate concentrations on
meteorology (Chapter 5); to determine the relationship between Hi-Vol data
and dichotomous data (Chapter 6); and, finally, to assess the contributions
of various source types to ambient aerosol concentrations (Chapter 7). The
following subsections summarize our findings and conclusions. For convenient
referal, the summary is organized according to the order of the chapters,
except Chapter 7 (Sources of Particulate Matter in St. Louis) is presented
first. We discuss the source assessment first because many of our most im-
portant conclusions concern the origins of particulate matter, and because
an understanding of sources facilitates the discussions of the other chapters.
Sources of Particulate Matter
3
Annual mean concentrations of IP range from 29 to 60 yg/m among the
RAMS sites* Averaged over the RAMS network, IP consists of 50% FINE and
50% COARSE. The chemical element balance of Dzubay (1979), modified to
include data for the entire year of 1976, implies that three-fourths of IP
consists of two components -- sulfate (29%) and crustal material (47%).
**
Note that all averages in this report represent arithmetic means rather
than geometric means.
All further allusions to chemical element balance in this summary will
refer to Dzubay"s results as modified herein to cover the entire year
of 1976.
-------
FINE consists basically of man-made participate matter, especially
secondary aerosols (aerosols formed from gas-to-particle conversion),
especially sulfates. Chemical element balance calculations imply that,
averaged over the year and over all RAMS sites, sulfate constitutes 53%
of FINE. That sulfur (or sulfate) is the major determinant of FINE is also
strongly evidenced by several other analyses in this report. These analyses
further indicate that sulfate is mostly secondary aerosol (formed from S02
emissions) rather than primary aerosol (directly emitted), and that sulfate
tends to be a large scale phenomenon partly air basin in scale and partly
synoptic in scale. Because electric power plants contribute approximately
80% of SO emissions in the St. Louis AQCR and 70% on the larger synoptic
scale, they are likely candidates as the major source of sulfate.
The chemical element balance and other analyses demonstrate that no
single type of aerosol dominates the non-sulfate 47% of FINE. Rather, the
non-sulfate part of FINE appears to be a mixture of fairly small contributions
from numerous sources: secondary organic aerosols, secondary nitrates,
automotive exhaust particles, particles from stationary source fuel com-
bustion, the fine fraction (i.e. the lower tail of the size distribution)
of suspended dust, etc.
The chemical element balance shows that 83% of COARSE is crustal mater-
ial, about two-thirds shale-type and one-third limestone-type. Numerous
spatial, temporal, and meteorological patterns in the data support the con-
clusion that COARSE is dominated by crustal material, mostly shale-type.
Our analyses further suggest that the crustal material arises primarily
from ubiquitous area sources of fugitive dust rather than industrial sources
of fly ash, and that most of the fugitive dust is man-made rather than wind-
blown. .Despite the omnipresence of automotive traffic, motor vehicles ap-
parently do not constitute a predominant source of fugitive dust. Rather,
a variety of dust sources -- paved roads, unpaved roads, quarries, construc-
tion, agriculture, soil dust, and certain industries -- seems implicated.
A reason why traffic is not the predominant source of dust may be that the
RAMS sites (unlike many typical urban particulate monitors) are located
away from major roads. Limestone-type dust appears to be more closely
-------
associated with commercial/industrial activity (possibly cement manufactur-
ing and construction) than shale-type dust.
The remaining, non-crustal 17% of COARSE stems from a wide variety of
sources. The data suggest that most of these sources are man-made.
In the sense that crustal material and sulfates are the two major
sources, the origins of IP are generally similar at the urban and rural
locations. The relative importance of crustal material and COARSE is slight-
ly greater at urban sites (basically because of more limestone-type dust),
while the relative importance of sulfate and FINE is slightly greater at
rural sites. There do exist, however, some significant urban/rural differ-
ences in particulate matter. Primary aerosols from auto exhaust, fuel com-
bustion, and industrial sources are typically three to five times greater
at urban locations. Particulate matter from these sources is not a domin-
ant contributor to IP mass, but it may be of concern because of its chemical
nature. The specific sources of fugitive dust also display distinct urban/
rural differences; dust in the center-city seems to stem mostly from urban
activities, while dust in the countryside likely arises more from agriculture
and natural sources (i.e. wind-blown dust).
Our analyses indicate that dust is the dominant source of TSP. Al-
though our analyses show that IP is less affected by fugitive dust (especially
wind-blown dust) than TSP, fugitive dust nevertheless contributes signifi-
cantly to IP (nearly one-half of IP is crustal material, most of which appears
to be dust). Air quality standards for IP, therefore, will not completely
eliminate the fugitive dust problems associated with the TSP standard.
Spatial Patterns
The one outstanding feature in the spatial patterns for TSP, IP, FINE,
and COARSE is a concentric urban bubble; all four particulate variables ex-
hibit nearly monotonic decreases from site 103 (near the center of the RAMS
network) radially out to the rural sites. One major reason for this smooth
concentricity is that the RAMS sites are generally isolated from strong
localized sources (fugitive dust or otherwise). In this sense, the RAMS
particulate sites are not fully representative of existing Hi-Vol sites
-------
nationwide, many of which are significantly affected by localized sources
(especially road dust).
As must be, the spatial trends for IP are intermediate to those for
its two components, FINE and COARSE. Because sulfate is the major compon-
ent of FINE, the spatial pattern of FINE agrees most nearly with that of
sulfur; these two variables exhibit the least urban/rural differences.
Sulfur at the rural sites averages 75% of sulfur at the urban sites, re-
flecting the large (air basin and synoptic) geographical scale of sulfate.
The spatial trends of COARSE closely resemble those of its predominant
constituent, crustal material (as traced by the elements Si, Al, and Ca).
On the average, urban sites are about 70% enriched over the rural sites in
crustal-shale and 90% enriched in crustal-limestone. The smooth concentric
pattern exhibited by COARSE and the crustal elements suggests that crustal
material originates more from widespread area sources of fugitive dust
than from concentrated industrial sources of fly ash.
The elements Pb (auto exhaust), V (fuel oil), Ti (paint pigment plant),
and Fe (iron/steel industry) all display erratic site-to-site fluctuations
and large urban/rural differences (averaging a factor of about three to
five). The dissimilarity in the spatial patterns for these elements versus
IP, FINE, and COARSE reflects the fact that none of the sources associated
with these elements are major contributors to IP.
To determine the effect of monitor site characteristics, the particu-
late data can be sorted with respect to both location in the metropolitan
area (urban vs. suburban vs. rural) and local environment (industrial vs.
commercial vs. residential vs. agricultural). For TSP, IP, FINE, COARSE,
and the eight elements, this sorting reveals systematic increases from rural
to suburban to urban locations but no substantial variation with type of
local environment.
The spatial scale of factors affecting particulate concentrations can
be investigated through interstation correlations (i.e. correlations of
daily particulate variations between pairs of sites). Interstation cor-
relations are extremely high for sulfur and high for FINE, reflecting the
-------
large spatial scale of sulfates and the importance of sulfates to FINE. The
correlations are moderately high for Si, Al, Ca, COARSE, and TSP, evidently
reflecting the common factor (area sources of fugitive dust) affecting all
these variables. Lead and iron exhibit moderate interstation correlations,
while titanium and vanadium demonstrate moderate to low correlations. One
would expect elements with low correlations to be associated with spatially
concentrated sources (e.g. point sources).
Background Concentrations and Particulate Transport
For days with predominantly northern and southern wind-flows, regional
upwind/downwind analysis is used to create particulate concentration profiles
across the St. Louis area. The most salient feature of these profiles is an
urban "bubble", with relatively lower particulate concentrations in upwind
and downwind rural areas.
The regional upwind/downwind studies and pollution rose analyses for
the rural sites indicate that the direct dowmwind transport effect of St.
3
Louis on rural areas is approximately 3 to 5 yg/m of aerosol, essentially
all in the fine size range. Because an individual rural location is not
always downwind of the center city, the effect of directly transported urban
aerosols on the annual mean at a rural site should be somewhat less, possibly
1 to 3 yg/m . Allowing also for "indirect" transport (from the sloshing of
air masses during periods of light or variable winds), we conjecture that
the total impact of the urban area on annual mean concentrations at the rural
sites is on the order of 2 to 5 yg/m3 of (basically fine) particulate matter.
The regional upwind/downwind analysis for sulfur suggests that much
(and possibly the majority) of sulfate in St. Louis is synoptic in nature.
Furthermore, the pollution rose analysis hints that important synoptic scale
factors may be the dense SOV emissions in the Ohio Valley toward the east
X
and high pressure systems moving 'up from the Gulf of Mexico (these high
pressure systems are known to be conducive to large scale sulfate episodes).
Pollution roses for IP, FINE, COARSE, and most of the elements tend
to be fairly isotropic (invariant with wind direction). This is expected
-------
because the two major components of IP should display rather weak wind-
directional dependencies; sulfates tend to be a large scale phenomenon, and
crustal material stems from widespread area sources of fugitive dust. Only
the pollution roses for titanium display very strong directional biases;
these roses implicate the paint pigment plant as the principal source of Ti.
Background particulate concentrations consist of aerosols from natural
sources and from man-made sources exterior to the St. Louis AQCR. Our es-
timate of background concentrations for St. Louis differs considerably from
estimates made by Record et al. (1976). Further analysis is needed to re-
solve this issue. The best guess, however, based on current information
3 3
would appear to.be approximately 35 yg/m for TSP and 20 yg/m for IP.
Temporal Patterns
Particulate concentrations in the St. Louis area tend to be highest
during the summer (June through September). The summer peak is strong for
FINE, moderately strong for IP, moderate for COARSE, and very weak for TSP.
Sulfa,te (sulfur), the major single component of FINE, displays a pronounced
summer maximum similar to that of FINE. Crustal material (as traced by Si,
Al, and Ca), the predominant source of COARSE, exhibits a moderate summer
maximum similar to that of COARSE.
On a weekly basis, the only salient feature in the aerosol data is a
significant weekend-weekday effect. For all twelve particulate variables
(TSP, IP, FINE, COARSE, and the eight elements), and for the averages over
all three types of site (urban, suburban, and rural), weekend mean concen-
trations are smaller than weekday mean concentrations. For most of the
particulate variables, the weekend-weekday differences are greatest for the
urban sites (typically about 20%), less for the suburban sites (typically
about 10%), and even less for the rural sites (typically about 5%). This
most likely reflects the fact that man-made sources are relatively more im-
portant in urban areas than rural areas.
Sulfur and FINE are the only variables that show rural weekend-weekday
percentage decreases about as large as the urban decreases. This again
-------
emphasizes the large (air basin or synoptic) scale of sulfates and other
secondary aerosols that constitute the major part of FINE. That substantial
(^ 15%) weekend-weekday differences do exist for sulfate indicates that a
significant part of sulfate is of air basin scale rather than synoptic
scale; we would expect synoptic sulfates to display little weekend effect
due to a time lag between emission changes and observed air quality changes.
The weekend-weekday differences for COARSE are fairly similar to
those of the crustal elements (Si, Al, and Ca). The significant weekend-
weekday differences for crustal material suggest that most of the crustal
material (at least for urban sites) stems from man-made sources. The
weekend-weekday differences are particularly large for Ca, indicating that
crustal limestone is more closely associated with conmerical and industrial
activities than crustal shale is.
Weekend-weekday differences are rather slight for TSP. This may
signify that TSP is more affected by wind-blown dust than is COARSE or IP.
The only particulate variable showing a pronounced diurnal pattern is
lead which demonstrates a distinct afternoon (noon-6 PM) minimum. There is
no obvious reason why lead uniquely exhibits this pattern. That crustal
material displays a different diurnal pattern (as well as different spatial
patterns) from lead hints that road dust is not the overwhelming source of
fugitive dust.
Decision-Tree Analysis of Particulate and Meteorological Data
The CART (Classification and Regression Trees) program developed at
Technology Service Corporation is used to conduct decision-tree studies.
The CART program offers several advantages over more conventional data-
analytic techniques such as multiple linear regression. CART is non-
parametric and nonlinear; the decision-trees are easily interpreted on
physical grounds; the CART results are evaluated by cross-validation tests
with independent data sets; and decision-trees tend to explain more variance
than multiple linear regression.
Using the CART program to relate daily concentrations of IP and FINE
to daily elemental concentrations yields very good statistical fits; the
-------
overall correlation coefficients are ^ 0.83 for FINE and ^ 0.81 for IP in
cross-validation tests. As expected, sulfur stands out as, by far, the
most important element explaining the variance in FINE. Also as expected,
sulfur and the crustal elements (Si, Al, and Ca) account for most of the
variance in IP.
The application of CART in relating particulate variables to eleven
meteorological variables produces less than excellent statistical fits.
Cross-validation tests with independent data sets yield correlation coef-
ficients on the order of 0.55 for FINE, IP, and S; 0.40 for COARSE, Si, Al,
Ca, and Fe; and less than 0.30 for TSP, Pb, V, and Ti. Despite the less
than excellent levels of percent variance explained, the forms of the
decision-trees do provide a reasonable and consistent picture of the type
of meteorology associated with high particulate levels.
Two basic themes stand out in the decision-tree classes for both high
sulfur and high FINE: prolonged stagnations and high temperatures. These
results are consistent with sulfate being the major component of FINE (worst-
case meteorology is similar for both variables); with sulfate being mostly
secondary aerosol (the worst-case meteorological classes are conducive to
secondary aerosol formation); and with sulfate and FINE exhibiting pro-
nounced seasonal patterns (prolonged stagnations and high temperatures are
basically summertime conditions). Some aspects of the decision-trees for
FINE point toward other man-made aerosols (both primary and secondary) as
the sources of the non-sulfate fraction of FINE.
The two meteorological conditions producing high levels of COARSE and
the crustal elements (Si, Al, and Ca) are dryness and low wind speeds. This
implicates anthropogenic sources of fugitive dust as the major contribitors
to the crustal material that dominates COARSE. The decision-trees also hint
that the small non-crustal component of COARSE is man-made.
The decision-tree results for IP are a mixture of the results for
FINE and COARSE. The decision-tree classes for high TSP point toward one
X
The square of the correlation coefficient represents the percent of
variance explained in the dependent variable.
-------
major factor -- dryness -- indicating that TSP tends to be dominated by
fugitive dust.
Relationship Between TSP and IP
On a day-to-day basis, TSP is not a very good predictor of IP, COARSE,
and (especially) FINE. Correlations between daily Hi-Vol and dichotomous
measurements at individual RAMS sites average 0.67 for TSP versus IP, 0.69
for TSP versus COARSE, and 0.47 for TSP versus FINE.
One way of characterizing the relationship between TSP and IP is to
study the properties of the IP/TSP ratio. The annual mean IP/TSP ratio,
averaged over all RAMS sites, is 0.61. At individual sites, the day-to-day
standard deviation away from the annual mean is ± 0.22 (t 36% of the annual
mean). The site-to-site standard deviation of the annual mean ratios,
however, is only ± 0.06 (± 10% of the network annual mean). Thus, for the
St. Louis RAMS stations, TSP tends to be a fair predictor of IP on an
annual mean basis even though it is a poor predictor on a daily basis.
The IP/TSP ratio does not display an obvious geographical pattern
within the St. Louis region nor does it show consistent variations accord-
ing to site characteristics. The IP/TSP ratio also lacks variation with
respect to day of the week. The ratio does, however, exhibit a seasonal
pattern, being higher in the summer months. The summer maximum in IP/TSP
(meaning that a greater fraction of the ambient aerosol is less than 20 ym
in the summer) is partly due to the pronounced sulfate peak during the
summer.
Decision-tree analyses relating the IP/TSP ratio to eight elemental
variables and to eleven meteorological parameters explain very little
variance in IP/TSP. The analyses do suggest, however, that days with high
IP/TSP ratios are days with high sulfate concentrations.
10
-------
1.0 INTRODUCTION
The Regional Air Monitoring System (RAMS) operated by EPA in St. Louis
has provided a wealth of data regarding ambient particulate matter in-and-
near a major metropolitan area. In order to obtain a basic understanding
of the data and to see what fundamental implications can be drawn from the
data, this report presents a series of statistical analyses conducted with
the RAMS particulate measurements. The analyses address not only particulate
mass concentrations, but also the size distribution and chemical composition
of the particulate matter. The specific purposes are to investigate the
spatial and temporal patterns of the aerosol in a major urban area; to de-
termine background particulate concentrations and examine particulate trans-
port relationships; to study the dependence of ambient particulate matter
on meteorology; to understand the relationship between Hi-Vol data and
dichotomous data; and, finally, to assess the contributions of various
source types to ambient aerosol concentrations.
1.1 THE RAMS PARTICULATE DATA
As shown in Figure 1.1, the RAMS air quality and meteorological net-
work consisted of 25 air monitoring stations (numbered 101 through 125).
Station 101 resided at the center of the network, and the remainder of the
stations formed four approximately concentric circles at radii of 4, 10, 20,
and 40 km (Hern and Taterka, 1977). Only ten of the RAMS stations (sites
103, 105, and 106 in the first circle; sites 108 and 112 in the second
circle; sites 115, 118, and 120 in the third circle; and sites 122 and 124
in the fourth circle) monitored particulate matter. The aerosol data from
these ten stations serves as the subject of this report.
Two techniques, the Hi-Vol sampler and the dichotomous sampler, were
used to measure particulate matter at the RAMS stations. The Hi-Vol measure-
ments are referred to as total suspended particulate mass (TSP) and
essentially represent the mass of particles less than approximately 50 urn
in diameter. The RAMS dichotomous samplers measure the mass of particles
less than approximately 2.4 urn in diameter as well as the mass of particles
11
-------
RAMS Parti oil ate Stations
e Other RAMS Stations
Figure 1.1 The RAMS monitoring network in St. Louis.
12
-------
in the approximate size range 2.4-20 ym (Dzubay, 1979).* Because EPA (Miller
et al., 1979) has recently stressed the special importance of inhalable par-
ticles'^ 15 ym in diameter) and fine particles U 2.5 ym in diameter), and
because the extensive set of dichotomous measurements is the most significant
novel feature of the RAMS data, this report emphasizes the dichotomous data
more than the Hi-Vol data. Even though the RAMS dichotomous samplers had an up-
per size cut-off of approximately 20 ym rather than 15 ym, we will refer to
the total dichotomous mass as inhalable particles (IP). The two size
fractions, less than 2.4 ym and 2.4 to 20 ym, will be called fine particle
mass (FINE) and coarse particle mass (COARSE), respectively.
The RAMS Hi-Vol samplers were run for 24 hours, every third day. Al-
though the dichotomous samplers essentially measured 12-hour averages on a
continuous basis*,* most of the analyses performed herein use only 24-hour
averaged dichotomous data. The RAMS network operated from mid-1975 to
early 1977, but this report includes data only for the year 1976.
The RAMS dichotomous data (both the fine and coarse fractions) were
analyzed by X-ray spectroscopy to determine the concentrations of 28 elements
(Goulding et al., 1978). In this report we will use data for eight of the
elements that are especially important in determining the nature of parti-
culate sources. These eight elements, and the source types in St. Louis
that the elements are useful in identifying, are listed in Table 1.1.
This study also utilizes data on eleven meteorological parameters.
These parameters are defined in Table 1.2.
1.2 DATA QUALITY ANALYSIS
As a preliminary to this study, Eldon et al. (1979) performed a data
quality analysis on the 1976 RAMS Hi-Vol and dichotomous data. The data
quality analysis consisted of three parts. First, the five highest and
*It should be noted that the upper size cut-off at 20 urn may be sensitive
to wind speed; specifically, the collection efficiency of particles near
the upper size cut-off may decrease with increasing wind speed.
**At two of the stations (103 and 105), 6-hour samples rather than 12-hour
samples were collected. Also, during an intensive study period in the
summer of 1976, sites 103, 105, and 112 had 2-hour samples while most of
the other sites had 6-hour samples.
-------
TABLE 1.1 TRACE ELEMENTS INCLUDED IN THIS STUDY.
ELEMENTS
Sulfur (S)
Silicon (Si)
Aluminum (Al)
Calcium (Ca)
Lead (Pb)
Vanadium (V)
Titanium (Ti)
Iron (Fe)
SOURCES IN ST. LOUIS THAT CAN BE IDENTIFIED BY THE
ELEMENTS*
(In order of decreasing importance as to usefulness
of the element in identifying the source. Under-
lined sources especially important.)
Sulfate aerosol (basically a secondary aerosol from
gaseous SOX emissions). Note: If one assumes all
sulfur is (NH4)2S04, then [sulfate] = 4.1[sulfur];
if one assumes a representative mixture of sulfates,
then [sulfate] * 3.9[sulfur](White and Roberts,
1977).
Soil dust (shale type), coal combustion.
Soil dust (shale type), coal combustion.
Cement dust, soil dust (limestone type), coal
combustion.
Auto exhaust.
Fuel oil combustion.
Paint pigment plant, soil dust, coal combustion.
Iron and steel industry, soil dust, coal combustion.
"in compiling this table, we relied mostly on a paper
but also considered the work of Miller et al. (1972),
Gatz (1975), and Kowlaczyk et al. (1978).
-------
TABLE 1.2 METEOROLOGICAL PARAMETERS INCLUDED
IN THIS STUDY.
SYMBOL
DESCRIPTION
UNITS
LOCATION
AV WD SP Average Uind Speed: daily 24-hour m/sec Individual RAMS
vector average surface wind speed Sites
WIND DIR Wind Direction: most common dir- no units Individual RAMS
ection (8-point compass) from Sites
which the hourly average surface
wind blew during each day (24-
hour period).
MIXHT AM Morning Mixing Height: six AM in- meters Lambert Field
version base height as estimated
from noon temperature sounding
and hourly ground temperature
readings.
MIXHT PM Evening Mixing Height: six PM in- metor: Lambert Field
version base height as estimated
from noon temperature sounding
and hourly ground temperature
readings.
MIN TEMP Minimum Temperature: lowest or Lambert Field
hourly average surface tempera-
ture reading for the day.
MAX TEMP Maximum Temperature: highest o,. Lambert Field
hourly average surface tempera-
ture reading for the day.
*REL HUM Relative Humidity: daily (24- % Lambert Field
hour) average of hourly relative
humidity readings.
PRECIPIT Precipitation: whether or not no units Lambert Field
there was measureahle precipi-
tation during the day (0 = no,
1 = yes).
»DAYS PP Days Since Previous Precipitation: no units Lambert Field
the number of days since the most
recent measurable precipitation,
irrespective of rain on day in
question (i.e. variable ranges
from 1 to »).
D7AM PRS Pressure Difference: today's 7 AM millibars Lambert Field
surface barometric pressure minus
yesterday's 7 AM pressure.
#OY CALM Days of Calm: the number of con- no units Lambert Field
secutive dalm days (from 0 to °°).
A calm day is one where vector
average wind speed is below 3
meters/second.
15
-------
five lowest values for each variable (i.e. TSP, FINE, COARSE, fine S, coarse
S, fine Si, coarse Si, etc.) were printed out along with several preceed-
ing and following measurements at the same site, and with simultaneous
measurements at nearby locations. These values were examined manually for
consistency and a few erroneous readings (e.g. zeros that should have been
blanks and some inexplicably very high values) were deleted.
Second, scatterplots of IP versus TSP at individual locations were
prepared. These scatterplots revealed that 10 to 15 TSP readings were
outliers. Also, it was noted that TSP samplers did not run for the full 24
hours on twenty occassions (Nelson, 1979). The lists of extreme outliers
and data with short sampling times overlapped considerably, both were
eliminated from the data set.
Third, a review was conducted of summary statistics at each site
(e.g. annual means, average percent fine, etc.), inter-species correlations
(e.g. Al vs. Si, S vs. FINE, etc.) at each site, and inter-site correlations
for individual aerosol variables to get an overall impression of data
quality. This review indicated that the data were generally of good quality
despite the existence of the few erroneous values noted earlier.
1.3 SUMMARY OF ANNUAL MEANS
Before proceeding with the statistical analyses of subsequent chapters,
it is wothwhile to get a feel for the magnitude of the aerosol variables by
examining summary statistics. Table 1.3 lists annual means*for the twelve
aerosol variables considered in this study. Appendix A presents more de-
tailed summaries of annual and monthly statistics for both the particulate
and meteorological variables.
As indicated in Table 1.3, annual mean TSP concentrations range from
53 yg/m3 to 96 yg/m3 and average 79 yg/m3 among the sites. To ascertain
the attainment status of the sites with respect to the present National
Ambient Air Quality Standards (NAAQS) for TSP, one must consider annual
geometric means and second highest daily maximum concentrations. Data pre-
sented in Appendix A indicate that the primary NAAQS for the annual geometric
*Note that arithmetic means rather than geometric means are used throughout
this report.
16
-------
TABLE 1.3 ANNUAL ARITHMETIC MEAN PARTICULATE CONCENTRATIONS.
ANNUAL MEAN CONCENTRATIONS (ug/m3) FOR 1976
TSP
IP
FINE
COARSE
Sulfur
Silicon
Aluminum
Calcium
Lead
Vanadium
Titanium
Iron
Site
103
96
60
(47%)
28
32
3.5
(83%)
4.8
( 7%)
1.5
(12%)
3.3
( 5%)
.68
(79%)
.007
(46*)
.16
(14%)
1.8
(13%)
Site
105
88
46
(SOX)
23
23
3.1
(88%)
3.6
( 8%)
1.1
(16%)
2.7
( 5%)
.81
(78%)
.006
(60%)
.21
(16%)
1.2
(19%)
Site
106
77
48
(50%)
24
25
3.5
(88%)
4.2
(10%)
1.1
(14%)
2.5
( 5%)
.87
(76%)
.006
(58%)
.53
(15%)
1.4
(20%)
Site
108
80
48
(50%)
24
24
3.3
(89%)
4.1
( 8%)
1.2
(13%)
2.9
( 5%)
.75
(80%)
.010
(51%)
.13
(16%)
1.7
(17%)
Site
112
79
44
(49%)
21
23
2.9
(89%)
4.3
( 8%)
1.3
(16%)
2.0
( 6%)
.89
(79%)
.007
(55%)
.23
(16%)
1.1
(17%)
Site
115
60
38
(52%)
19
19
2.8
(92%)
3.5
( 9%)
0.9
(14%)
1.7
( 8%)
.40
(83%)
.005
(59%)
.09
(17%)
0.8
(19%)
Site
118
66
35
(51%)
18
17
2.7
(9U)
2.9
( 9%)
0.8
(14%)
2.1
( 9%)
.37
(79%)
.003
(54%)
.17
(17%)
0.7
(19%)
Site
120
53
37
(55%)
20
17
2.7
(91%)
2.7
( 8%)
0.8
(17%)
1.5
( 6%)
.65
(81%)
.003
(58%)
.09
(16%)
0.6
(17%)
Site
122
55
33
(51%)
17
17
2.5
(92%)
3.0
( 7%)
0.8
(11%)
1.9
( 6%)
.21
(86%)
.002
(57%)
.06
(14%)
0.6
(18%)
Site
124
53
29
(56%)
16
13
2.4
(93".}
2.2
(20%)
0.6
(22%)
1.0
(24%)
.19
(82%)
.002
(52%)
.04
(15°.)
0.4
(31%)
AVERAGE
ALL SITES
79
42
(50%)
21
21
2.9
(90%)
3.5
( 9%)
1.0
(15%)
2.2
( 8%)
.58
(80")
.005
(55%)
.17
(16%)
1.0
(19%)
*Nunbers in parentheses represent percent of the dichotomous concentration that is fine.
-------
mean (75 yg/m3) was exceeded in 1976 at site 103 (89 yg/m3) and site 105
(82 yg/m3). The primary NAAQS for daily maximal concentrations (260 yg/m )
was not exceeded at any of the RAMS sites in 1976.
The concentrations of IP average 42 yg/m among the RAMS sites. Approx-
imately half of IP (slightly less than half at center-city sites, slightly
more than half at rural sites) consists of fine particles. Sulfur and lead
occur predominantly among the fine particles. Silicon, aluminum, calcium,
titanium, and iron are found predominantly among the coarse particles.
Vanadium occurs nearly equally in COARSE and FINE.
1.4 CHEMICAL ELEMENT BALANCE
One of the objectives of this study is to perform a qualitative assess-
ment of the sources contributing to ambient particulate matter based on the
spatial/temporal patterns in aerosol mass and composition, the dependence of
aerosol mass and composition on meteorology, particulate transport relation-
ships, etc. A more direct and quantitative empirical technique for assessing
source contributions is the chemical element balance (tracer element) method
(Miller et a!., 1972; Friedlander, 1973; Gatz, 1975; Kowalczyk, 1978). Work
on the St. Louis RAMS data with the chemical element balance method is being
carried out separately by other researchers (Dzubay, 1979; Gordon, 1979;
Hopke, 1979). Although chemical element balance is not within the planned
scope of this study, a brief review of preliminary conclusions reached by
the tracer element technique is useful for placing our results in better
perspective.
Dzubay (1979) performed a tracer element analysis with the RAMS dicho-
tomous data for July and August 1976. He included six types of sources in
his analysis: ammonium sulfate, crustal shale (soil, road dust, quarry dust,
fly ash, etc.), crustal limestone (calcium-rich soil, cement manufacturing,
vehicle generated cement dust, etc.), motor vehicle exhaust, iron/steel
industry, and paint pigment. The results are summarized in Figure 1.2.
The most important sources of IP were found to be ammonium sulfate, which
constituted 59% of FINE and 35% of IP, and the crustal components (mostly
shale-type), which constituted 83% of COARSE and 43% of IP.
18
-------
Average Concentration at 10 RAMS sites (pg/m )
en
en
ro
CD
en
co
o
co
en
-P*
o
en
en
0
en
en
(O
c.
n>
cr to co
t< _.. O
rt C
o n> -5
3- w o
-o o
o -s o
O) 3
' _ rt
C -J
ro _!.
<< cr
fD C
3 O» rt
n> 3 -"
3 Q. O
rt 3
CT c
O> td Q)
C <
cu u> n>
n o»
I-'IQ
IO 0)
^^ *^J O
O Ol
N O
c <
cr at (t>
BJ in -J
» Q. rt
O) 3-
o
O
co
rt
O
-5
fD
O)
l/l
O
Cu
CO
3"
fD
o
-J
ft)
rt
O
3
n>
o
rt
3"
fD
-5
c:
3
7T
3
O
3 >_,
CO
-------
Dzubay's results would change slightly if the whole year, rather than
July and August were considered. The greatest change would involve
sulfates which show a distinct summer maximum. From calculations using
data for the entire year of 1976, we find that ammonium sulfate constitutes
53% (rather than 59%) of FINE and 29% (rather than 35%) of IP. We also
estimate that, for the entire year, crustal material still constitutes
about 83% of COARSE but around 47% (rather than 43%) of IP.
The most significant question left unanswered by Dzubay's analysis
concerns the specific origins of the crustal material: Is the crustal material
basically related to soil, road dust, quarries, cement manufacturing, fly
ash, or what? Although the present report does not resolve the question
completely, we do reach one important conclusion relevant to this issue
the crustal material appears to be predominantly related to various fugitive
dust sources rather than to fly ash. The strongest evidence supporting
this conclusion is found in Chapter 2, where ubiquitous area sources
(e.g. fugitive dust sources) rather than industrial point sources are im-
plicated as the originators of the crustal material, and in Chapter 5, where-
dryness is found to be the meteorological condition most closely related to
high ambient concentrations of crustal elements.
1.5 ORGANIZATION OF THE REPORT
This report is organized in seven chapters. The present chapter pro-
vides an introduction to the RAMS participate data. Chapter 2 discusses
the spatial aspects of particulate concentrations in St. Louis based on
geographical patterns in the data, the relationship of aerosol concentra-
tions to site characteristics, and the magnitude of interstation correla-
tions. In Chapter 3, regional upwind/downwind analyses and pollution
roses are used to study background particulate concentrations and particu-
late transport. Chapter 4 examines temporal patterns -- seasonal, weekly,
and diurnal. Chapter 5 presents the results of pattern-recognition/
decision-tree analyses relating aerosol concentrations to meteorological
and elemental variables. In Chapter 6, the relationship between Hi-Vol
and dichotomous sampler data is studied. Finally, Chapter 7 draws from
several parts of the report to synthesize and summarize conclusions re-
garding the sources of ambient particulate matter in St. Louis.
20
-------
2.0 SPATIAL PATTERNS OF ST. LOUIS PARTICULATE DATA
The dichotomous and Hi/Vol data from the RAMS network provide a good
opportunity to investigate the spatial patterns of ambient participate matter
in-and-near a large metropolitan area. Specific issues that can be addressed
include the geographical distribution of ambient particulate matter, the
variations among site types, and the spatial scale of aerosol sources.
2.1 GEOGRAPHICAL DISTRIBUTION OF AMBIENT PARTICULATE MATTER
Figures 2.1 through 2.3 illustrate the geographical variation of TSP,
*
IP, and FINE data, respectively. Concentrations of all three parameters
are highest at station 103 and decrease in a fairly concentric manner out to
the rural stations (122 and 124).
Because the aerosol spatial distribution appears to be concentric, it
**
is instructive to plot the data versus distance from station 103. Such
plots are presented in Figure 2.4 for all 12 particulate variables considered
in this study. To facilitate reading and interpreting the graphs, the vari-
ables are divided into three groups: TSP, IP, COARSE, and FINE; S, Si, Al,
and Ca; and Pb, V, Ti, and Fe. All the plots are normalized with respect to
the concentration at station 124 (the site experiencing the lowest average
concentration for all variables except TSP).
Figure 2.4 reveals that TSP, IP, COARSE, and FINE all increase rather
monotonically from the rural stations to station 103. This implies that the
spatial pattern of the data is indeed very concentric about station 103. One
major reason for the smooth concentric pattern is that the RAMS sites tend to
be isolated from strong localized sources such as heavily used roads, other
intense fugitive dust sources, or industrial point sources (PEDCO, 1978).
With the influence of local sources minimized, the gradual increase in
ambient particulate matter from rural areas to the center city becomes a
dominant spatial feature in the data.
*
Similar maps for eight particulate elements (S, Si, Al, Ca, Pb, V, Ti, and
Fe) are presented in Appendix B.
**
We also tried plotting the data versus distance from station 101, the center
of the entire RAMS network. The plots based on distance to station 101 were
not quite as smooth or monotonic as those based on distance to station 103;
this apparently indicates that the center of the St. Louis particulate bulge
is closer to station 103 than it is to station 101.
21
-------
MississippiJLiver
Figure 2.1 Geographical distribution of arithmetic mean TSP mass concentration
in the St. Louis area (yg/m3).
22
-------
Mississippi River
Figure 2.2 Geographical distribution of arithmetic mean IP mass concen-
tration in the St. Louis area
23
-------
Mississippi River
Figure 2.3 Geographical distribution of arithmetic mean FINE mass con-
centration in the St. Louis area (yg/mj).
24
-------
Concentration Normalized to Site 124 Concentration Normalized to Site 124
fD
ro
ro i
-e=> o
fD O
-S -h
(/> O
0-1
O> N
3 ft)
O Q.
fD
O <-h
3
O
10 C
r+ '
0) Oi
O
3 n
o
I =S
o o
CO fD
3
rt-
o>
o
3
CO
fD
(/»
XJ
fD
O
rt-
Oi
O
3
Concentration Normalized to Site 124
-------
The increase in TSP from station 124 to station 103 is by a factor of
1.8. The factor for IP (2.1) is slightly larger and is, of course, inter-
mediate to the factors for COARSE (2.5) and FINE (1.8) which together com-
prise IP.
Sulfur, silicon, aluminum, and calcium also display nearly monotonic
increases (i.e. a concentric pattern). Sulfur (or sulfate), the major
component of FINE (see Chapter 1), increases by a factor of only 1.5 from
site 124 to 103. This likely reflects the large spatial scale of sulfates;
a major portion of sulfates can arise from "background" sources such as
power plants. As expected based on the chemical element balance results in
Chapter 1, the crustal elements - especially the crustal-shale elements
(Si and Al) - exhibit patterns similar to that of COARSE. The greater
urban/rural differences for Ca (as compared to COARSE, IP, or TSP) signify
that crustal-limestone contributions are relatively larger in the center .
city than at rural locations. That the crustal elements show rather smooth
spatial gradients suggests that the crustal material stems more from wide-
spread area sources of fugitive dust rather than concentrated industrial
sources of fly ash. As discussed in the next paragraph, elements associated
with point sources tend to display erratic spatial patterns.
The elements Pb, V, Ti, and Fe exhibit large fluctuations among the
sites. The erratic patterns suggest that the sources of these elements are
point sources or areas of concentrated emissions (e.g. dense traffic areas).
For example, the spatial peaks in some of these elements can be readily ex-
plained as follows: Ti peaking at 106 due to a nearby paint pigment factory
(Dzubay, 1979); Fe peaking at 103 and 108 due to nearby steel mills (Dzubay,
1979); and Pb peaking at 112 because that site is closest to heavily
travelled streets (PEDCO, 1978). Because the large site-to-site fluctuations
of these elements are not apparent in the TSP or IP data, we conclude that
none of the source categories associated with these elements Pb (auto-
mobile exhaust), V (fuel oil),Ti (paint pigment plant), and Fe (iron/steel
industry) - constitute a single predominant source of TSP or IP throughout
St. Louis. Rather, TSP and IP evidently show gradual spatial gradients be-
cause one large fraction of particulate mass stems from ubiquitous sources
of fugitive dust, and because another significant fraction consists of
26
-------
sulfates and other secondary aerosols (which tend to exhibit slight spatial
gradients due to the mixing and transport that occurs in the time required
for secondary aerosol formation).
2.2 EFFECTS OF SITE CHARACTERISTICS
It is of interest to determine if the spatial variations in ambient
particulate matter can be explained, in part, by site characteristics. In
examining this question, we will investigate two types of site character-
istics: location with respect to the metropolitan area (e.g. urban vs. sub-
urban vs. rural) and local environment (e.g. industrial vs. commercial vs.
residential vs. agricultural). We have chosen to classify locations within
the first two rings of RAMS stations (sites 103, 105, 106, 108, and 112) as
urban, locations in the third ring of RAMS stations (sites 115, 118, and 120)
as suburban, and locations in the outer ring of RAMS stations (sites 122 and
124) as rural. The classification as to local environment is based on de-
tailed site descriptions compiled by PEDCO (1978).
Table 2.la indicates the distribution of the RAMS particulate stations
with respect to site characteristics. Because of the extremely small number
of stations involved, there is no hope of obtaining a statistically defini-
tive statement of how particulate levels vary among site types; in fact,
half of the site characteristic matrix is barren of stations. Rather, we
will attempt only to describe the main qualitative features in the data
sorted according to site characteristics.
Tables 2.1b through 2.Id summarize annual mean concentrations of TSP,
IP, and FINE, averaged for all sites of each type. Two main tendencies ap-
pear in these tables: (1) particulate concentrations increase systematically
from rural to suburban to urban locations, and (2) particulate concentrations
do not seem to vary substantially with type of local environment. This re-
sult reemphasizes a conclusion of the previous section; specifically, we
again find that the principal spatial feature of the RAMS data is the con-
centric decrease in concentrations from urban to rural locations.
We have also examined how the concentrations of the eight trace elements
vary with site type (see Appendix C). For all eight elements, we find the same
two tendencies apparent in Table 2.1 (i.e. systematic increases from rural to
The reader is referred to Figure 1.1 for a map of the RAMS network.
27
-------
TABLE 2.1 AVERAGE PARTICULATE CONCENTRATION
AS A FUNCTION OF SITE TYPE.
Urban
Suburban
Rural
Urban
Suburban
Rural
Urban
Suburban
Rural
Tabl
Industrial
0
1
0
Table Z.lb
Industrial
53
Table Z.lc
Industrial
37
Table 2. Id
Industrial
20
e 2. la Number of Sites
Commercial Residential
1 2
0 0
0 0
Annual Arithmetic Mean TSP
Commercial Residential
88 78
Annual Arithmetic Mean IP
Commercial Residential
46 46
Agricultural
2
2
2
All
5
3
2
o
(yg/m ).
Agricultural
88
63
54
All
84
59
54
(yg/m3).
Agricultural
54
36
31
All
49
37
31
o
Annual Arithmetic Mean FINE (yg/m ).
Commercial Residential
23 22
Agricultural
26
19
16
All
24
19
16
28
-------
suburban to urban locations, and lack of substantial variations with respect
to local site environment).
Table 2.2 summarizes the urban/suburban/rural differences for all
twelve variables. Many of the data features noted in the previous section
are again apparent in Table 2.2. Specifically, sulfur shows the least
spatial variation; FINE shows the second least spatial variation; patterns
are similar for COARSE, Si, Al, and Ca; and Pb, V, Ti, and Fe show the
greatest urban/rural differences.
2.3 SPATIAL SCALE OF PARTICULATE PHENOMENA
The spatial scale of factors affecting particulate concentrations can
be investigated through interstation correlations (i.e. correlations of day-
to-day particulate variations between pairs of sites). Table 2.3 summarizes
the results of three types of correlation/regression analysis applied to the
24-hour particulate data: (1) the average correlation between each pair of
RAMS sites, (2) the average correlation obtained in a multiple regression of
each site against its three nearest neighbors, and (3) the average correlation
obtained in a multiple regression of each site against its six nearest neigh-
bors. The particulate parameters in Table 2.3 are ordered according to the
magnitude of interstation correlations.
As evidenced by Table 2.3, sulfur exhibits extremely high interstation
correlations (i.e. all sites exhibit very similar day-to-day fluctuations).
This is expected because sulfur basically represents sulfate, a secondary
aerosol that tends to show spatially uniform fluctuations on an air basin
scale and sometimes even on a synoptic scale. FINE, which consists of ap-
proximately 53% sulfate (see Chapter 1), also demonstrates high interstation
correlations, although not as high as sulfate (i.e. sulfur). Evidently,
the other components of FINE -- other secondary aerosols and primary fine
particles -- involve more localized phenomena than does sulfate, leading to
lower interstation correlations.
As expected, IP displays a level of correlation intermediate to FINE
and COARSE. The variables Si, Al, Ca, COARSE, and TSP all show about the
29
-------
CO
o
TABLE 2.2 RURAL VS. SUBURBAN VS. URBAN CONCENTRATIONS OF ST. LOUIS PARTICULATE DATA.
Urban
Suburban
Rural
Urban
Suburban
Rural
Urban
Suburban
Rural
TSP
84 (1.6)
59 (1.1)
54
S
3.3 (1.3)
2.7 (1.1)
2.5
Pb
0.80 (4.0)
0.47 (2.4)
0.20
IP
49 (1.6)
37 (1.2)
31
Si
4.2 (1.6)
3.0 (1.2)
2.6
V
0.007 (3.5)
0.004 (2.0)
0.002
COARSE
25 (1.7)
18 (1.2)
15
Al
1.26 (1.8)
0.85 (1.2)
0.69
Ti
0.26 (5.2)
0.12 (2.4)
0.05
FINE
24 (1.5)
19 (1.2)
16
Ca
2.7 (1.9)
1.7 (1.2)
1.4
Fe
1.44 (2.7)
0.69 (1.3)
0.53
*First number is concentration in (yg/m3); second number is ratio to rural.
-------
TABLE 2.3 IMTERCORRELATIONS FOR ST. LOUIS
PARTICULATE DATA
AVERAGE CORRELATION COEFFICIENTS
Univariate Regression Multiple Regression Multiple Regression
of Each Site Versus of Each Site Versus of Each Site Versus
All Others Its Three Nearest Its Six Nearest
PARAMETER
Sulfur
FINE
IP
Silicon
Aluminum
Calcium
COARSE
TSP
Lead
Iron
Titanium
Vanadium
0.90
0.81
0.71
0.68
0.61
0.56
0.56
0.58
0.40
0.44
0.24
0.19
Neighbors
0.95
0.90
0.86
0.81
0.77
0.77
0.77
0.71
0.70
0.64
0.53
0.36
Neighbors
0.96
0.91
0.90
0.83
0.80
0.80
0.80
0.75
0.76
0.74
0.65
0.47
same, moderately high levels of interstation correlation. Because Si,
Al, and Ca are basic tracers for crustal material, this result supports in-
dications elsewhere in this report that the majority of COARSE and TSP con-
sists of crustal material. That the interstation correlations for the
crustal elements are much lower that those for sulfates implies that the
crustal phenomena are more localized that sulfate phenomena; that these
correlations are much higher than for industrial or point source tracers
(e.g. V, Ti, and Fe) suggests that the crustal material arises more from
31
-------
area sources of fugitive dust than from industrial or point sources of fly
ash.
Lead and iron exhibit moderate interstation correlations, while
titanium and vanadium demonstrate moderate to low interstation correlations.
We expect elements with low correlations to be associated with spatially
concentrated sources (e.g. point sources). Point sources lead to low inter-
station correlations because there is less tendency for simultaneous high
concentrations at all stations (wind direction with respect to the point
sources, in addition to overall dispersive conditions, becomes a key meteor-
ological condition).
The larger scale nature of FINE relative to IP, and of IP relative to
TSP, may have significant implications with respect to particulate monitor-
ing. Specifically, the scale of representativeness of particulate monitors
should be greatest for FINE, less for IP, and least for TSP. This implies
that, in characterizing the spatial and temporal patterns of air quality,
FINE requires fewer monitors than IP which in turn requires fewer monitors
than TSP.
32
-------
3.0 ANALYSIS OF BACKGROUND CONCENTRATIONS
AND PARTICULATE TRANSPORT
The relationship between participate concentrations and wind direction
can provide information about background aerosol concentrations for St. Louis
and about particulate transport phenomena that occur in-and-near St. Louis.
The wind direction dependence of aerosol concentrations can also identify
significant individual sources of particulate matter. This chapter uses two
techniques to analyze the relationship between particulate levels and wind
direction: (1) regional upwind/downwind analysis and (2) wind rose and
pollution rose analysis.
Before starting the technical discussion of this chapter, it is worth-
while to digress a moment and discuss the definition of "background" concen-
tration. For our purposes, background aerosol concentration refers to the
aerosol concentration that would exist if all man-made sources in the St.
Louis AQCR* were eliminated; background therefore includes contributions
from natural sources as well as from man-made sources outside the St. Louis
AQCR. Background concentrations tend to be greater than "natural background"
but less than concentrations measured at rural locations near St. Louis
(because rural aerosol concentrations include particulate matter generated
by local man-made sources as well as particulate matter transported from
the urban area).
3.1 REGIONAL UPWIND/DOWNWIND ANALYSIS
Regional upwind/downwind analysis is a useful technique for identify-
ing particulate sources, estimating background concentrations, and assessing
particulate transport from an urban area. Given a linear set of monitoring
sites and data corresponding to wind flow parallel to the line of sites,
large sources might be located by noting abrupt concentration increases
from one site to the next. Upwind of the source, the concentration naturally
should be lower than immediately downwind of it. Upwind/downwind analysis
can also help estimate regional background levels of aerosols, since
*
Air Quality Control Region
33
-------
the readings at a rural site, if taken when that site is upwind of the city,
should be near the background level. Particulate transport from the urban
area can be estimated by the difference between downwind rural concentra-
tions and upwind rural concentrations.
Of the ten RAMS sites equipped with aerosol samplers, six lie approxi-
mately in a straight, north-south line through the center of the city.
These six sites are 122 (the northernmost rural site), 108, 103, 105, 118,
and 124 (the southernmost rural site). Data for these six sites, under pre-
dominantly northern and southern flows, serves as the subject of this
section. The methodology for calculating concentration cross-sections under
northerly and southerly flows 'is explained in Appendix E.
Figures 3.1 through 3.4 are concentration cross-sections for TSP, IP,
FINE, and COARSE. Each figure contains two graphs, one the averages for
wind'predominantly from the north, the other the averages for wind predom-
inantly from the south. Average concentrations at the central urban site
(103), the upwind site (122 for north, 124 for south), and the downwind site
(124 for north, 122 for south) are presented in Table 3.1 along with downwind-
upwind differences and urban-upwind differences.
The main feature in the cross-sections is the excess of the urban con-
centrations over the rural concentrations. This feature is demonstrated by
the urban "bubble" effect evident in Figures 3.1 through 3.4 and by the
significant urban-upwind differences in Table 3.1. The downwind-upwind
differences are comparatively minor, indicating that particulate transport
from the urban area to rural areas is not an extremely important phenomenon.
Although it is difficult to reach definitive conclusions because we are
dealing with small differences between large numbers (and therefore the
errors tend to be large), the downwind-upwind differences in Table 3.1 gen-
erally indicate that the direct downwind transport effect of St Louis on
rural areas is approximately 5 yg/m3 of aerosol, predominantly in the fine
size range.
Because an individual rural location is not always downwind of the
center city, the effect of directly transported urban aerosols on the arm^l
mean at a rural site should be less than 5 yg/m3, possibly on the order of
34
-------
80-,
60-
:£ 40 4
ra
§ 20
c
o
10
20
30
60
80
40 50 60 70
Distance Downwind (South) in Km
90
100
110
120
Figure 3.la TSP concentration cross-section, wind from north.
Distance Downwind (North) in Km
Figure 3.1b TSP concentration cross-section, wind from south.
Figure 3.1 Regional upwind/downwind analyses for TSP.
35
-------
Concentration (pg/m )
Concentration (vig/m )
n>
CO
ro
70
n>
IQ
o'
3
o>
CO
cr>
CL
o
3
O-
CU
3
Cu
CD
I/)
31
-s
c
-s
ro
CO
ro
or
O
O
o
(D
3
O
3
O
O
in
o
ri-
3
O-
-h
I
O
c:
C
-5
fD
CO
ro
O
O
3
O
(T>
3
ri-
O
3
O
O
1/1
fD
O
O
3
3
ex
-h
o
3
O
-------
CD
C
0)
o
c
o
o
10
30 40 50 60 70 80
Distance Downwind (South) in Km
i
90
100
110 120
Figure 3.3a FINE concentration cross-section, wind from north.
10
Distance Downwind (North) in Km
Figure 3.3b FINE concentration cross-section, wind from south.
Figure 3.3 Regional upwind/downwind analyses for FINE.
37
-------
Concentration (yg/m )
Concentration (yg/m )
o
CO
CO
10
c
n>
CO
yo
n>
o
3
C
1.
3
O.
Q.
i
3
O.
cu
3
-h
o
-5
O
§
73
CO
-s
co
t
a-
o
o
CO
O
o
3
O
n>
3
ft-
O
3
O
O
01
fD
O
rt
O
3
-h
O
i/)
O
cz
co
-5
fD
CO
O>
O
o
3=
co -
o
o
3
O
n>
3
rt
-S
o
3
O
-S
O
(/I
(/)
I
in
n>
o
o
3
S
3
-S
O
3
O
-s
-------
TABLE 3.1 SUMMARY OF UPWIND/DOWNWIND ANALYSES FOR TSP, IP, FINE, AND COARSE.
AEROSOL CONCENTRATIONS (yg/rrT)
Wind Direction
Average Concentration
122
103
124
*
Downwind-Upwind
^ Absolute Change
Percent Change
**
Urban-Upwind
Absolute Change
Percent Change
TSP
N S
44.9 59.0
71.5 89.2
45,4 56.3
0.5 2.7
1% 5%
26.6 32.9
59% 58%
IP
N S
25.7 39.8
46.7 54.8
27.2 29.3
1.5 10.5
6% 36%
21.0 25.5
82% 87%
FINE
N S
12.1 21.9
22.8 28.0
16.1 16.6
4.0 5.3
33% 32%
10.7 11.4
88% 69%
COARSE
N S
13.6 17.9
23.9 26.8
11.1 12.7
-2.5 5.2
-18% 41%
10.3 14.1
76% 111%
**
"Downwind-Upwind" is difference between downwind and upwind averages.
r
"Urban-Upwind" is the difference between the site 103 reading and upwind rural site concentration.
-------
1 to 3 yg/m3. There exists, however, what might be called an "indirect"
transport effect that cannot be measured by the regional upwind/downwind
analysis. The indirect effect, which should be particularly important for
secondary aerosols, involves sloshing of air masses in the AQCR during
periods of light or variable winds. Making an allowance for indirect trans-
port, we conjecture that the total impact of the St. Louis urban area on ^
annual mean concentrations at the rural sites is on the order of 2 - 5 yg/m
of (mostly fine) particulate matter.
Figures 3.5 through 3.7 illustrate concentration profiles for sulfur,
lead, and silicon. The first two elements, which are associated with FINE,
do show a positive transport effect downwind of the city for both northerly
and southerly flows. Silicon, which is associated with crustal material
and COARSE, does not show a consistent transport effect. These results
reinforce our above conclusion that the urban plume basically contains
fine particles. The enrichment of the FINE/COARSE ratio in the urban plume
compared to the center-city (where FINE and COARSE are about equal) likely
results from two factors: (1) coarse particles settle during transport and
(2) some fine particles (secondary aerosols) are produced during transport.
The urban versus rural concentrations of sulfur (or sulfate) deserve
special comment. The annual mean concentration of ammonium sulfate (4.125 x^
sulfur) range only from 9.9 yg/m3 at the lowest RAMS site (124) to 14.4 yg/m
at the highest RAMS site (103). That the rural concentrations are 70% of
the urban concentrations implies that the sulfate phenomenon is air basin
in scale if not synoptic in scale. This large spatial scale is rather
typical for sulfates (EPA, 1975; Frank and Posseil, 1976) and occurs because
sulfates are mostly secondary aerosols slowly produced from SOX emissions
(most of which arise from nonurban, tall-stack sources, i.e. power plants).
The upwind/downwind analysis (averaged for northerly and southerly flows)
yields the following averages for ammonium sulfate : upwind = 8.9 yg/m ,
center-city (103) = 12.3 yg/m3, and downwind = 11.0 yg/m . This illustrates
transport of sulfate from the urban center to downwind sites. More import-
antly, the upwind concentration being 70% of the center-city concentration
40
-------
Concentration (yg/m )
Concentration (pg/m )
(D
CO
en
70
fD
IQ
o
3
QJ
C
I
Q.
O
3
Cu
3
CU
n>
o
-s
-h
-$
fD
CO
en
cr
oo
c ->
ro
o
co
o
O
O
3
O
a>
3
r+
-j
OJ
o
3
O
O
(/>
V)
l
to
0)
O
o
n>
o en
O C"*
S
l.
Q- a
00 _
-h
o
c.
<-t-
o
o
ro
o
ro
co
I
CQ
-s
n>
co
en
o
o
3
O
fD
3
r+
Cu
O
3
n
-s
o
(ft
t/>
I
CO
CD
n
O)
3
o
fD
ro ,
o
co
o
-p.
o
s
co
O
5.
3
a.
-h
3
O
-s
H-
CXI
o
o
o-
o
ro.
o
-------
r
100 110
10 20 30 40 50
Distance Downwind (South) in Km
Figure 3.6a Lead concentration cross-section, wind from north.
120
20 30 40 50
100 110 120
70 80 90
Distance Downwind (North) in Km
Figure 3.6b Lead concentration cross-section, wind from south.
Figure 3.6 Regional upwind/downwind analyses for lead.
42
-------
10 20 30 40 50 60 70 80 90 100 110 120
- Distance Downwind (South) in Km
Figure 3.7a Silicon concentration cross-section, wind from north.
T
10
T
40
1
50
T
70
30 40 50 60 70 80 90 100 110
Distance Downwind (North) in Km
Figure 3.7b Silicon concentration cross-section, wind from south.
Figure 3.7 Regional upwind/downwind analyses for silicon.
120
43
-------
implies that much (and possibly the majority) of sulfate in St. Louis is
synoptic in nature, due to regional sources rather than to sources within
the St. Louis AQCR.
3.2 POLLUTION ROSE ANALYSIS
This section discusses wind/pollution roses and their use in identify-
ing sources and examining particulate transport. Three types of roses are
presented: wind roses, pollution roses, and dosage roses. Wind roses il-
lustrate the percent of time that hourly winds are from each direction.
Pollution roses illustrate average particulate concentrations when the wind
is from given directions.* Theoretically, the pollution roses for an atmos-
pheric constituent emitted from a single source should all point approxi-
mately toward that source. In contrast, pollution roses for a constituent
emitted or produced throughout a broad area should be isotropic (invarient
with direction). A dosage rose represents the product of wind direction
frequencies (i.e. the wind rose) and average pollution levels for each wind
direction (i.e. the pollution rose); for each wind direction the dosage
rose measures the contribution of that wind direction to the overall average
concentration.
3.2.1 wind/Pollution Roses for IP and FINE
Figure 3.8 presents hourly wind roses for the RAMS sites. As evidenced
by the wind roses at all ten locations, the most frequent wind direction in
the St. Louis area is from the south.
Figures 3.9 and 3.10 present pollution roses for IP and FINE, respec-
tively. Both sets of roses are fairly isotropic. This result is consistent
with earlier conclusions that much of FINE is sulfate aerosol (which occurs
on a large spatial scale), and that the two dominant sources of IP are sul-
fates and ubiquitous fugitive dust sources. Both sets of roses do, however,
appear to show some directional biases toward the industrial areas in East
St. Louis and along the Mississippi River. Furthermore, the pollution roses
for FINE seem generally to have a slight but detectable south to east bias
(the reason for this will be seen in the next subsection).
*In this chapter, we use the six-hour dichotomous data at sites 103 andi 105
and the twelve-hour dichotomous data at other sites to construct pollution
roses.
44
-------
Wind Frequency (%)
Figure 3.8 Wind roses for the ten RAMS particulate sites.
45
-------
Figure 3.9 Pollution roses for IP at the ten RAMS particulate sites,
46
-------
$
(yg/m3)
Figure 3.10 Pollution roses for FINE at the ten RAMS particulate sites,
47
-------
At rural site 122, the IP and FINE pollution roses both exhibit a
slight bias (%3yg/m3) toward the urbanized area. At rural site 124, neither
the IP nor the FINE rose exhibit a bias toward the urban area. This supports
the conclusions of the previous section that only small amounts of aerosol
are directly transported from the city to rural sites, and that the trans-
ported aerosol is mostly FINE.
Figures 3.11 and 3.12 present dosage roses for IP and FINE at the ten
RAMS sites. Because the pollution roses are fairly isotropic, the dosage
roses resemble the wind roses (i.e. Figure 3.8).
3.2.2 Wind/Pollution Roses for All Dichotomous Variables
Figures 3.13 through 3.15 are collections of wind and pollution roses
for sites 103, 105, and 124 respectively. The first two sites are located
in the center city, while the last is the southernmost rural site. Wind,
pollution, and dosage roses are presented for FINE, COARSE, IP, and the
eight elements.
Most of the elements (as well as FINE, COARSE, and IP) exhibit iso-
tropic patterns at all three sites. The most obvious exception is the
highly wind-biased pattern of Ti to the south-southwest at sites 103 and
105. In contrast, there is a slight Ti spike from the north at site 124.
The pollution rose analysis can be used to help identify sources of
trace elements. For example, although station 103 lies a few km south of a
secondary lead smelter, its Pb rose exhibits no pronounced northerly spike.
The highest Pb levels correspond to wind from the south, the direction of
maximum vehicular traffic density. This suggests that motor vehicles,
rather than the lead smelter, are the primary source of particulate lead in
the St. Louis area. The strong spike of the Ti rose for site 103 points
toward several major coal-fired facilities and a large paint pigment plant
(see Figure 3.16). The south and southwest spikes of the Ti rose at site 105
are caused by the paint pigment plant alone, since the nearest coal burners
lie northwest and southeast of this site. The primary spikes of the Ti
rose at station 124 point north, toward the city, the highest concentration
of coal burners and the paint pigment plant. This analysis indicated that
the paint plant is the principal source of Ti emissions in St. Louis.
48
-------
Figure 3.11 Dosage roses for IP at the ten RAMS particulate sites,
49
-------
Figure 3.12 Dosage roses for FINE at the ten RAMS participate sites,
50
-------
FINE
COARSE
IP
WIND DIRECTION FREQUENCY-
ALUMINUM
X'
AVERAGE CONCENTRATION
AVERAGE DOSAGE
Figure 3.13 Wind and pollution roses for site 103.
-------
t r.
V
T3
O)
CJ
-------
VANADIUM
IRON
LEAD
x'
WIND DIRECTION FREQUENCY
X. T-
en
CO
AVERAGE CONCENTRATION
x I
X
AVERAGE DOSAGE
Figure 3.13 Wind and pollution roses for site 103 (continued).
-------
FINE
COARSE
IP
4-t
3
V
V
WIND DIRECTION FREQUENCY
ALUMINUM
AVERAGE CONCENTRATION
T
I X
AVERAGE DOSAGE
M
Figure 3.14 Wind and pollution roses for site 105.
-------
SULFUR
CALCIUM
4-1
4-1-
WIND DIRECTION FREQUENCY
TITANIUM
,\
\-'
H -( ' '
..v.
AVERAGE CONCENTRATION
AVERAGE DOSAGE
Figure 3.14 Wind and pollution roses for site 105 (continued),
I'-n^
-------
VANADIUM
I *
/.
WIND DIRECTION FREQUENCY
AVERAGE CONCENTRATION
LEAD
H-
AVERAGE DOSAGE
Figure 3.14 Wind and pollution roses for site 105 (continued).
-------
FINE
COARSE
IP
WIND DIRECTION FREQUENCY
ALUMINUM
i \
xx
' "
X
AVERAGE CONCENTRATION
AVERAGE DOSAGE
Figure 3.15 Wind and pollution roses for site 124.
\
-------
SILICON
SULFUR
CALCIUM
X" »-]
WIND DIRECTION FREQUENCY
TITANIUM
en
oo
\i
X
+H-
AVERAGE CONCENTRATION
AVERAGE DOSAGE
Figure 3.15 Wind and pollution roses for site 124 (continued),
-------
Ul
VANADIUM
IRON
WIND DIRECTION FREQUENCY
4-
AVERAGE CONCENTRATION
\. 4- \
AVERAGE DOSAGE
LEAD
X
Figure 3.15 Wind and pollution roses for site 124 (continued).
-------
Note: for selected sources, emission rates in tons/year
are shown in parentheses or within the symbols.
4280
4260
720
'
740
760
Figure 3.16 Emission sources in the vicinity of St. Louis
(Dzubay, 1979).
60
-------
At all three sites, FINE and sulfur exhibit a moderate easterly to
southerly bias. Because sulfate constitutes approximately 50% of FINE, the
agreement between the pollution roses for sulfur and FINE is not unexpected.
The east to south bias may reflect large scale sulfate phenomena, i.e. the
dense SO emissions in the Ohio Valley toward the east, or high pressure
J\
systems moving from the south which are known to be conducive to sulfate
formation.
The isoptropic nature of the pollution roses for COARSE and the crustal
elements (Si, Al, and Ca) supports the contention made in Chapter 2 that the
crustal material stems mostly from ubiquitous area sources of fugitive dust
rather than point sources of fly ash. The isoptropic property also reflects
the fact that the RAMS sites tend to be located away from strong local
influences (local fugitive dust or other local sources).
3.3 ESTIMATE OF BACKGROUND CONCENTRATIONS
Annual mean background particulate concentrations for the St. Louis
area can be calculated by taking average concentrations for the rural sites
(122 and 124) and subtracting both transported aerosol from the urban area
and local man-made aerosol. The results of the previous two sections in-
dicate that the influence of the urban area on the rural sites amounts to
o
approximately 2 to 5 yg/m of essentially fine aerosol. Record et al.
(1976) concluded that the local source influence on TSP at rural locations
is on the order of 2 to 4 yg/m . The local influence should be basically
coarse particles, and the mass contribution should be somewhat less for IP
(or COARSE) than it is for TSP (because IP and COARSE do not include
particles greater than 20 ym in size).
From the above considerations, we make an initial estimate that back-
ground aerosol concentrations for the St. Louis area are approximately 45
yg/m3 for TSP, 25 yg/m3 for IP, 13 yg/m3 for FINE, and 12 yg/m3 for COARSE.
As noted previously in this chapter, these background concentrations repre-
sent natural sources plus man-made sources exterior to the St. Louis AQCR.
61
-------
Record et al. (1976) estimated a background TSP concentration of
approximately 30 yg/m for the Midwest. The large difference between their
estimate and ours is probably due to three factors: (1) Record et al. seem
to have too low of an estimate for transported secondary aerosols (only
3 yg/m ); (2) the local scale contribution that we assumed, 2 to 4 yg/m3
based on Record et al., may be too low (i.e. local scale contributions at
the rural RAMS sites may be elevated due to fugitive dust from agriculture,
roads, construction, etc.); and (3) the transport contribution from the
urban area may be higher than we were able to detect with rather crude
wind-direction analyses. Until these issues are resolved by further
analysis, the background aerosol concentration for St. Louis must be
regarded as somewhat uncertain. The best guess, however, based on current
information would appear to be as follows:
TSP ... 35 yg/m3
IP ... 20 yg/m3
FINE ... 10 yg/m3
COARSE ... 10 yg/m3
*
Actually, Record et al. (1976) estimated a background contribution of 25
yg/m3 in terms of annual geometric mean (AGM). By examining data at non-
urban locations, we have found that a 25 yg/m3 AGM typically corresponds
to a 30 yg/m3 arithmetic average.
62
-------
4.0 TEMPORAL PATTERNS OF ST. LOUIS PARTICULATE DATA
In response to seasonal and diurnal patterns of meteorology as well as
to seasonal, weekly, and diurnal patterns of human activities, ambient aerosol
concentrations exhibit significant temporal variations. These temporal pat-
terns can frequently provide insight into the nature and sources of particu-
late matter. This chapter discusses the temporal patterns of several meteor-
ological variables and several components of particulate, as monitored at
Lambert Field and the ten RAMS particulate sites.
4.1 SEASONAL PATTERNS
This section discusses the seasonal patterns of twelve particulate
parameters and eight meteorological parameters. The particulate and weather
patterns are then analyzed jointly to assess the effects of meteorology on
ambient aerosol concentrations.
4.1.1 Particulate Parameters
Figure 4.1 illustrates the seasonal patterns of TSP, IP, COARSE, and
FINE. Each pattern consists of monthly averages for the year 1976. The
data are grouped according to four urban sites (103, 105, 108, and 112),
three suburban sites (115, 118, and 120), and one rural site (122). As
noted in Chapter 2, urban concentrations are higher than suburban con-
centrations which in turn are higher than rural concentrations. Suburban
levels are generally closer to rural levels than urban levels.
All four particulate parameters in Figure 4.1 tend to peak during the
summer (June through September). The summer peak is strong for FINE, moder-
ately strong for IP, moderate for COARSE, and very weak for TSP.
Figure 4.2 displays the seasonal patterns for the eight elements. The
urban versus suburban versus rural differences are again obvious. As noted
in Chapter 3, the spatial differences are especially pronounced for Pb, V,
Ti, and Fe and least pronounced for S.
*
Sites 105 and 124 are excluded from the seasonal analysis because
dichotomous measurements were not taken at these sites from September
through December.
63
-------
TSP
Urban
Suburban
Rural
- Urban
Suburban
Rural
Suburban
Rural
.-Urban
Suburban
FEE MAR APR MAY JUN JUL AUG SEP OCT NOV
Figure 4.1 Seasonal patterns of TSP, IP, COARSE, and FINE.
64
-------
3 3
Concentration (jig/m ) Concentration (yg/m )
Concentration (yg/m ) Concentration (vg/m )
crt
en
(O
c
-5
fD
n>
o<
w
o
TD
CU
fD
(D
0)
3
OJ
O
o
3
n
fD
3
c-t-
-S
o>
g
^
-------
3 333
Concentration (yg/m ) Concentration (yg/m ) Concentration (yg/m ) Concentration (yg/m )
c.
-5
co
(D
Oi
(/)
o
3
D>
o
Ca
c-l-
rt
-$
(/>
O
-h
cr> ft)
rc
O)
o>
n
o
o
n>
O
3
in
n
O
3
C
Q.
cn
^ t
4
73 CO
C C
-S CT
oj e
cr
-s
O"
-------
Figure 4.2 indicates that sulfur has a strong summer peak, similar to
the pattern of FINE. The parallel behavior of sulfur and FINE, a phenomenon
noted throughout this report, is consistent with the results of the chemical
element balance (Chapter 1) that showed sulfate contributes about one-half
of FINE.
The crustal elements -- Si, Al, and Ca -- display moderate summer
peaks, similar to COARSE. This is consistent with other findings in this
report that crustal material constitutes the dominant part of COARSE. Ti-
tanium and iron show seasonal patterns fairly similar to those for Si, Al,
and Ca, indicating that crustal material (in addition to the paint pigment
plant and iron/steel mills) may be a major source of Ti and Fe as well.
The tendency for vanadium to peak in the winter reflects increased use
of fuel oil for space heating during the winter. The element Pb, appears to
lack a significant seasonal pattern. This evidently implies that gasoline
consumption, as well as weather conditions governing the dispersion of lead,
tend to be fairly constant during the year. An alternative explanation,
that traffic and weather correlate inversely, such that periods of high
vehicle use (e.g. summer) are accompanied by favorable (dispersive) meteor-
ology, seems ruled out by the weather pattern studies of the next section.
4.1.2 Meteorological Parameters
Figure 4.3 illustrates the seasonal trend in daily average wind speed,
showing a peak in the winter and a trough in the summer. Not surprisingly,
Figure 4.4, the average value of number days persistent calm, shows an in-
verse pattern, peaking in the summer with a minimum during the winter.
Figure 4.5 shows that afternoon mixing height reaches a maximum in
the summer months, falling to a minimum during the winter. Morning mixing
height exhibits a less pronounced and inverse pattern, with a summer minimum.
The daily spread between morning and afternoon mixing height, a measure of
the amount of ground heating by the sun, is highest in summer, as one would
anticipate.
Figure 4.6 shows the expected seasonal trends in daily minimum and
maximum temperatures, with highs in the summer and lows in the winter. The
67
-------
ea
o
M-
O
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Figure 4.3 Monthly averages of daily (vector) average wind speeds.
2.5-1
2.0.
1.5-
JAN FEB MAR APR
DEC
Figure 4.4 Monthly averages of number of days persistent calm (#DY
CALM) (Lambert Field data).
68
-------
to
s_
(U
M-
O
a
a;
-a
CD
c
03
a
JAN
MAR APR MAY
JUN JUL
Month
AUG
1
SEP
OCT NOV DEC
Figure 4.5 Monthly averages of morning and afternoon mixing heights.
69
-------
30 -i
-10
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Month
Figure 4.6 Monthly averages of daily minimum and maximum temperatures.
80 -,
60-
0)
O)
40 -
20 _
T
T
T
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Month
Figure 4.7 Monthly averages of relative humidity.
70
-------
gap between the minimum and maximum temperature lines is very consistent
from month to month.
Figure 4.7 indicates that relative humidity undergoes only a slight
seasonal variation, with somewhat higher levels in winter that in summer.
Figure 4.8 presents the number of precipitation days per month, showing a
slight peak in winter and spring. Figure 4.9, the number of days since
last precipitation, exhibits peaks in April and November.
4.1.3 Comparison of Particulate and Meteorological Patterns
The relationship between particulate concentrations and meteorology
will be discussed at length in Chapter 5. For sake of completeness, however,
it is useful to note that certain relationships observed in Chapter 5 are
also apparent in the seasonal patterns. For example, Chapter 5 demonstrates
that prolonged stagnations (high #DY CALM) and high temperatures are the two
weather conditions closely tied to elevated concentrations of sulfur and FINE.
The seasonal patterns are consistent with this since all of these variables
show a pronounced summer maximum. Also, Chapter 5 demonstrates that elevated
levels of COARSE and the crustal elements occur under dry, calm conditions.
The seasonal patterns reflect this because COARSE and the crustal elements
display a moderate summer peak, while dryness (as measured by the precipi-
tation variables and relative humidity) and calmness (as measured by wind
speeds and calm days) also tend to display summer maxima.
4.2 WEEKLY (HEBDOMADAL) PATTERNS
Since human activities tend to follow pronounced weekly cycles, while
natural processes do not, it is informative to look for weekly patterns in
particulate concentrations. Of particular interest are differences between
weekday and weekend concentrations, since anthropogenic emissions are higher
on weekdays than on weekends.
4.2.1 Weekly Patterns at Two Selected Sites
Figures 4.10 through 4.20 present the hebdomadal patterns of IP, FINE,
COARSE, and the eight elements at two urban sites, 103 and 105. Of the ten
71
-------
ITS
Q
H-
O
i.
Ol
.a
12 -
10 -
8 -
6 -
4 -
2 -
JAN
FEB MAR APR MAY JUN JUL AUG SEP OCT NOV
Month
Figure 4.8 Number of days of precipitation per month.
DEC
72
-------
12 -,
10 -
8 ~
-------
HEBDOMADAL
DIURNAL
150 -i
75-
SITE 103
100-
0
1 12 BB 12 03 12 00 12 OB 12 W 12 08 12 21
SUN MQN TUE WED THU FRI SflT
SITE 105
i 1 1 1
00 06 12 18 24
1 * i ' ' ' i * ' * i ' ' > i ' i ' i i i i i i i i i
BH 12 BB 12 BB 12 00 12 08 12 00 12 00 12 21
SUN MQN TUE WED THU FRI SRT
00 06 12 18 24
50-
25-
Figure 4.10 Hebdomadal and diurnal patterns of IP (
SITE 103
aa 12 aa 12
I ' ' ' i ' i ' i
12 m 12 aa 12 00
12 w is si
SUN MON TUE WED THU FRI SflT
SITE 105
00 06 12 18
25-
0-
' ' ' i ' ' ' i ' ' ' i ' ' ' i ' i ' i i i i i i i i i
> 12 ea 12 0a 12 ea 12 00 12 0a 12 on 12 m
SUN MON TUE WED THU FRI SflT
00 06 12 18 24
Figure 4.11 Hebdomadal and diurnal patterns of FINE (jag/m3).
74
-------
HEBDOMADAL
DIURNAL
100-i
50-
SITE 103
1 ' I ' ' ' l ' ' ' I ' ' ' I
aa 12 ea 12 00 12 ea 12 aa 12 aa 12
1IliiIiir~
50-|
25
a 21
SUN MQN TUE WED THU FRI SflT
SITE 105
06 12 18 24
i r -fr r-T | i i- i | i i - i | i i i | i i i | i i i |
ea 12 ea 12 00 a aa 12 aa 12 00 12 00 12 21
SUN MQN TUE WED THU FRI SflT
00 06 12 18 24
Figure 4.12 Hebdomadal and diurnal patterns of COARSE (yg/m ).
8350 -i
4925-
1500
SITE 103
iijiii|-H[' ' i ' ' ' i ' * ' i ' ' ' i ' ' ' i
aa 12 at 12 an is aa 12 ea 12 aa 12 ea 12 21
SUN MQN TUE WED THU FRI SflT
1 1 1 1
00 06 12 18 24
6750-,
4050-
1350
SITE 105
ea 12 ea 12 00 12 ea 12 aa 12 00 12 00 12 21
SUN MQN TUE WED THU FRI SflT
00 06 12 18 24
Figure 4.13 Hebdomadal and diurnal patterns of sulfur (ng/m3).
75
-------
HEBDOMADAL
DIURNAL
8400-1
5000-
1600
SITE 103
ea 12
i ' > ' i
is 00 12 ea 12 aa 12 00 12
SUN MON TUE WED THU FRI SRT
12 21
1 1 1 1
00 06 12 18 24
6750 -i
4000-
1250
SITE 105
i i i i i i i i i i | i i i | i i i | i i i | ii i |
aa 12 00 12 ea 12 eg 12 BB 12 ea 12 00 12 24
SUN MON TUE WED THU FRI SRT
00 06 12 18 24
Figure 4.14 Hebdomadal and diurnal patterns of silicon (ng/m3).
2650-
1575-
500
SITE 103
1 ' i ' ' ' i
fla 12 at 12 eg 12
' ' I
12 00 12 00 12
2300-
1350-1
SUN MON TUE WED THU FRI SRT
SITE 105
12 2t
00 06 12 18 24
400
' ' I ' ' ' i ' ' i i ' i i i i i i | i i i i i i i i
ea 12 aa 12 0a 12 ea 12 ea 12 00 12 0a 12 21
SUN MON TUE WED THU FRI SRT
00 06 12 18 24
Figure 4.15 Hebdomadal and diurnal patterns of aluminum (ng/m3).
76
-------
HEBDOMADAL
DIURNAL
7050-,
1150-
1250
SITE 103
5450-
3150-
850
' i i"1 ' i ' i i ' ' | i i i i i i i I i i i
12 ea 12 oa 12 aa is aa 12 0a 12 aa 12
SUN MON TUE WED THU FRI SflT
SITE 105
i 1 1 1
00 06 12 18 24
1 r i ' ' ' I ' ' ' I ' ' i i i i i i i i i i i i i
BB 12 m 12 aa 12 ee 12 aa 12 00 12 00 ia
SUN MQN TUE WED THU FRI SflT
-i 1 1 1
06 12 18 24
Figure 4.16 Hebdomadal and diurnal patterns of calcium (ng/m3).
1700 n
925-
150
SITE 103
2050-,
1125-
t'a ' ^ ' ' ' I ' ' ' I ' i ' | i i i ! i i
'SUN HQN "TUE MWED "THU "FRI
J,
06 12 18 24
SITE 105
"TUE "WED "THU "FRI VT
00 06 12 18 24
Figure 4.17 Hebdomadal and diurnal patterns of lead (ng/m3).
77
-------
HEBDOMADAL
DIURNAL
25-
SITE 103
aa is
i
12 08 12 ea 12
12 m 12
SUN MQN TUE WED THU FRI SflT
12 2t
00 06 12 18 24
50-
25-
0
SITE 105
"T ' I ' ' ' i ' ' i i ' ' i i ' t i i i i i i i i
aa 12 ea 12 0a 12 aa 12 oa 12 w 12 0a 12
SUN MON TUE WED THU FRI SflT
00 06 12 18 24
Figure 4.18 Hebdomadal and diurnal patterns of vanadium (ng/m3).
700-
350-
0-
SITE 103
ea 12 0a 12
1 ' ' I ' r ' I ' ' ' I ' ' '
12 aa 12 0a 12 aa 12
1750-
SUN MON TUE WED THU FRI SflT
i 1 1 1
00 06 12 18 24
875-
fl-
SITE 105
ea 12 aa 12 ea 12 ca 12 ae 12 00 12 aa 12 24
SUN MON TUE WED THU FRI SflT
00 06 12 18 24
Figure 4.19 Hebdomadal and diurnal patterns of titanium (ng/m3).
78
-------
HEBDOMADAL
DIURNAL
3300 -i
2000-
700
SITE 103
12 H3 12 03 12 00 12 Ba 12 00 12 00 12 21
SUN MON TUE WED THU FRI SflT
00 06 12 18 24
2700-
1525-
350
SITE 105
82 12 23 12 00 12 00 12 03 12 08 12 00 12 2
-------
dichotomous sampler sites, these two are the only ones at which routine 6-
hour data were taken. Also displayed in the figures are the corresponding
average diurnal patterns, which will be discussed in the next subsection.
Appendix F presents the actual numbers used to construct Figures 4.10 through
4.20 as well as hebdomadal and diurnal patterns disaggregated by quarter of
the year.
Few significant patterns emerge among the hebdomadal traces of Figures
4.10 through 4.20. Lead, the only element exhibiting a pronounced diurnal
pattern, displays a corresponding series of seven "bumps" in the hebdomadal
pattern. The main general feature among the figures, that particulate levels
are apparently lower on weekends, will be discussed at length in the next
subsection.
4.2.2 Weekend-Weekday Differences
Table 4.1 summarizes weekend-weekday differences in particulate con-
centrations at the RAMS sites (more detailed data on weekend-weekday dif-
ferences are presented in Appendix G). Percent differences averaged over
urban, suburban, and rural sites are listed, and the statistical signifi-
cance of the changes as determined by t-tests at individual sites are
indicated.
Table 4.1 illustrates that the average percent differences, averaged
over each type of site, are always negative. Evidently, decreased human
activity levels during weekends do lower particulate concentrations. The
differences tend to be greatest for the urban sites (especially stations
105, 106, and 108 which are located near the heavily industrialized zones
close to the river, towards the southern and northern ends of the city),
less for suburban sites, and even less for rural sites. This indicates, as
expected, that man-made sources are relatively more important in urban areas
than rural areas. The lesser rural weekend-weekday differences may also
reflect a shift in the spatial distribution of activity (e.g. traffic);
specifically, recreation related activities during the weekends may be rela-
tively more intense in rural areas than are work related activities during
weekdays.
80
-------
TABLE 4.1 WEEKEND-WEEKDAY DIFFERENCES IN ST. LOUIS PARTICULATE CONCENTRATIONS
PARTICULATE
VARIABLE
TSP
IP
FINE
COARSE
Sulfur
Sil icon
Aluminum
Calcium
Lead
Vanadium
Titanium
Iron
Average perc
*
Average week
confidence 1
AVERAGE PERCENT DIFFERENCE1"
Five Three Two
Urban Suburban Rural
Sites Sites Sites
- 9% - 5% -12%
-18% - 8% - 8%
-15% - 7% -12%
-20% - 9% - 4%
-18% -10% -20%
-15% - 6% - 4%
-17% -11% - 7%
-26% -18% - 5%
-22% -18% - 1%
-29% -11% - 5%
-19% - 9% - 3%
-29% -14% -10%
STATISTICAL
Urban
Sites
103 105 106 108 112
** *
* ** ** ** *
** ** **
* ** ** ** *
*****
** ** *
** ** **
* ** ** ** *
** ** *
* ** ** **
*
* ** ** **
SIGNIFICANCE
Suburban
Sites
115 118 120
*
*
**
** **
**
**
*
* **
Rural
Sites
122 124
* *
*
°nt difference i- weekend - weekday
weekday
end concentration is less than average weekday concentration
eve! .
at a 95 percent
oo
**
Average weekend concentration is less than average weekday concentration at a 99 percent
confidence level.
-------
The weekend-weekday differences are lower for TSP than for the dichot-
omous variables. As will be indicated by results in the next chapter, and
as makes sense because TSP includes particles in the 20 to 50 urn size range,
TSP tends to be more dominated by dust sources than IP. The lower weekend-
weekday differences may be a reflection that TSP is also significantly more
dominated by wind-blown dust than IP. We cannot be sure of this, however,
because the TSP data base is less robust than the IP data base (because of
every third day sampling rather than daily sampling); the lesser weekend-
weekday differences for TSP may thus be partly a statistical artifact.
As must be, the weekend-weekday differences for IP are intermediate
to those for FINE and COARSE. FINE and its most important component, sulfate
(or sulfur), are the only particulate variables that show rural weekend-
weekday percentage decreases about as large as the urban decreases. This
reflects the large spatial scale of sulfates and the other secondary aerosols
that constitute the major part of FINE. It is not known how much of the
weekend sulfate decrease is related to weekend changes in St. Louis air basin
SO emissions, larger scale SO emissions, and/or photochemical precursor
/\ | /\
emissions (photochemical smog from hydrocarbons and NO promotes more rapid
/\
production of sulfate from S02). The weekend-weekday differences for sulfur
do imply, however, that a significant fraction of sulfate is not related to
synoptic scale phenomena. For large scale phenomena, we would expect a con-
siderable time lag between emission changes and air quality changes; this
time lag would lessen the likelihood of observing weekend-weekday changes.
The weekend-weekday differences for COARSE are fairly similar to those
for the crustal elements (Si, Al, and Ca). This is expected other parts
of this report indicate that crustal material is the predominant source of
COARSE. That significant weekend-weekday differences do exist for the
crustal material suggests that most of the crustal material (at least for
urban! sites) stems from man-made sources; this result is consistent with a
major conclusion of Chapter 5 (that the crustal material arises mostly from
anthropogenic fugitive dust).
At urban and suburban locations, the weekend-weekday differences for
crustal limestone (Ca) are significantly larger than those for crustal shale
82
-------
(Al and Si). This effect could be explained by the hypothesis that crustal
limestone arises relatively more from cement manufacturing and construction
(both of which would tend to show very strong weekend changes),
while crustal shale arises relatively more from traffic, agriculture, and
wind-blown dust. The above hypothesis is also supported by the analysis of
spatial patterns (Chapter 2) which showed that Ca has a greater urban/nonurban
difference than Si and Al, indicating that Ca is more closely associated with
commercial and industrial activities.
The weekend-weekday changes for lead should be fairly representative
of patterns in automotive traffic. That these changes are not extremely
different from the changes in the crustal elements indicates that traffic
may be a significant source of fugitive dust.
Two of the elements that may have significant industrial components,
V (fuel oil) and Fe (iron/steel industry), demonstrate the greatest urban
weekend-weekday differences. In fact, the large weekend-weekday difference
for vanadium implies that a major portion of fuel oil is burned in industrial
and commercial sources (residential fuel oil use would show little weekend-
weekday change).
4.3 DIURNAL PATTERNS
Figures 4.10 through 4.20 (in the previous section) present diurnal
patterns for IP, FINE, COARSE, and the eight elements at sites 103 and 105,
the only two sites with routine 6-hour average data. The only parameter
showing a pronounced diurnal pattern is lead which demonstrates a distinct
afternoon (noon - 6 PM) minimum. This pattern is somewhat puzzling because
noon - 6 PM traffic levels should be as great as midnight-6 AM and 6 PM -
midnight traffic levels. An alternative explanation, high afternoon mixing
heights, seems contradicted by the fact that none of the other elements
exhibit lead's strong diurnal pattern (although Fe and Ti do have a small
afternoon dip). Perhaps the explanation is mixing height, with the diurnal
patterns for other elements being smoothed by higher source activity in the
afternoon or by a complex interaction of diurnal source strength, diurnal
mixing height, and particle settling (note that lead, a fine aerosol, would
not have the settling effect).
83
-------
As noted earlier in this report, road dust is suspected of being a
significant source of the crustal material that dominates COARSE. Although
possibly a significant source of crustal material, road dust appears not
to be a predominant source because the diurnal patterns for Al, Si, Ca, and
COARSE do not reflect the strong diurnal features exhibited by the automo-
tive tracer, Pb.
84
-------
5.0 DECISION-TREE ANALYSIS OF PARTICULATE
AND METEOROLOGICAL DATA.
One goal of this study is to gain an understanding of the relation-
ship between particulate air quality and meteorology and to see if this
understanding can provide insights into the major sources of ambient par-
ticulate matter. This chapter investigates the relationship between parti-
culate concentrations and meteorology by using a very general and powerful
statistical tool, decision-tree analysis. For three locations (urban sites
103 and 105, and rural site 124), decision-trees are constructed relating
each of the twelve particulate parameters (TSP, IP, FINE, COARSE, and the
eight elements) to eleven meteorological parameters. Decision-trees are
also formed relating IP and FINE to the eight elemental variables.
5.1 THE CART DECISION-TREE PROGRAM
The decision-tree program used in this study is the CART (Classification
and Regression Trees) program developed at Technology Service Corporation.
The purpose of the program is to explain the variation in a dependent vari-
able (e.g. particulate concentrations) by sequentially splitting the data
according to ranges of the independent variables (e.g. meteorological para-
meters). As illustrated by the example in Figure 5.1, the net result is a
decision-tree that accounts for the variance in the dependent variable
according to groups (e.g. meteorological classes) defined by the independent
variables.
The CART program consists of a rather complicated set of algorithims.
The four basic parts of the program are as follows:
I. A set of decision-trees is generated with the entire data base by
sequentially splitting the data according to ranges of the inde-
pendent variables. Each sequential split is performed on the sub-
group with the greatest variance; the split is selected to maximize
the variance explained, weighted by a factor which favors robust
(many data point) splits. The processs is terminated by limits on
85
-------
Mean Sulfur Cone, (ng/m3)'Samp1e Size
Standard Deviation
N= 287
M- 3524.51
S= 2510.89
N- 153
PI- 2455.21
S- 1337.25
M= 4745.48
S= 3003.71
Meteorological Variable
Spl it Point for
Meteorological
Variable
N= 67
M- 3653.28
S= 2334.89
RV W
N= 67
M- 5837.66
6= 3210.54
oo
O1
3.81
*RfiNK
3.81
1.00
*RRNK= 4*
<=
1.00
N= 32
M= 4816.47
S- 2613.70
N- 35
M= 2589.79
S- 1384.33
N= 43
J1= 4717.63
8= 2558.53
N- 24
f1= 7844.40
S- 3328.23
«DY CRLfl
2.00
N= 11
M- 5605.81
8= 1926.27
N= 13
M- 9738.61
S= 3107.76
Rank Order of Terminal Node with Respect to Sulfur Concentration
Figure 5.1 Example decision-tree, sulfur at site 103 versus meteorological variables.
-------
the allowable number of terminal nodes and on the minimal allowable
improvement in variance explained.
II. The final, most complex decision-tree from part I is subjected to
a pruning procedure (a procedure that reduces the complexity of
the tree while losing minimal amounts of variance explained). The
pruning is carried out sequentially, with the decision-tree at each
step graded according to a complexity parameter.
III. The data base is randomly divided into ten parts, each containing
10% of the data. Withholding each of these parts in turn, ten new
sets of decision-trees are generated with the remaining ninety
percents of the data following the procedures in parts I and II.
Each decision-tree in these sets is rated according to percent
variance explained using the independent 10% of the data. An
optimal complexity parameter is determined by maximizing the average
**
percent variance explained over the ten sets of trees. The pruned
tree from part II above (based on all the data) corresponding to the
optimal complexity parameter is chosen as the final CART decision-
tree. The percent variance explained is taken as the average level
achieved by the ten cross-validation tests using independent data
sets.
IV. The final aspect of CART allows one to test the importance of each
independent variable to the overall decision-tree. For each inde-
pendent variable, the entire data set is divided into quartiles
*
The first part of CART is similar to the AID decision-tree program developed
at the University of Michigan (Sonquist et al., 1973), except CART involves
three refinements: (1) AID requires the independent variables to be specified
in discrete form (i.e. continuous variables must be divided into pre-selected
ranges), while CART has no such requirement; (2) CART includes special pro-
cedures to handle cyclical variables (e.g. wind direction); and (3) AID
selects splits solely on percent variance explained, while CART adds a
weighting function to favor robust splits.
**
Note: at a small level of complexity, little variance is usually explained
because the data have not been divided enough; at a large level of complexity,
loss in variance explained occurs due to over-fitting of the (ninety percent)
data sets used to construct the trees.
87
-------
ranked according to that variable. The variance explained in the
predicted values (from the decision-tree) by this quartile split
is used as a measure of the importance of the variable to the
decision-tree. An example output from this part of the program
(corresponding to the decision-tree in Figure 5.1) is presented in
Table 5.1.
TABLE 5 1 PERCENT OF VARIANCE IN THE DECISION-TREE FOR
' SULFUR AT SITE 103 ACCOUNTED FOR BY INDIVIDUAL
METEOROLOGICAL VARIABLES.
METEOROLOGICAL VARIABLE PERCENT VARIANCE EXPLAINED
oo/
#DAYS PP L'0
#DY CALM 16%
%REL HUM l%
AV WD SP ll%
D7AM PRS 5%
MAX TEMP 27%
MIN TEMP 24%
MIXHT AM 9%
MIXHT PM 13%
PRECIPIT 1%
WIND DIR 6%
For the problem of relating particulate air quality to meteorological
variables, the CART decision-tree program offers several advantages over more
conventional data-analytic techniques such as multiple linear regression.
The advantages include the following:
The CART program is a non-parametric technique based on a general
form of the least-squares principle; it does not involve restric-
tive assumptions such as additivity and linearity.
The meteorological classes defined by CART can be interpreted on
physical grounds more easily than a multivariate regression equation,
88
-------
t Because the CART trees are evaluated by a cross-validation test
with independent data sets, an unbiased estimate is obtained of
the percent variance explained.
Previous experience (Trijonis et al., 1979), indicates that
decision-tree analysis of aerosol/meteorology relationships
generally explains more variance than multiple linear regression.
5.2 DECISION-TREES RELATING IP AND FINE TO ELEMENTAL CONCENTRATIONS
Before we proceed to anlayze the decision-trees relating particulate
variables to meteorology, it is worthwhile to digress slightly and examine
decision-trees relating IP and FINE to elemental concentrations. Because
the elemental variables serve as tracers for sources, the relationship of IP
(or FINE) to the elements will itself shed light on the major sources contri-
buting to IP (or FINE). Also, knowing how IP (or FINE) depends on the elements
will be useful for interpreting the decision-trees relating all the particu-
late variables to meteorology.
Decision-trees relating IP and FINE to the eight elemental variables at
sites 103, 105, and 124 are presented in Appendix H. Table 5.2 summarizes the
percent variance explained by the trees; Tables 5.3 and 5.4 summarize the
relationships of FINE and IP to the elemental variables.
TABLE 5.2 PERCENT VARIANCE EXPLAINED BY DECISION-TREES
RELATING IP AND FINE TO ELEMENTAL VARIABLES.
PERCENT VARIANCE EXPLAINED
SITE IP FINE
103 71+3 68 i 3
105 68+3 73 t 3
124 59 + 3 68 + 3
The percent variance explained by the elemental concentrations is quite
high, 59% to 71% for IP and 68% to 73% for FINE. These are equivalent to
correlation coefficients of 0.77 to 0.84 for IP and 0.83 to 0.85 for FINE.
Note that the percent variance explained is the square of the correlation
coefficient.
89
-------
TABLE 5.3 RELATIONSHIP OF FINE TO ELEMENTAL CONCENTRATIONS AS FOUND IN THE DECISION-TREE ANALYSIS,
SITE
103
105
124
PERCENT OF VARIANCE IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL ELEMENTS*
S Si Al Ca Pb V Ti Fe
55% 13% 10% 10% 10%
56% 15% 17% 16% 10% 19%
55%
*
Includes only those elements explaining at least 10% of the variance
SITE
103
105
124
if if
CONDITIONS ASSOCIATED WITH HIGH FINE CONCENTRATIONS
high S (6); high Pb (2); high Si (1); high Ti (1).
high S (7); high Pb (1); high Ca (1); high Si (1); high Fe (2).
high S (5)
**
Elements are listed in approximate order of appearance in decision-tree. Number in parentheses
represents the number of splits in the decision-tree made on this element.
-------
TABLE 5.4 RELATIONSHIP OF IP TO ELEMENTAL CONCENTRATIONS AS FOUND IN THE DECISION-TREE ANALYSIS
SITE
103
105
124
PERCENT OF VARIANCE
S Si
41% 27%
41% 33%
37% 27%
IN DECISION-TREE ACCOUNTED
Al Ca Pb
27%
32%
27%
36%
35%
24%
19%
10%
FOR
V
19%
BY INDIVIDUAL ELEMENTS
Ti Fe
*Includes only those elements explaining at least 10% of the variance
**
SITE
103
105
124
24%
23%
29%
in the
33%
42%
28%
decision- tree
CONDITIONS ASSOCAITED WITH HIGH IP CONCENTRATIONS
high
high
high
S (5);
Fe (4)
S (5);
high Ca
; high S
high Ca
(4);
(5);
(3);
high Si (1);
high Si (2);
high Si (1),
high Fe
high Ca
high Fe
(2);
(i);
(i);
high
high
high
v (1);
Al (3)
Ti (1)
high Al
; high V
; low Ti
(1).
(1).
(1).
Elements are listed in up^/i v/^nnvivv, ~.-^. ~. ~r, --
parentheses represents the number of splits in the decision-tree made on this element.
-------
The high degrees of percent variance explained are especially notable because
they are unbiased (i.e. they have been determined by cross-validation tests
with independent data sets).
As indicated by Table 5.3, sulfur is by far the single most important
element explaining the variance in FINE. The major reason for this strong
relationship is that sulfate aerosol (traced by the element sulfur) alone
constitutes approximately one half of FINE (see Chapter 1). Also, the con-
ditions that produce high sulfate may also lead to high concentrations of
other fine aerosols, especially other secondary aerosols; this effect would
further strengthen the statistical association between sulfate (or sulfur) and
FINE. It is notable that, although the other elements do show some link with
FINE, none of them consistently bears a strong relationship with FINE. This
may imply that the non-sulfate fraction of FINE is a mixture of fairly small
contributions from several types of primary and secondary aerosols.
As illustrated by Table 5.4, sulfur and the crustal elements (e.g. Si,
Al, and Ca) account for most of the variance in IP. This is because, as
noted in Chapter 1, sulfate is the major constituent of FINE while crustal
material is the major constituent of COARSE. That the element Ke also is
important may be caused by any of three factors: (1) the iron/steel industry
may be a significant source of IP, (2) at sites 103, 105, and 124, Fe may
represent crustal material rather than the iron/steel industry, or (3) due to
intercorrelations in the day-to-day fluctuations of .the elements, Fe may
be acting as a surrogate for some of the other elements.
5.3 DECISION-TREES RELATING PARTICULATE VARIABLES TO METEOROLOGY
The relationship between ambient particulate concentrations and meteor-
ology can provide insights regarding the sources of particulate matter
(Throgmorton and Axetell, 1978; Trijonis et al., 1979; Price et al., 1977).
For example, high particulate levels associated with dry conditions and very
high wind speed implicates wind-blown dust; high particulate levels associated
with dry conditions, high mixing height, and low to moderate wind speed im-
plicates man-made dust sources; high particulate levels associated with low
mixing height, low wind speeds, and preferential wind directions tends to
92 *
-------
imply conventional (e.g. industrial) sources. The reader is referred to
Trijonis et al. (1979) for an extensive discussion of the emission source
implications that can be drawn from the relationship between aerosol con-
centrations and meteorological parameters.
This section uses decision-tree analysis to characterize the meteor-
ological conditions associated with high particulate levels and to gain in-
formation on sources of ambient particulate matter. Decision-trees are con-
structed for twelve particulate variables (TSP, IP, FINE, COARSE, and the
eight elements) at three RAMS sites (numbers 103, 105, and 124). The
eleven meteorological parameters used are summarized in Table 1.2 (Section
1.1). The decision-trees are presented in Appendix I.
Table 5.5 summarizes the percent variance explained by the decision-
trees. The results are good to fair for sulfur, FINE, and IP (average per-
cent variance explained on the order of 30%, equivalent to a correlation
coefficient of ^ 0.55); fair to poor for silicon, aluminum, calcium, iron,
and COARSE (average percent variance explained on the order of 18%, equiva-
lent to a correlation coefficient of ^ 0.42); and poor for lead, vanadium,
titanium, and TSP (percent variance explained less than 10%).
Despite the less than excellent levels of percent variance explained,
the forms of the decision-trees do tend to provide a reasonable and consis-
tent picture of the type of meteorology associated with high particulate
levels. Tables 5.6 through 5.17 summarize the particulate/meteorology
relationships for the twelve aerosol variables as observed in the decision-
trees. The principal conclusions that can be drawn from these tables are as
follows:
t Sulfur (Table 5.10)
Two basic themes stand out in the decision-tree classes for
high sulfur: prolonged stagnations (high #DY CALM) and high temper-
atures. This is consistent with the evidence indicating that par-
ticulate sulfur basically represents the secondary aerosol, sulfate
(Stevens et al., 1978). The greatest sulfate concentrations are
likely to accumulate when calm conditions lead to low dispersal of
SOX emissions, and when intense photochemical activity (due to high
temperatures and associated high solar radiation) elevates the
sulfate formation rate.
93
-------
TABLE 5.5 PERCENT VARIANCE EXPLAINED BY DECISION-TREES
RELATING PARTICULATE CONCENTRATIONS TO
METEOROLOGICAL PARAMETERS.
VARIABLE PERCENT VARIANCE EXPLAINED*
TSP
FINE
COARSE
IP
Sulfur
Silicon
Aluminum
Calcium
Lead
Vanadium
Titanium
Iron
Site 103
-2 + 7
37 +. 6
25 + 6
37 + 6
31 i 6
25 i 6
14 + 6
20 + 6
7 ± 4
7 + 5
-40 + 57
39 +. 8
Site 105
16+6
33+8
21+8
25 + 10
36 i 7
15 1 6
26+6
23 + 11
-13 + 17
11+3
-4 + 17
7 +. 5
Site 124
5 ± 15
34 + 12
3 + 10
29 + 10
15 ± 12
15 i 12
8 + 8
17 + 10
-25 ± 55
10+5
16+6
15 i 10
*
The percent variance explained is determined by cross-validation tests
with independent data sets and can therefore be negative.
94
-------
TABLE 5.6 RELATIONSHIP OF TSP TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
SITE
103
105
124
*
Includes
SITE
103
8 105
124
PERCENT
#DAYS PP #DY CALM
13%
OF VARIANCE
&REL HUM
14%
10%
115,
IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES*
AV WD SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
11% - 39% 32%
41% 37%
19% 21% 22% 27%
only those variables explaining at least 10% of the variance in the decision-tree.
CONDITIONS ASSOCIATED WITH HIGH TSP CONCENTRATIONS**
low ;=REL HUM (1).
high *DAYS PP (1).
high MAX TEMP (1); low
°«REL HUM (1)
; low D7AM PRS (1); high iCDAYS PP (1).
Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits i-
the decision-tree made on this parameter.
-------
TABLE 5.7 RELATIONSHIP OF IP TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
SITE
103
105
124
PERCENT
'DAYS PP *DY CALM
it
Includes only those
SITE
103
105
124
20%
21%
14%
OF VARIANCE IN
3SREL HUM
variables explaining at
-
-
-
least
DECISION-TREE ACCOUNTED
AV WD SP D7AM PRS
172
14%
10%
10% of the variance
in
FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES
MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
the
CONDITIONS ASSOCIATED WITH HIGH
high
high
N to
E and
MAX TEMP (1); low AV WD
MAX TEMP (1); low AV WD
S WIND DIR (1);
SE WIND DIR (1)
high MAX
; low AV
SP (2)
SP (1)
TEMP
WD SP
; high #DY
; high #DY
CALM (1);
CALM (2);
(1); low D7AM PRS (2)
(1); high #DY CALM (
low
low
26%
21%
19% 21% 24%
15% 23% 22% 11%
26% 267, 141
decision-tree.
IP CONCENTRATIONS**
%REL HUM.
MIXHT AM (1)
; high
; high #DAYS PP (1); N
1); high D7AM PRS (1).
MIXHT PM (1); high SREL HUM.
to SE WIND DIR (1); low %REL HUM (1);
Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits in
the decision-tree made on this parameter.
-------
TABLE 5.8 RELATIONSHIP OF FINE TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
<£>
SITE
103
1G5
134
PERCENT OF VARIANCE IN
ffDAYS PP ttm CALM %REL HUM
20%
25%
32%
Includes only those variables explaining at least
SITE
103
105
124
DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES*
AV WD SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
14% - 26% 24% 15% 16%
16* - 21% 18% 167, 16%
20% 15% 23% 20% -
10% of the variance in the decision-tree.
CONDITIONS ASSOCIATED WITH HIGH FINE CONCENTRATIONS**
high MIN TEMP (1); low AV WD SP (2);
high MIN TEMP (1); HE to S WIND DIR
high #DY CALM (4); high MIN TEMP (1)
high MAX TEMP (1); high #DY CALM (3); ESE to S WIND DIR (1).
(1); high *DY CALM (3); low AV WD SP (5); S to SW WIND DIR (1); high D7AM PRS (1); low %REL HUM (1).
; high i»DAYS PP (1).
**
"
taber '" »«"«.«« represents the nu.ber of splits in
-------
TABLE 5.9 RELATIONSHIP OF COARSE TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
00
PERCENT OF VARIANCE IN
SITE #DAYS PP #DY CALM %REL HUH
103
105 - 13%
124
DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES
AV WD SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
14% 20% 14% 19% 19%
16% 13% 22% 19%
- 13% 12%
'includes only those variables explaining at least 10% of the variance in the decision-tree.
SITE CONDITIONS ASSOCIATED WITH HIGH COARSE CONCENTRATIONS
103 high MAX TEMP (2); low AV WD SP (3);
105 high MAX TEMP (2); high *DY CALM (1);
124 N to S WIND DIR (1); low %REL HUM (2)
high IDAYS PP (1).
low AVWDSP (3); low MIXHT AM (1).
; low MIN TEMP (3); high MAX TEMP (1); low D7AM PRS (1); high #DAYS PP (1).
"Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits in
the decision-tree made on this parameter.
-------
TABLE 5.10 RELATIONSHIP OF SULFUR TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
SITE
103
105
124
PERCENT OF VARIANCE IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES
#DAYS PP #DY CALM %REL HUM AV WD SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
16%
25%
35%
*
Includes only those variables explaining at least
SITE
103
105
124
11% 272 24% 13%
25% zn 10% 11%
14% 12% 11% 10%
10% of the variance in the decision-tree.
CONDITIONS ASSOCIATED WITH HIGH SULFUR CONCENTRATION**
high MIN TEMP (1); high MAX TEMP (1)
high MIN TEMP (2); high #DYCALM(2);
high *DY CALM (4).
; low AV WD SP (1); high *DY CALM (2).
high rfDAYS PP (1).
Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits in
the decision-tree made on this parameter.
-------
TABLE 5.11 RELATIONSHIP OF SILICON TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
o
o
SITE
103
105
124
*
Includes
SITE
103
105
124
PERCENT OF VARIANCE
^OAYS PP #DY CALM %REL HUH
14S
10%
only those variables
low %REL HUM (4)
low %REL HUM (1)
low %REL HUM (2)
-
explaining
28%
19%
19%
IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES
AV WD SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
16% 16% 17%
18% 13% 20% 21%
at least 10% of the variance in the decision-tree.
CONDITIONS ASSOCIATED WITH HIGH SILICON CONCENTRATION**
; high #DAYS PP (2)
; low D7AM PRS (1);
; low MIN TEMP (1);
; high MAX TEMP (1); high MIXHT PM (1); low AV WO SP (1).
high MAX TEMP (3); high #DAYS PP (1); high MIN TEMP (1); low AV WD SP (1).
high #DAYS PP (1); E to S WIND DIR (1); high %REL HUM (1).
Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits in
the decision-tree made on this parameter.
-------
TABLE 5.12 RELATIONSHIP OF ALUMINUM TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
SITE
103
105
124
it
Includes
SITE
103
105
124
?
-------
TABLE 5.13 RELATIONSHIP OF CALCIUM TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
o
ro
SITE
103
105
124
PERCENT
#DAYS PP #DY CALM
22%
13%
12%
OF VARIANCE IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES
%REL HUM AV WO SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIP1T KINO DIR
16%
"includes only those variables explaining at least 10% of the variance in the
SITE
103
105
124
high
high
low
low
#DAYS PP (2); low
#DAYS PP (1); low
MIXHT AM (2), high
WIN TEMP (1); high
CONDITIONS ASSOCIATED WITH HIGH
%REL HUM (1); high MAX TEMP (1); low MIN
3.REL HUM (2); low AV WD SP (2); NE to S
mixht PM (1).
#DAYS PP (1); high MAX TEMP (1); low AV
12% 25% 19%
lit 25% 19%
18% 11%
decision-tree.
CALCIUM CONCENTRATION**
TEMP (1); low AV WD SP (3); high MIXHT PM (1).
WIND DIR (1); high #DY CALM (2); S and SW WIND DIR (1); low PRECIPIT (1);
WD SP (1).
^Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits 1n
the decision-tree made on this parameter.
-------
o
co
SITE
103
105
124
*DAYSPP
TABLE 5.14 RELATIONSHIP OF LEAD TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
PERCENT OF VARIANCE IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES*
%REL HUM AV WD SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
13% -
11%
10%
12%
14%
Includes only those variables explaining at least 10% of the variance in the decision-tree.
SITE
103
105
124
CONDITIONS ASSOCIATED WITH HIGH LEAD CONCENTRATION*
low AV UD SP (1).
low AV WO SP (2); high #DY CALM (1 ); high D7AM PRS (1); high MAX TEMP (1); low #DY CALM (1)
low MAX TEMP (1); low MIN TEMP (2); SW and W WIND DIR (1); WEST WIND DIR (1); high MIXHT PM (1); high tfDAYS PP (1); low D7AM PRS (1).
°f
1n decision-tree. Nu.ber in parentheses represents the number of splits
-------
SITE
TABLE 5.15 RELATIONSHIP OF VANADIUM TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
PERCENT OF VARIANCE IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES
*DAYS PP #OY CALM %REL HUM AV WO SP 07AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT WIND DIR
103
105
124
10*
12%
*Includes only those variables explaining at least 10% of the variance In the decision-tree.
SITE
CONDITIONS ASSOCIATED WITH HIGH VANADIUM CONCENTRATION
103
105
124
high MIN TEMP (1); low AV WD SP (1).
low MAX TEMP (1); low MIN TEMP (1).
low MIN TEMP (1).
*Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits in
the decision-tree made on this parameter.
-------
SITE
TABLE 5.16 RELATIONSHIP OF TITANIUM TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
PERCENT OF VARIANCE IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES*
iOAYS PP rfOY CALM '/.REL HUM AV WD SP D7AK PRS MAX TEMP M1N TEMP M1XHT AM MIXHT PM PRECIPIT WIND DIP.
103
105
124
132
10',.
Includes only those variables explaining at least 10% of the variance in the decision-tree.
SITE
CONDITIONS ASSOCIATED WITH HIGH TITANIUM CONCENTRATION
O
01
103
105
124
high MAX TEMP (2); low SREL HUM (1); low MIN TEMP (2); low AV WO SP (1); low #OY CALM (1).
SE to SW WIND DIR (1); low AV WD SP (1); high MAX TEMP (1); high D7AM PRS (1); low MIN TEMP (2).
low XREL HUM (1); low MIN TEMP (1).
Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits in
the decision-tree made on this parameter.
-------
TABLE 5.17 RELATIONSHIP OF IRON TO METEOROLOGICAL CONDITIONS
AS FOUND IN THE DECISION-TREE ANALYSIS.
o
CT>
SITE
103
105
124
*
Includes
SITE
103
105
124
PERCENT OF VARIANCE IN DECISION-TREE ACCOUNTED FOR BY INDIVIDUAL METEOROLOGICAL VARIABLES
*DAYS PP rDY CALM '{REL HUM AV WD SP D7AM PRS MAX TEMP MIN TEMP MIXHT AM MIXHT PM PRECIPIT V.INQ DIR
19% - - 13%
12% - - - 13%
15%
only those variables explaining at least 10% of the variance in the decision-tree.
CONDITIONS ASSOCIATED WITH HIGH IRON CONCENTRATION**
low XREL HUM (3); low AV WD SP (3); low MIXHT AM (I); high #DAYS PP (2); high MIXHT
low #DAYS PP (1).
high MAX TEMP (1), high tm CALM (1).
low MIN TEMP (1); high *DAYS PP (1); E to S WIND DIR (1); low ?=REL HUM (4); E and SE
low AV WD SP (1); high MAX TEMP (1); low MAX TEMP (1); low D7AM PRS (1).
23% 20%
19% 17%
11%
PM (1); high MAX TEMP (1); high *DY CALK (1);
WIND DIR (1); NE to SE WIND DIR (1);
*Meteorological parameters are listed in approximate order of appearance in decision-tree. Number in parentheses represents the number of splits in
the decision-tree made on this parameter.
-------
FINE (Table 5.8)
The two most salient features of the decision-tree for FINE,
high #DY CALM and high temperatures, are the same as for sulfur.
This is consistent with earlier findings (e.g. Chapter 1) that sul-
fate comprises approximately half of FINE. Prolonged stagnations
and high temperatures would also promote the formation of other
secondary aerosols (e.g. nitrates and secondary organics) which
likely contribute to FINE. A condition that is important to high
FINE at sites 103 and 105 -- low average wind speed -- may reflect
localized contributions of primary (or possibly secondary) fine
aerosols from industrial sources and fuel combustion. The signi-
ficance of easterly to southerly flows for elevated FINE at sites
103 and 105 may point toward the industrial areas of East St. Louis
or, possibly, synoptic sulfate episodes (from the dense SOx emissions
in the Ohio Valley and/or high pressure systems from the Gulf which
are known to be conducive to high sulfate). All in all, the
decision-trees for FINE implicate man-made sources, especially
secondary aerosols, especially sulfates.
Silicon, Aluminum, Calcium (Tables 5.11, 5.12, and 5.13)
The overwhelming factor contributing to high levels of the'
crustal elements is dryness -- low relative humidity, high number
of days since precipitation, and low minimum temperature with high
maximum temperature (an indication of low relative humidity). The
importance of dryness implies that fugitive dust, rather than fly
ash, is probably the major source of these elements. A secondary
feature of the decision-trees for the crustal elements, especially
at the urban sites and especially for calcium, is calm winds (low
AV WD SP and high #DY CALM). This probably* implicates anthropo-
genic sources of dust (e.g. traffic, industries, quarries, con-
struction, etc.) rather than wind blown dust, particularly for the
urban sites and particularly for the limestone component.
COARSE (Table 5.9)
The meteorological conditions producing high levels of COARSE --
dryness (low %REL HUM, high #DAYS PP, and low MIN/high MAX TEMP) and
low wind speed are similar to those producing high levels of the
crustal elements. This agrees with earlier statements (e.g. Dzubay's
results in Chapter 1) that crustal material is the predominant con-
contributor to COARSE. As noted in the above paragraph, the suspended
crustal material appears to be mostly fugitive dust rather than fly
ash. Low wind speeds are more important to COARSE than they are to
the crustal elements, apparently reflecting the fact that other
*
Another explanation for the negative dependence of crustal elements on wind
speed involves a sampling problem. Specifically, the collection efficiency
of particles near the upper size cut-off (^20 ym) tends to decrease with
increasing wind speed.
107
-------
man-made sources (in addition to the crustal component) make some
contribution to COARSE.
IP (Table 5.7)
As expected, the meteorological conditions associated with
high IP appear to be a mixture of the conditions associated with
high FINE and high COARSE. High temperatures and calm conditions
reflect the contributions from FINE (and sulfates). Dry conditions
and low wind speeds reflect the contributions from COARSE (and
crustal materials).
TSP (Table 5.6)
The decision-tree classes for high TSP point toward one major
factor -- dryness -- indicating that TSP tends to be dominated by
fugitive dust. We would expect TSP to contain relatively more
dust than IP because TSP includes larger particles (i.e. particles
in the 20 ym to 50 ym size range as well as those below 20 ym).
Other Elements: Pb, V, T, and Fe (Tables 5.14 to 5.17)
The decision-trees for the last four elements point toward the
specific sources associated with these elements, but these decision-
trees are not important in explaining IP or TSP because none of
these sources appear to be major contributors to IP or TSP. The
decision-tree classes for high lead (a tracer for automotive ex-
haust) show poor dispersion as the major theme. The one prevalent
feature in the decision-trees for vanadium (a tracer for fuel oil)
is low temperature; this is expected because fuel oil use for space
heating is a winter phenomenon and because dispersion of primary
contaminants is often poor in the winter due to strong nocturnal
inversions. The decision-trees for iron and titanium appear rather
similar to those for the crustal elements (Si, Al, and Ca); dryness
and calm winds are the major features. This seems reasonable be-
cause Fe and Ti also serve partly as tracers for crustal materials.
Calm conditions are more important for Fe and Ti, however, because
these elements also serve as important tracers for industrial
sources, iron/steel facilities, and a paint pigment plant.
108
-------
6.0 RELATIONSHIP BETWEEN TSP AND IP
The existing National Ambient Air Quality Standards (NAAQS) for
participate matter pertain to total suspended particulate mass (TSP) as
measured by the Hi-Vol sampler. Because of compliance monitoring for the
NAAQS, and because of the widespread use of Hi-Vol samplers even before
promulgation of the NAAQS, there is a wealth of historical data on ambient
TSP concentrations. In the future, EPA may revise the NAAQS by establish-
ing a size-specific particulate standard for inhalable particulate matter
(IP) and possibly fine particulate matter (FINE) as measured by the
dichotomous sampler (Miller et al., 1979). If this is done, there will
be a need to understand the relationship between TSP and IP so that the
historical Hi-Vol data can be viewed in better perspecitve relative to the
new dichotomous data. This chapter provides some information on the
relationship between TSP and IP by analyzing the simultaneous Hi-Vol and
dichotomous data at the ten RAMS particulate monitoring sites.
6.1 SCATTERPLOTS OF DICHOTOMOUS DATA VERSUS HI-VOL DATA
One way of investigating the relationship between dichotomous and
Hi-Vol measurements is to compare day-to-day variations in the respective
data at individual monitoring sites. This can be done by preparing scatter-
plots and computing least-squares regression lines relating dichotornous
variables to Hi-Vol mass at each of the RAMS locations.
Table 6.1 summarizes the results of regression/correlation analyses
relating IP, FINE, and COARSE to TSP. Although the correlations for each
variable are statistically significant at all locations, TSP explains (on
the average among sites) only 47%, 50%, and 24% of the daily variation in
IP, COARSE, and FINE, respectively. The lack of an extremely close
association between TSP and the dichotomous variables can also be illus-
trated by scatter plots. Figures 6.1, 6.2, and 6.3 present such plots for
sites 108 (typical correlation), 115 (best-case correlation), and 124
(worst-case correlation), respectively. It is obvious from the above
analysis that TSP measurements alone cannot adequately explain the day-to-
day fluctuations in IP, COARSE, and especially FINE.
109
-------
TABLE 6.1 CORRELATION/REGRESSION ANALYSIS BETWEEN DAILY
VALUES OF DICHOTOMOUS AND HI-VOL DATA.
Table 6.la TP Versus TSP
STATION
103
105
106
108
DEGREE OF
Correlation
Coefficient
R
.65
.67
.69
.65
112 ! .84
115 . .87
118 .45
120 .55
122
124
Average
.88
.42
.67
CORRELATION
Percent of
Variance Explained
R2
42%
45%
48%
42%
70%
76%
20%
30%
78%
17".
47%
REGRESSION LINE
Slope Intercept
(pg/m3)
0.49
0.51
0.57
0.43
0.57
0.74
0.27
0.47
0.68
0.20
0.49
11.1
-1.5
6.2
11.0
2.3
-4.6
17.1
11.7
-3.4
18.3
fi.8
Table 6.
STATION
103
105
106
108
112
115
118
120
122
124
Average
DEGREE OF
Correlation
Coefficient
R
.72
.77
.71
.70
.86
.89
.44
.57
.89
.30
.69
Ib COARSE Versus TSP
CORRELATION*
Percent of
Variance Explained
R2
52%
59%
51%
49%
75%
79%
20%
33%
79%
9%
50*
REGRESSION LINE
Slope
0.33
0.29
0.29
0.28
0.37
0.44
0.14
0.26
0.40
0.09
0.23
Intercept
(ug/m3)
-0.4
-3.8
1.6
1.5
-5.1
-6.2
7.2
3.0
-5.1
7.6
.03
Table 6.1c FINE Versus TSP
STATION
103
105
106
108
112
115
118
120
122
124
Average
DEGREE
Correlation
Coefficient
R
.41
.51
.55
.53
.59
.57
.31
.29
.62
.33
.47
OF CORRELATION
Percent of
Variance Explained
R2
17%
26%
30%
28%
35S
33%
9%
8%
39"
in
24:;
REGRESSION LINE
Slope Intercept
(yg/m3;
0.17
0.22
0.25
0.16
0.21
0.26
0.12
0.14
0.25
0.11
0.19
10.4
2.9
5.9
8.8
6.4
4.0
9.9
11.5
3.7
10.7
7.4
'tote that all correlation coefficients are statistically significant
at 95:. confidence level.
110
-------
EJ
LO
in
c\i
cs>
13
in _
S)
in
in _
CM
tsi
J I J I I
tsi
LO
CM
ta
a: ^ -
o
CJ
in
in
CM
A A
tSJ
in
CM
I I 1 1
0 50 100 150 200 250
JSP
.0.
in
ir> _
CM
IS3
50 100 150 200 250
JSP
0 50 100 150 200 250
JSP
Figure 6.1 Scatterplot of dichotomous variables
versus TSP at site 108.
Ill
-------
LU
r-j
If)
L'5
CM
IS
cs
in _
IS _
LT)
LO _
CM
IS
O
o
IS)
in
LO
CM
s
IS
in
in
CM
is ~~
oo in _
I 1 I I
50 100 150 200 250
TSP
E3
in
in
CM
i
ts
~"
in _
^-
IS _
in
LO _
CM
IS)
2
1 1 1 1
0
o
°/
o /
o/
°°/ <,
o$> °
o y
o c^^g o
o
-------
63 I 1 1 1
in i ' ' j 1
r\j
ta
IS
i
LJ-I 10
i i
ii_
K)
in
10 _
CM
£
J
A
A £ ^A^ ^^
1 1 1 1
5 50 100 150 200 2E
-
-
-
-
_
is
in -
LO
tv
;0 -
TSP i _
Q. L0
t i ^
(S3
LO -
in
CM
1
KS
i
LU
c2 LO
O
O
S3 _
in
LO _
CM
1 ' ' 1
o
0 O
&5f5^T"o
12 1 1 1 1
IS
LO
LO _
CM
is
0
-L L__1_JL
o
0 O
0
1 1 1 -1 1
50 100 150 200 25
-
-
-
0
TSP
50 100 150 200 250
TSP
Figure 6.3 Scatterplot of dichotomous variables
versus TSP at site 124.
113
-------
6.2 RATIO OF IP TO TSP
Although TSP is not a good predictor of IP in a day-to-day basis,
there may be certain ways (e.g. long-term averages) in which the variables
relate more closely. This section examines the relationship between IP and
TSP in terms of the ratio IP/TSP. This ratio can be used to obtain crude
estimates of IP from Hi-Vol data. Also, the ratio provides information on
the size distribution of particulate matter, because it basically represents
the fraction of particulate mass in the size range of 20-50 microns.
Table 6.2 presents annual and monthly statistics for the IP/TSP ratio.
Three interesting summary statistics for IP/TSP are readily computed from
this table: (1) the average of the annual means among the sites is 0.61,
(2) the site-to-site standard deviation of the annual means is 0.06, and
(3) the day-to-day standard deviation averaged over all sites is 0.22.
These results indicate that, although the ratio varies by ± 36% daily at a
given location, the annual mean ratio varies only ± 10% among stations. Thus,
in the case of St. Louis, TSP tends to be a fairly good predictor of IP on
an annual mean basis even though it is a poor predictor on a daily basis.
The fairly close relationship between annual mean TSP and annual mean IP is
also illustrated by the scatterplot shown in Figure 6.4.
The remainder of this section discusses spatial, temporal, and meteor-
ological patterns in the IP/TSP ratio. The purposes are to examine vari-
ations in the size distribution of particulate matter and to see if patterns
in the ratio suggest better methods of estimating IP from TSP.
6.2.1 Spatial Patterns of IP/TSP
As indicated by Figure 6.5, the IP/TSP ratio does not show an obvious
spatial pattern within the St. Louis region. The lack of a consistent
spatial trend from the center city to rural locations is also illustrated
in Figure 6.6 which plots IP/TSP ratio versus distance from station 103.
Table 6.3 summarizes IP/TSP ratios as a function of site type. Again,
no consistent pattern is obvious. The one urban/commercial site has a low
ratio, and the one suburban/industrial site has a high ratio; definitive
conclusions, however, cannot be reached based on data from single locations.
*It does not exactly represent the mass in the size range less than 20-50 ym
because of particle size - sampling efficiency considerations with both the
dichotomous and Hi-Vol samplers, and because of artifact aerosol formation
on Hi-Vol filters.
-------
TABLE 6.2 ANNUAL AND MONTHLY STATISTICS FOR IP/TSP RATIO.
STATIC:!
KLTO23
AR17H tlSAfl
ARITII SO
2T-3 HIG-SST
AH ITU nSi\?|
FE3
AP.T
122
73
.63
.23
2.07
1.13
.23
7
.44
7
.44
9
.6-3
3
.62
S
.Go
S
1 . IS*
3
So
3
.37
7
.69
f "">
4
.43
7
.33
ins
z\
.49
.20
1.11
1.C3
.20
10
.34
5
.45
10
9
.44
_3
.73*
10
.31ft*
7
.43
5
34-
7
.31
7
.45
5
. TO
5
.37
ins
53
.64
.13
1.13
.93
.14
9
.50
5
.54
3
.53
6
.53
3
.57
7
,C3*
o
.75
3
.73
10-3
60
.50
.27
1.50
1.59
.10
3
.50
3
. 31**
3
.43
-3
.43
3
.33
4
.37
1
V3-
2
.31
5
.53
2
.43
3
.34
4
.73*
uz
75
.61
.20
1.28
1.16
.27
6
.62
i
.48
9
.57
9
.48
8
.57
0
.22*
5
.74
6
.63
4
.46
n
U
.SSi-nv
6
.53
4
.67
115.
67
.65
.13
1.13
1.11
.30
9
7
.62
0
.64
5
.61
3
.53
3
.32
G
.75
S
.63
3
.53
5
.G2
9
.61*
112.
73
.55
.23
1.23
1.13
.11
S
.49
4
.39
3
.43
9
.S2-.V
3
.60
4
.73
3
.53
3
.53
3
.61
7
.42
3
.70
7
.Sift*
120
63
.71
.25
1.46
1.33
.18
3
.52
7
.53
3
.63
7
.70
3
.76
3
.33
4
.37
6
1.07*
2
.74
4
.63
3
.40
5
.55
122
S5
.61
.17
1.02
1.01
.16
3
.52
4
.64
7
.51
6
.45
7
. bo
7
. 34-:.-*
3
.57
2
.57
5
.60
2
.37*
3
.53
9
.51
12A
53
.61
.31
1.46
1.21
.14
11
.38*
5
.43
7
.33
3
.33
.63
7
.55
5
.47
4
.as
:': Month in ::hich iiaxi-un OCC-JPS.
:--.v Ilorrth in uhich sacond highaat occurs.
(It th3 r,ont;i3 coincids only ana month is flsggod.)
115
-------
o.
tO
O)
3
C
c
<:
70-,
60-
50-
40-
30-
20-
' 10-
i
10
I
20
103
108
106
115
120
105
112 f
Station Number
'IP = 3.1 + 0.55 TSP
Correlation Coefficient = 0.92
i
30
I
40
i
50
I
60
70
80
90
100
Annual Mean TSP
Figure 6.4 Relationship between annual mean IP and annual
mean TSP among the RAMS sites.
116
-------
Mississippi River
^^M
.61
Figure 6.5 Geographical distribution of IP/TSP
ratio in the St. Louis area.
117
-------
cc.
.7-
.6-
.5-
.4-
r- -3-
to
3
< .2-
112 118
,?n
124 122
I T
10 20 30 40
Distance to Station 103 (km)
Figure. 6.6 IP/TSP ratio versus distance to site 103,
TABLE 6.3 AVERAGE IP/TSP RATIO AS A
FUNCTION OF SITE TYPE.
Urban
Suburban
Rural
Industrial
mt 1 M
.71
Commercial
.49
Residential
.63
Agricultural
.62
.60
.61
All
.59
.64
.61
118
-------
6.2.2 Temporal Patterns of IP/TSP
Figure 6.7 presents the seasonal pattern of the IP/TSP ratio, averaged
over all 10 RAMS sites. It can be seen that the IP/TSP ratio tends to be
highest in the summer (especially in June), meaning that a greater fraction
of the ambient aerosol is less than 20 microns in the summer. Part of this
effect is due to the high sulfate concentrations during the summer (see
Chapter 4); however, even subtracting out the sulfate influence, we have
found that a summer peak in the ratio still exists (possibly reflecting
photochemical aerosols--other than sulfatewhich may also peak during the
summer).
Figure 6.8 illustrates the lack of a weekly pattern in the IP/TSP
ratio. We have also compared weekend versus weekday levels of the IP/TSP
ratio and have found no significant differences.
6.2.3 Decision-Tree Analysis of IP/TSP
In an attempt to gain a better understanding of the factors affecting
the IP/TSP ratio on a day-to-day basis, two decision-tree analyses were per-
formedone relating daily IP/TSP to dichotomous elemental concentrations,
and the other relating daily IP/TSP to meteorological parameters. The
analyses were conducted for three sites: stations 103, 105, and 124. The
resulting decision-trees are presented in Appendix J.
Table 6.4 summarizes the percent variance explained in IP/TSP by the
decision-trees . It is obvious that, with the possible exception of the
elemental concentration analysis at site 105, the daily variance in IP/TSP
is not explained well by either elemental concentrations or meteorological
parameters. Despite this lack of success in explaining much of the variance
in IP/TSP, the decision-trees did provide some insight into at least one
*
To factor out the influence of sulfate on the IP/TSP ratio, we examined
the variable (IP - SULFATE)/(TSP-SULFATE), where SULFATE = 4.1 x (di-
chotomous sulfur).
**
See Chapter 5 for a discussion of the decision-tree method.
Note that, because the performance of the decision-trees is determined
using independent data sets, the percent variance explained can be
negative.
119
-------
<:
o
o
I- 1/1
+-> O)
03 +->
O£ -i-
oo
D_
i/) oo
D. I
H-4
O)
S-
O)
1.0-
.9-
.8-
.7-
.6
.5-
.4-
.3-
.2-
.1-
M
M
T~
J
"T
A
Month
1 1 1
0 N D
Figure 6.7 Seasonal pattern of the IP/TSP ratio.
s_
o -
oo oo
O)
CT>
(O
s_
a;
.7 -
.6
.5
.4 -
.3-
.2
.1 -
Mon.
Tue.
T
Sat.
Wed. Thr. Fri.
Day of Week
Figure 6.8 Weekly pattern of the IP/TSP ratio.
Sun.
120
-------
factor that affects the IP/TSP ratio. Specifically, the elemental decision-
trees for sites 103 arid 105 indicated that high IP/TSP ratios correlated
with high sulfate (sulfur). Also, high IP/TSP ratios correlated with meteor-
*
ology known or suspected to cause high sulfate (e.g. high temperatures, pro-
longed stagnation, and high relative humidity). This finding agrees with the
results of the seasonal analysis which indicated that elevated levels of sul-
fate and other secondary aerosols might be associated with the occurrence of
high IP/TSP ratios.
TABLE 6.4 PERCENT VARIANCE EXPLAINED IN THE IP/TSP
RATIO BY DECISION-TREE ANALYSES.
PERCENT VARIANCE EXPLAINED1"
Decision-Tree Analysis Using Decision-Tree Analysis Using
STATION NUMBER 8 Elemental Concentrations 11 Meteorological Parameters
103
105
124
17% + 10%
40% t 13%
-28% + 16%
6% + 12%
15% + 10%
-31% ± 37%
fThe reader is referred to Chapter 1 for explicit definitions of the
elemental and meteorological variables.
6.2.4 Conclusions Regarding the IP/TSP Ratio
In this section, we have found that the annual mean IP/TSP ratio aver-
aged among RAMS sites is 0.61, with a site-to-site standard deviation of
t 0.06. Also, the day-to-day deviation away from the annual mean at individ-
ual stations is + 0.22. Beyong these rather simple conclusions in regard to
the IP/TSP ratio in St. Louis, we have found very little. The geographical
patterns, temporal patterns, and decision-tree analyses do not support refined
estimates of the ratio with respect to position, time, or meteorology. The
only possible exception arises from the seasonal pattern, i.e. the ratio ap-
pears to be higher in the summer than in the winter. Even this exception,
however, may be in doubt because recent data for Topeka (an urban area in the
same region of the country) show the opposite of the seasonal pattern observed
in St. Louis (Spengler, 1979).
See Chapter 5 for a discussion of the analysis relating sulfur concentrations
to meteorology.
to meteorology.
121
-------
7.0 SOURCES OF PARTICULATE MATTER IN ST. LOUIS
Evidence concerning the sources of ambient participate matter is
spread throughout this report: Dzubay's chemical element balance in Chapter
1, spatial patterns of the aerosol in Chapter 2, temporal patterns in Chapter
4, aerosol/meteorology relationships in Chapter 5, etc. Because knowledge of
aerosol sources is key to planning effective control strategies, this chapter
synthesizes the evidence and discusses our overall conclusions regarding
aerosol sources. The discussion is organized in two parts -- (1) an overview
of particulate sources for the entire RAMS network, and (2) urban versus
nonurban patterns including considerations of background concentrations and
particulate transport.
7.1 SOURCES OF FINE, COARSE, IP, AND TSP
This section describes the sources of particulate matter in St. Louis,
viewing the RAMS network in aggregate. Secondary aerosol sources are dis-
tinguished from primary aerosol sources, with the primary aerosol sources
further distinguished as to specific source type, e.g. automotove exhaust,
fuel oil combustion, iron/steel industry, various source types of crustal
material, etc.
7.1.1 Fine Fraction of IP
o o
Annual mean concentrations of FINE range from 16 yg/m to 28 yg/m and
3
average 21 yg/m over the RAMS network. Extending Dzubay's chemcial element
balance to the full year of 1976 (see Chapter 1) we find that slightly over
o
50% of FINE, 11 yg/m averaged over the network is ammonium sulfate. That
sulfate aerosols are the single most important source of FINE is also
evidenced by our analysis of spatial patterns, temporal patterns, and
aerosol/meteorology relationships; specifically, the features in the data
for FINE correspond rather closely to the features for sulfur.
Sulfur concentrations at the rural RAMS sites are approximately 75%
of the concentrations at the center-city sites, signifying that the sources
of sulfate are of air basin scale if not synoptic scale. Analyses performed
in Chapter 3 indicate that the majority of sulfate in the St. Louis area
123
-------
apparently results from large scale phenomena and that many of the sources may
be exterior to the St. Louis AQCR. That the sulfate phenomenon is also part-
ly air basin in scale (rather than synoptic in scale) is evidenced by the
moderate urban/nonurban differences, and by the existence of a weekend-
weekday effect (which implies a fairly short response time for the sulfate).
The air basin or synoptic scale nature of sulfate implies that it consists
predominantly of secondary aerosol produced from SC^ emissions rather than
primary sulfate emissions. Electric power plants are implicated as the major
source of sulfate because they contribute about 80% of SO emissions in the
/\
St. Louis AQCR (EPA, 1978) and 70% of SO emissions synoptically (Husar,
A
1979). Sulfate levels are somewhat elevated in the center-city relative to
the countryside due to any of three factors: (1) greater urban SO,, concen-
trations which oxidize to form local sulfate; (2) the greater oxidizing
potential of the urban atmosphere due to the presence of photochemical
smog; and/or (3) the possible importance of primary sulfate emissions from
urban sources.
Dzubay's calculations (Chapter 1), the decision-tree analyses (Chapter
5), and the spatial/temporal patterns (Chapters 2 and 4) imply that no
single type of aerosol dominates the non-sulfate portion of FINE. Rather,
the non-sulfate part of FINE seems to be a mixture of fairly small contri-
butions from numerous (mostly man-made) sources: secondary organic aerosols,
secondary nitrates, automotive exhaust particles, particles from stationary
source fuel combustion, the fine fraction (i.e. the lower tail of the size
distribution) of suspended dust, etc. As discussed in Chapter 5, FINE con-
sists essentially of man-made particulate matter, especially secondary
aerosols, especially sulfates.
7.1.2 Coarse Fraction of IP
3 3
Annual mean concentrations of COARSE range from 13 yg/m to 32 yg/m
3
and average 21 yg/m among the RAMS sites. Dzubay's chemical element bal-
ance (extended to cover the entire year) shows that 83% of COARSE is crustal
material: about two-thirds shale type material and one-third limestone type
material.
The various patterns that we have observed in the data for COARSE and
the crustal elements (Si, Al, and Ca) support Dzubay's conclusion that COARSE
124
-------
is dominated by crustal material, mostly of the shale-type. The analyses in
this report have also provided a more refined assessment of the origins of
this crustal material. The smooth spatial patterns in the crustal elements
(Chapter 2) and the correlation of the crustal elements with dryness (Chapter
5) indicate that ubiquitous area sources of fugitive dust, rather than in-
dustrial sources of fly ash, are the dominant contributors to.the crustal
material. Furthermore, the existence of significant weekend-weekday effects
(Chapter 4) and the correlation of the crustal material with low wind speeds
(Chapter 5) signify that most of the fugitive dust is man-made rather than
wind-blown.
Considering the omnipresence of motor vehicle traffic, one might sus-
pect that road dust constitutes the dominant source of fugitive dust in St.
Louis. However, the strong dissimilarities between the spatial/temporal
patterns for the crustal elements (or COARSE) and lead suggest that motor
vehicles are not the predominant dust source. Rather, an assortment of dust
sources paved roads, unpaved roads, quarries, construction, agriculture,
soil dust, and certain industries -- seems implicated.
Shale-type and limestone-type crustal material each probably arise
from several dust sources in the above list. There is evidence, however,
that the limestone-type dust is more closely tied to urban commercial/industrial
activity than shale-type dust. The greater urban/rural spatial gradient for
*
One can invent scenarios whereby traffic could be the predominant source
of crustal material, yet crustal material and lead would have incongruous
spatial/temporal patterns. For example, the spatial patterns of lead and
road dust could diverge because lead emissions are determined by traffic
levels and percent gasoline vehicles, while road dust depends on traffic
and road conditions (paved versus unpaved, road cleanliness, vehicle
speed, etc.); the two independent factors (percent gasoline vehicles and
road conditions) could destroy the spatial correlation of road dust and
lead. Also, the diurnal patterns of lead and crustal elements may differ
because high afternoon mixing heights will disperse the fine lead particles
producing the observed afternoon miminum in Pb, while the afternoon effect
on coarse dust particles will be to keep more suspended (due to unstable
conditions) as well as to disperse them, leading to little net impact
on the crustal elements. Despite the potential validity of these argu-
ments, we interpret the data to imply that road dust is not likely the
predominant source of the crustal material.
125
-------
calcium (Chapter 2) and the greater weekend effect for calcium (Chapter 5)
suggest that crustal limestone stems relatively more from cement manufactur-
ing and construction, while crustal shale stems more from traffic, agri-
culture, and wind-blown dust.
Dzubay's chemical element balance shows that the remaining, non-crustal
17% of COARSE stems from a wide variety of sources (mostly man-made). Our
analyses do not reveal very much about this remaining 17% except to re-
emphasize Dzubay's findings that it is mostly man-made (see decision-tree
analyses in Chapter 5). This non-crustal fraction may not be very signifi-
cant in terms of the total mass of COARSE, but it may nevertheless be
important to health and welfare effects because its chemical composition
differs from that of crustal material.
7.1.3 Total IP
Annual mean concentrations of IP range from 29 to 60 yg/m among the
RAMS sites and average 42 yg/m3. The major sources of IP are the major
sources of its two constituents, FINE and COARSE, each of which comprises
50% of IP. The chemical element balance, modified to encompass all of 1976,
implies that over three-fourths of IP consists of sulfate (29%) and crustal
material (47%). Discussions of the specific sources of sulfate, other parts
of FINE, crustal material, and other parts of COARSE are presented in the
previous two subsections.
It is noteworthy that nearly half of IP in St. Louis consists of
crustal material and that most of the crustal material is apparently dust
related. One implication of this result is that going from a TSP air
quality standard to an IP air quality standard will not completely eliminate
the fugitive dust problems associated with the TSP standard.
7.1.4 TSP
3
Annual mean concentrations of TSP range from 53 to 96 yg/m and
average 79 yg/m over the RAMS network. The relationship between parti-
culate air quality and meteorology (Chapter 5) indicates that TSP is more
influenced by dust than IP; in fact, dust appears to be the dominant source
of TSP. This makes sense because TSP essentially consists of IP plus
126
-------
particles in the 20 to 50 pm size range; and one would expect these larger
particles to be composed almost completely of dust.
The spatial pattern of TSP concentrations among the RAMS stations ex-
hibits very smooth spatial gradients, suggesting that the influence of
localized sources tends to be small. This reflects the fact that the RAMS
stations are generally sited away from strong local sources (fugitive dust
or other types). In this sense, the RAMS particulate sites are not repre-
sentative of existing Hi-Vol sites nationwide, many of which are signifi-
cantly affected by local sources (especially road dust) because of their
proximity to streets, parking lots, and other particulate sources (Record
et al., 1976; Pace, 1978; Price etal., 1977)?
7.2 URBAN VERSUS RURAL SOURCES OF IP
In the sense that crustal material and sulfates are the two major
components of IP, the origins of IP are generally similar at the five urban
sites (103, 105, 106, 108, and 112) and the two rural sites (122 and 124).
As evidenced by the urban/rural ratios in crustal elements (see Table 2.2),
the relative importance of crustal material is minutely greater at urban
sites than rural sites, basically because limestone-type dust is relatively
more important at the urban sites. Sulfate, on the other hand, is slightly
more significant at nonurban locations; ammonium sulfate comprises 27% of
urban IP but 33% of rural IP. Essentially because of the sulfate pattern,
FINE makes up 49% of urban IP but 53% of rural IP.
There are, however, some notable differences in the sources of urban
and rural particulate matter. The spatial patterns of lead, vanadium,
titanium, and iron imply that primary aerosols from auto exhaust, fuel
combustion, and industrial sources are typically a factor of three to five
times greater at urban sites than rural sites. Although the particulate
matter from these sources is not a major portion of IP mass, it may be of
concern because of its special chemical composition (i.e. it may be more toxic
In a study of 154 Hi-Vol sites in 14 cities, Record et al. (1976) found
that 45% of the sites had some degree of local impact, 22% had major
local impacts, and 8% had "undue" local impacts.
127
-------
than crustal material). Also, as evidenced by air quality/meteorology
relationships (Chapter 5) and weekend-weekday effects (Chapter 4) for the
crustal elements, the specific sources of crustal material appear to differ
significantly from urban to rural sites. Urban dust sources (road dust,
construction, cement manufacturing, industrial dust, etc.) are relatively
more important in the city, while agriculture and wind-blown dust are rela-
tively more important in the countryside. Of course, we generally expect
natural sources to contribute a larger fraction of the particulate matter
at rural locations.
The analysis of background concentrations and particulate transport
(Chapter 3) indicated that transport of aerosols from the urban area did
not have a major effect on the rural sites. On an annual mean basis,
transport from the urban area appears to contribute less than 5 yg/m3 of
aerosol (basically FINE) to the rural sites. Considerable uncertainty
existed in our estimates of the transport effect as well as in our estimates
of background concentrations. Annual mean background concentrations
(representing contributions from natural sources and from man-made sources
exterior to the St. Louis AQCR) appear to be approximately 35 yg/m3 for TSP
and 20 yg/m3 for IP.
128
-------
8.0 REFERENCES
Dzubay, R.G., "Chemical Element Balance Method Applied to Dichotomous
Sampler Data," Prepared for the Symposium on Aerosols: Anthropogenic and
Natural-Sources and Transport, January 9-12, 1979.
Eldon, J. A., et al., "Statistical Analyses of the St. Louis Regional Air
Pollution Study Data - Description and Validation of Dichotomous Sampler
Data," Interim Report prepared at Technology Service Corporation, Santa
Monica, CA, under EPA Contract #68-02-2391 Task 4, April, 1979.
EPA, "Position Paper on Regulation of Atmospheric Sulfates," EPA-450/2-75-
007, September 1975.
EPA, "1978 National Emissions Report," EPA-450/2-78-020, May 1978.
Frank, N and N. Posseil, "Seasonality and Regional Trends in Atmospheric
Sulfates," Presented before the Division of Environmental Chemistry,
American Chemical Society, San Francisco, CA. September 1976.
Friedlander, S.K., "Chemical Element Balances and Identification of Air
Pollution Sources," Environmental Science and Technology. 7: 235-240, 1973.
Gatz, D.F., "Relative Contributions of Different Sources of Urban Aerosols,"
Atmospheric Environment, 9: 1-18, 1975.
Gordon, G., Department of Chemistry, University of Maryland, College Park,
Maryland, Personal Communication, 1979.
Goulding, F.S., et al., "Aerosol Analysis for the Regional Air Pollution
Study," EPA-600/4-78-034, July 1978.
Hern, D.H. and M.H. Taterka, "Regional Air Monitoring System Flow and
Procedures Manual," Prepared at Rockwell International, Creve Coeur, MO,
for EPA Environmental Sciences Research Laboratory, under Contract #68-
02-2093, August 1977.
Hopke, P., Institute for Environmental Studies, University of Illinois,
Urbana, Illinois, Personal Communication, 1979.
Husar, R., Department of Mechanical Engineering, Washington University,
St. Louis, MO, Personal Communication, 1979.
Kowalczyk, G.S., et al., "Chemical Element Balances and Identification of
Air Pollution Sources in Washington, D.C.," Atmospheric Environment, 12:
1143-1153, 1978.
Miller, M.S., et al., "A Chemical Element Balance for the Pasadena Aerosol,"
Journal Colloid Interface Sci.. 39(1): 165-176, 1972.
129
-------
Miller, F J., et al.t "Size Considerations for Establishing A Standard for
Inhalable Particles," Journal of the Air Pollution Control Association.
29(6): 610-615, 1979.
Nelson E , "Regional Air Pollution Study - High Volume Filter Measurements
of Suspended Particulate Matter," EPA-600/4-79-003, January 1979.
Pace, T.G., "An Approach for the Preliminary Assessment of TSP Concentra-
tions," EPA-450/2-78-016, July 1978.
PEDCO Environmental, "TSP Source Inventory Around Monitoring Sites in
Selected Urban Areas, St. Louis (Draft)," Prepared under EPA Contract #68-
02-2603 Task 15, Cincinnati, OH, August 1978.
Price, J.P., et al., "Attainment Analysis, Volume I, Causes of Nonattainment,"
Texas Air Control Board, Austin, TX, January 1977.
Record, F.A., et al., "National Assessment of the Urban Particulate Problem,
Volume I, Summary of National Assessment," EPA-450/3-76-024, July 1976.
Spengler, J., Environmental Health Sciences Department, Harvard School of
Public Health, Harvard University, Boston Mass., Personal Communication,
1979.
Stevens, R.K., et al., "Sampling and Analysis of Atmospheric Sulfates and
Related Species," Atmospheric Environment. 12: 55-68, 1978.
Throgmorton, J.A. and K. Axetell, "Digest of Ambient Particulate Analysis
and Assessment Methods," EPA-450/3-78-113, September 1978.
Trijonis, J., et al., "Statistical Analysis of TSP and Meteorological Data
in EPA Region 6," EPA-906/9-79-005, February 1979.
White, W.H. and P.T. Roberts, "On the Nature and Origins of Visibility-
Reducing Aerosols in the Los Angeles Air Basin," Atmospheric Environment,
11: 803, 1977.
130
-------
TECHNICAL REPORT DATA
(I'leair read ln\trui luuis on the revrrii- before ftim/ilrlmxj
1 RFPORI NO
EPA- 4 50/4-80-0 0_6a
4 TITLE AND SUBTITLE
Analysis of the St. Louis RAMS Ambient Particulate
Data, Volume I: Final Report
6. PERFORMING ORGANIZATION CODE
3 RFCIPItNT'S ACCESSION NO.
5. REPORT DATE
February 1980
7 AUTHOR(S)
John Trijonis et al.
8. PERFORMING ORGANIZATION REPORT NO
9 PERFORMING ORGANIZATION NAME AND ADDRESS
Technology Service Corp.
2811 Wilshire Blvd.
Santa Monica, CA 90403
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
SPONSORING AGENCY NAME AND ADDRESS
US Environmental Protection Agency
Research Triangle Park, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
15 SUPPLEMENTARY NOTES
Project Officer: Thompson G. Pace
is ABSTRACT In th1s report, a variety of data analysis methods are used to study the 1976
particulate data from the Regional Air Monitoring System (RAMS) in St. Louis. The
aerosol data, collected at ten sites, include hi-vol measurements of total suspended
particulate mass (TSP), as well as dichotomous sampler measurements of inhalable parti -
culate mass (IP). IP is subdivided into fine particles (less than 2.4 ym in diameter)
and coarse particles (between 2.4 and 20 ym in diameter). This study also includes
dichotomous sampler data for eight trace elements (S, Si, Al, Ca, Pb, V, Ti, and Fe) an
data for 11 meteorological parameters.
The analyses characterize the spatial pattern of particulate matter in-and-near
St. Louis; background aerosol concentrations and particulate transport; temporal
patterns of particulate concentrations, the dependence of aerosol concentrations on
meteorology; and the relationship between hi-vol data and dichotomous data.
Averaged over the RAMS network, IP mass consists of 50% fine particles and 50%
coarse particles. Sulfates, secondary aerosols occurring on a large (air basin and
synoptic) scale, constitute 53% of fine mass and 29% of IP. Crustal material mostly
consisting of manmade fugitive dust, comprises 83% of coarse mass and 47% of IP That
sulfate is the major source of fine mass and that crustal material dominates coarse
mass are the two major themes apparent in the spatial, temporal, and meteorological
Patterns Of the dirhntotnnuf; Hata y
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
particulate matter
aerosol characterization
Inhalable Particulate
b.IDENTIFIERS/OPEN ENDED TERMS
hi-volume sampler
dichotomous sampler
RAPS
sources
spatial patterns
temporal patterns
c. COSATI ricld/Group
19. SECURITY CLASS (This Report)
21. NO. OF PAGES
Unlimited
20. SECURITY CLASS (Thispage)
22. PRICE
EPA Form 2220-1 (Rev. 4-77) PREVIOUS EDITION is OBSOLETE
------- |