March, 1975
Final Report
HAZE FORMATION:
ITS NATURE AND ORIGIN
pBattelle
Columbus Laboratories
A report of research conducted for the
Coordinating Research Council Inc.
and
U.S. Environmental Protection Agency
-------
FINAL REPORT
HAZE FORMATION:
ITS NATURE AND ORIGIN
to
COORDINATING RESEARCH COUNCIL INC.
and
U.S. ENVIRONMENTAL PROTECTION AGENCY
March, 1975
by
David F. Miller, Warren E. Schwartz,
James L. Gemma, and Arthur Levy
BATTELLE
Columbus Laboratories
505 King Avenue
Columbus, Ohio 43201
-------
ACKNOWLEDGMENTS
We would like to acknowledge the help and guidance of the people who served
on the CAPA 6-68 Project Committee:
Mr. B. S. Bailey — Chairman
Dr. R. L. Bradow
Dr. M. K. Gest
Dr. L. C. Gibbons
Dr. T. P. Goldstein
Dr. P. J. Groblicki
Mr. W. Lonneman
Dr. H. W. Otto
Mr. R. Patterson
Dr. T. R. Powers
Mr. J. W. Shiller
Dr. J. Vardi
Dr. J. Wagman
Dr. E. E. Weaver
Dr. F. T. Weiss
Mr. A. E. Zengel
IATTELLE — COLUMBUS
-------
TABLE OF CONTENTS
Page
INTRODUCTION 1
BACKGROUND 2
Characteristics of Haze 2
Chemical Characteristics of Aerosols 5
SCOPE OF PROGRAM 6
First Year 6
Second Year 7
Third Year 7
MAJOR FINDINGS AND CONCLUSIONS 8
Light-Scattering Interpretations 8
Features of Aerosol Composition 10
RESULTS AND DISCUSSION 12
Light-Scattering Measurements 12
Statistical Analysis of Light Scattering 12
Preliminary Analyses 14
Time-Series Analysis 15
Forecasting 16
Summary 19
Chemical Composition of Light-Scattering Aerosols 35
Inorganic Composition 35
Organic Composition 41
Solvent-Extractable Paniculate Matter 44
Weight Percent C, H, and N 46
Infrared Spectra 45
Aromatic/Aliphatic Ratio 53
Acid/Base Neutral Distribution 53
Functional Group Analyses 55
Summary 57
REFERENCES 61
APPENDIXES
A. STATISTICAL METHODS A-1
B. ORGANIC ANALYTICAL PROCEDURES B-1
C. INFRARED SPECTRA OF ORGANIC PARTICULATE C-1
SATTELLE — COLUMBUS
-------
FINAL REPORT
on
HAZE FORMATION: ITS NATURE AND ORIGIN
to
COORDINATING RESEARCH COUNCIL INC.
(CAPA 6-68)
and
U.S. ENVIRONMENTAL PROTECTION AGENCY
(Contract No. 68-02-0792)
from
BATTELLE
Columbus Laboratories
March, 1975
INTRODUCTION
In 1970, when this program was undertaken by Battelle-Columbus for the Coordinating Research
Council and the Environmental Protection Agency, very little was known regarding the nature of
aerosols responsible for haze — a condition which accompanies other manifestations of smog and
often serves as a visual indicator of the severity of smog. About that time an important aerosol
study was conducted in the Pasadena area in which nearly all available tools were utilized in an
intensive monitoring and analytical effort to characterize the airborne paniculate in the Los
Angeles Basin. In spite of the breadth of that project, many gaps remained in our knowledge,
particularly regarding the chemical composition of aerosols and the predominant factors influ-
encing aerosol formation. Frustrated by the initial goal of developing a direct means of determin-
ing the total automotive contribution to haze in urban areas, the subject program was redirected
to investigate more generally the aerosols responsible for haze with emphasis placed on organic
analyses and analyses of the factors influencing light scattering. Although we feel that the
results of this program have significantly advanced our overall understanding in these areas of
aerosol science, analytical and interpretive difficulties still abound, and a thorough accounting of
the nature and origins of haze requires continued investigations.
The program was conducted as a 3-year study beginning in June, 1970, and ending December,
1974. There was a 1-year interruption in the work between the first-year and second-year pro-
grams. Reports summarizing the progress during the first 2 years were issued separately in 1972^)
and 1973(2). This report covers the third year of progress, and for convenience of discussion,
includes summaries of data from the Second Year Report. The principal sections of the report
cover results on statistical analyses of light-scattering data and chemical analyses of light-scattering
aerosols. Preceding these sections we have included short sections covering the phenomena of haze,
some related,research results, the scope of the program, and the major findings of the study.
BATTELLE — COLUMBUS
-------
BACKGROUND
Characteristics of Haze
Haze, by definition, is a quality of indistinctness — in the context of atmospheric science it
refers to a blurring or reduction in the perception of distant objects. The visual effect of haze is
similar to that of fog, although a color effect is more frequently observed with haze. In contrast
with haze, fog consists principally of water droplets, although it is well established that foreign
particles (condensation nuclei) are usually required to nucleate water vapor in advance of conden-
sation to fog. Aerosols associated with haze may also contain water but are comprised primarily
of solids and other liquid materials which have either been emitted into the atmosphere from a
variety of sources, or produced in the atmosphere as a result of photoinduced reactions of gaseous
pollutants. The latter category of aerosols is commonly referred to as secondary or photochem-
ical aerosols.
Visibility problems associated with haze have been recognized for some time. The signifi-
cance of haze in comparison with fog is shown in Figure 1 in which the yearly occurrences of
days with visibility <7 miles in New York City is plotted from 1934 to 1960.^ The number
of poor visibility days attributable to fog has remained fairly constant while the occurrence of
haze has increased dramatically since 1946.
»• L -^ —
34 36 38 40 42 44 46 48 50 52 54 56 58 60
Year
FIGURE 1. FREQUENCY OF POOR VISIBILITY DAYS IN
NEW YORK CITY ATTRIBUTABLE TO HAZE
AND FOG*3'
Evaluation of visibility is mo$t frequently performed by human observation, and, aside from
the limitations of subjective measurements, visual quality has some inherent complexities. Among
the complications is the fact that particles scatter more light in a forward direction than in a
backward direction. Thus the same aerosol cloud may appear more dense when viewed toward
the sun than away from the sun. As alternatives to direct observations, light-transmission instru-
ments (transmissometers) are commonly used for estimating visual quality. The instruments
BATTELLE — C O I. U M B U
-------
measure the extinction of visible light over a large distance which must in turn be related to ob-
served visibilities. In most cases the Koschmeider Formula (visual range = 3.9/extinction coeffici-
ent) is used for such calculations. Assumptions inherent in the formula and conditions affecting
its application have been reviewed.^) Light extinction, as measured directly or by inference from
such instruments, is necessarily an average value over a considerable distance in which many in-
homogeneities may exist. The shortcoming of such determinations is that they cannot be readily
related to other air-quality parameters determined at point sources.
In this program, haze or visual quality was estimated using an integrating nephelometer -
a method conceived by Beuttell and Brewer'^) and later modified for air monitoring by Ahlquist
and Charlson'6). The instrument measures the volume scattering coefficient of a discrete air
sample pulled through an optical chamber. The volume scattering coefficient {bscat) is deter-
mined by integration of the scattering over nearly half a solid angle.
It should be noted thet the nephelometer measures the extinction of light due to scattering
alone while the total extinction of light along an atmospheric path occurs by both scattering and
absorption of gases and aerosols. Rayleigh scattering, which dominates in clean air, occurs from
air molecules and particles approaching molecular size. Mie scattering is important for particles
whose dimensions are roughly comparable to the wavelength of visible light. Thus aerosols in the
size range 0.1 to 1-jum diameter are extremely effective Mie scatterers. Absorption of light by
gases and particles may be significant in polluted atmospheres, but, except for unusual cases,
Mie scattering by aerosols largely dominates the effect of visibility degradation. Rayleigh
scattering as determined by the integrating nephelometer is relatively constant and corresponds
to a bscat value of 0.2 x 10"^m"^ (bscat units of 10"^m~^ are commonly used with the
nephelometer).
Many investigations have been conducted to associate nephelometry data with visibility and
aerosol mass-loading data. While the relationships of bscat with total participate mass concentra-
tions are not always good, as might be expected, the relationships with visibility are generally quite
good. In a recent study conducted in the Los Angeles, Oakland and Sacramento areas, the corre-
lation coefficients between bscat and prevailing visibility* ranged from 0.84 to 0.91 for the three
sites. (?)
For convenience in relating bscat values to visual range, a scale is provided in Figure 2 based
on the empirical relationships developed by Charlson for the nephelometer.^) Light scattering
values for clean air and Freon 12 calibration are indicated.
Aside from the studies relating light scattering to aerosol concentrations and visibility, few
studies have concentrated on interpreting light-scattering data in relation to other air-pollution
phenomena. In some early work by Buchan and Charlson in Seattle'9^ significant correlations
were established between 10-minute averages of NOX concentrations and the nephelometer light-
scattering coefficient. In more recent studies-by Covert etal.^O), growth related to deliquescent
properties of simulated aerosols was demonstrated, and humidity effects observed while monitoring
light scattering in the coastal regions of California were attributed to this behavior. Lundgren^D
conducted a study in which size-classified aerosols were collected on a 4-hour basis to assess
diurnal variations in paniculate composition and concentration. Results of the study indicated
that light scattering correlates well with particulate nitrate, peroxyacetyl nitrate (PAN) and total
paniculate concentration when windy days are omitted from the analysis.
^•Prevailing visibility is defined as the greatest visibility which is attained or surpassed around at least half of the horizon circle.
Visual range is the distance at which it is just possible to visualize with the unaided eye the contrast of a prominent dark object
against the daytime horizon sky.
BATTELLE — COLUMBUS
-------
bscat at
Effective
Wavelength Visual
of Nephelometer, Range,
l0-4 m.| miles
Clean air — ^
Freon 12
calibration — ^
-o a
— 1
— 2
—3
—4
-5
-6
-7
-8
r9
— 10
)
-100
-50
-30
r20
-15
-12
-10
-9
-8
-7
-6
-5
-4
-3
— 1 1
— 12 .
— 13
FIGURE 2. RELATIONSHIP BETWEEN VISUAL RANGE AND LIGHT SCATTERING
MEASURED BY AN INTEGRATING NEPHELOMETER*8'
BATTELLE — COLUMBUS
-------
Results of the 1969 Pasadena study^1^ showed that the most pronounced increases in light
scattering and aerosol volume are coincident with maximum solar radiation intensity. Analyses
of data from the 1972 Air Resources Board project in California indicated that the relationship
of light scattering with ozone is not well established, and autocorrelation results show complex
relationships for light scattering that are not well understood.'^'
Chemical Characteristics of Aerosols
Numerous studies have been conducted in an effort to characterize the chemical composition
of aerosols. Most of the work has been on total suspended particulates, and many analyses have
dealt only with obtaining the distribution of chemical elements in the particulates. A comprehen-
sive elemental analysis of "hi-vol" samples from six CAMP stations (Cincinnati, St. Louis, Washing-
ton, D.C., Denver, Philadelphia and Chicago) was reported by Blosser.^4) John et al.^S) reported
on trace-element concentrations of "hi-vol" paniculate collected in San Francisco. More important
to our understanding of the nature of haze have been studies involving the chemical analysis of
aerosols according to size. Elemental analyses of cascade-impactor collections have been made by
several investigators.'16-18) The size distributions of phosphate, nitrate, chloride, and ammonium^)
and that of sulfate'20) have also been studied using cascade impactors. Some of the aforementioned
studies, and particularly studies conducted by Friedlander and his associates^21'2**), have been
directed at identifying sources of aerosols on the basis of emissions data and elemental tracers of
certain origin. The overall results of most of this work show that in polluted atmospheres most of
the naturally occurring aerosols are in the >2 jum-size range together with emitted and/or reen-
trained matter such as fly ash, tire dust, and cement dust.
There appears to be general agreement that the aerosols <2 jtim are composed predominantly
of sulfur and organic compounds, with nitrogen compounds and some metals associated with com-
bustion (Pb, Zn, and V) making minor contributions.
An interesting study by Novakov et alJ24) described diurnal variations in the chemical states of
sulfur and nitrogen paniculate in Los Angeles. The S^+ (sulfate) to S4+ (sulfite) ratio for particles
<2 nm was shown to be > 1 for only a few early-morning hours. Throughout most of the day the
sulfate/sulfrte ratio was near 0.5.
Far less information is available on the nature and origin of the organic matter of small aerosols,
owing primarily to difficulties of organic micro analyses and the fact that the organic constituency
does not contain compounds traceable to specific sources. While few detailed analyses have been
performed it has become routine to determine the fraction of total particulate which is extractable
by organic solvents, most often benzene. The benzene fraction is reported to range from 4-14
weight percent over a range of U.S. cities^25' The aliphatic fraction of benzene-soluble matter
has been studied by gas chromatographyj2^) in a recent study by Ciaccio et al.^7)( a variety of
extracting solvents and spectroscopic methods were employed to analyze the organic fraction of
aerosols collected along a highway complex in New York City. Their analyses, however, were
performed on aerosols in the size range 0.4-7 jim, and they may be only marginally relevant to
light-scattering aerosols. One of the more pertinent studies in organic analysis was conducted
some 20 years ago by Mader et al.'28' Analyzing organic fractions of Los Angeles aerosols in
the light-scattering range (0.3-0.8 jum), they observed sizeable amounts of oxygenated and
peroxidic organic materials, and they demonstrated similarities between the absorption spectra
of the Los Angeles aerosols and those generated synthetically upon irradiating gasoline hydro-
carbons and NOX in a smog chamber.
BATTELLE — COLUMBUS
-------
SCOPE OF PROGRAM
First Year
The initial objective of this program was to develop a method of distinguishing between light-
scattering aerosols attributable to automotive sources as opposed to those from nonautomotive
sources. Both chemical and optical approaches were considered initially, but upon reviewing the
possibilities in more detail, emphasis, was placed on chemical methods of analysis. Thus the first
year of the program was devoted to seeking a chemical basis which would permit differentiation
among the organic composition of aerosols according to precursors. For this purpose aerosols
associated with the following sources were collected for analysis:
(1) Rural atmospheres, Blue Ridge Mountains, N.C.
(2) Controlled atmospheres, Battelle-Columbus smog chamber
(3) Diluted primary auto exhaust
(4) Urban atmospheres, Bronx, New York.
Chemical fractionation of particulate extracts was performed after which organic constituents
were analyzed by gas chromatography combined with mass spectrometry. It became apparent
that spectra of organic material present from each of the above sources is enormously complex.
This is true even when a single hydrocarbon and NOX are irradiated to produce aerosols.
Irradiation of a-pinene (one of several terpenes emitted in abundance by trees and plants)
and nitrogen oxides in a smog chamber produced aerosols having chemical properties similar to
that of aerosols collected in a forested region. In this case, pinonic acid was identified as a
product of both the natural-rural (Blue Ridge Mountains) and the simulated-rural (smog chamber)
aerosols. This finding served to confirm the utility of the smog chamber in simulating atmo-
spheric conditions conducive to secondary aerosol formation.
Analyses of primary auto-exhaust aerosols revealed two predominant compounds, namely
benzoic and phenylacetic acid, among the complex acidic fraction of the aerosol matter. However,
these aromatic acids were not detected in urban particulate similarly analyzed. This result is
rationalized in terms of dilution of auto exhaust in the atmosphere, or on the basis of removal of
these acids from the atmosphere by photochemical reactions. Indeed, decarboxylation of organic
acids was found to occur during the progress of aerosol formation under smog-chamber conditions.
The principal mission during the first year was to determine if a chemical basis could be
established for estimating the atmospheric burden of secondary aerosols related to automotive
emissions. Combined gas chromatography — mass spectrometry analyses revealed that the
organic constituency of atmospheric aerosols'was enormously complex — so complex that com-
plete resolution was a formidable task. It was also obvious that automobile exhausts do not
contain hydrocarbons uniquely related to automotive operations*. Thus at the conclusion of
the first year, the approach of establishing unique precursor relationships between automotive
emissions and the chemical composition of aerosols was abandoned, in spite of the evidence
that aerosol precursor relationships could be established in smog-chamber simulations of simple
systems. The problem as it related to auto exhaust was too complex, and too little was known
about the chemical nature of aerosols in the atmosphere.
'There are a few hydrocarbons, such as acetylene and ethylene, which might be regarded as unique to automotive emissions on a
relative basis, but these hydrocarbons do not participate in the formation of organic matter condensing into aerosols.
BATTELLE — COLUMBUS
-------
Second Year
At the beginning of the second year, a somewhat broader objective was defined which sought
to improve on the understanding of the composition and sources of aerosols contributing to urban
haze. The overall objective was to determine the relationships which might exist between the con-
centration and composition of the aerosols responsible for reduced visibility and the sources and
conditions under which these aerosols form in the atmosphere. In seeking these relationships, it was
apparent that the customary methods of characterizing paniculate composition were incomplete,
particularly with respect to the organic constituency. Secondly, it was obvious that a complete
analysis of the air was necessary concomitant with the aerosol collections. Thus the second year's
program had two specific objectives (1) to develop and apply an analytical scheme for characterizing
the broad chemical classes and functional groups present in the organic fraction of light-scattering
aerosols and (2) to conduct field-sampling programs for collecting aerosols and characterizing the
prevailing atmosphere conditions under which the aerosols accumulated.
Field sampling was conducted in downtown Columbus during the latter half of July, 1972, in
New York City (Welfare Island) throughout August, and in Pomona, California, for 10 days in mid-
November. The sampling involved continuous monitoring of the meteorological conditions, inte-
grated light scattering (visibility), solar radiation intensity and the gas-phase composition of the
air (including CO, NO, N02, S02, ozone, methane, ethylene, acetylene, and total hydrocarbon)
while simultaneously collecting aerosols for chemical analyses. All aerosol samples were collected
on a diurnal basis. To provide the aerosol mass needed for organic analyses, three samplers
operated continuously at 20 cfm, sampling 25 feet above ground level. Because the program is
concerned with small particles which cause light scattering, the samplers were equipped with size-
fractionating devices which provided particle separation (based on aerodynamic size) near 2 urn
diameter.
An analytical scheme was developed which provides semi-quantitative data on the organic
features of aerosols. Diurnal differentiation of samples limited the quantities of material available
for analyses. The organic matter extracted from the daily aerosol collections «2 urn diameters)
averaged approximately 14 mg per extraction. Within such limitations, both spectroscopic and wet-
chemical analyses were conducted. The analyses provide numerical data on the weight-percent
solvent extractable (methylene chloride and dioxane) components of the aerosols, weight-percent
CHN in the extractables, infrared spectroscopic band intensities for specified absorptions, aromatic/
aliphatic ratios and the concentration of total alcohol and total carbonyl in specified sample frac-
tions. During the second year's program, six aerosol samples were analyzed; two samples from each
of the three sampling sites. In addition, inorganic analyses were performed on aerosol samples
selected from 22 of the 43 sampling days at the three sites. The analyses included trace metals
by optical emission spectrometry, Cl, Br, S, and Pb by X-ray fluorescence, and, SO^, NOo, and
NH^, and CHN by other chemical methods.
Third Year
To fulfill the overall objective set forth in the second year, two specific objectives were
defined for the third year of the program: (1) to apply the organic analytical scheme to addi-
tional aerosol samples and to estimate the precision of the analytical procedure, and (2) to
conduct a statistical analysis of the aerometric and aerosol compositional data with emphasis on
the development of an empirical model of light scattering based on air pollution data.
SATTE1.LE — COLUMBUS
-------
8
Organic analysis were conducted on three atmospheric samples collected by the groups
noted in parentheses: Denver, Col. 11-17-73 (EPA), West Covina, Ca. 9-21-73 (Battelle-Columbus),
Rubidoux, Ca. 9-21-73 (EPA/California Air Resources Board). A week-long collection of aerosol
in Pomona, Ca. (Oct., 1972) by EPA was composited to obtain enough aerosol mass for replicate
analyses in estimating analytical accuracy. In addition, organic analyses were conducted on an
aerosol sample representative of primary and primary plus secondary auto-exhaust aerosol. The
latter sample was a composite of auto exhaust aerosol collected after irradiating diluted exhaust
in Battelle's smog chamber.
Utilizing data from the 1972 field study in New York, an empirical model was developed
relating light scattering to other measures of air quality. A computer program for multivariate
classification was used to select the independent variables to be considered for model develop-
ment. From the preliminary information, time-series models were created which could predict
the degree of light scattering on the basis of changes in CO, N02, total hydrocarbons and rela-
tive humidity. Applications of the model to short-term forecasting of visibility in other localities
(Denver, Col. and Dayton, Ohio) were successful. The modeling results were summarized in a
paper entitled "A Model of Urban Visibility Based on Air Quality" presented at the 168th
National Meeting of the American Chemical Society, Atlantic City, New Jersey, 1974.(29)
The procedures and results of the statistical analysis of ambient data, and those of the
chemical analyses of aerosols, comprise the major portion of this report.
IATTELLE — COLUMBUS
-------
9
MAJOR FINDINGS AND CONCLUSIONS
Light-Scattering Interpretations
During the combined air-sampling period of more than 3 months in New York and Ohio, the
maximum hourly-average light-scattering coefficient was 12 x 10"^m"', corresponding to a visual
range of about 2 miles (Figure 2). It is noted that inferences here to visual range relate to single
point measurements. During the same sampling period, the lowest hourly-average scattering
coefficient was 0.9 x 10"^m"1 (visibility of 35 miles), and the grand average value was 3.4 x
10"^m"^ (visibility of 9 miles). In contrast, light-scattering measurements performed in the Blue
Ridge Mountains during the season of blue haze indicated a maximum value of 2 x 10~^m~^
(visibility of 15 miles) and a minimum value approaching 0.2 x 10"^m"1 {(visibility >100 miles);
a minimum value corresponding to the coefficient of molecular {Rayleigh) scattering of clean air.
Unlike frequent conditions of maximum visibility in the Blue Ridge Mountains, the air in the
urban areas was always significantly polluted with light-scattering particles.
The average diurnal pattern of light scattering in New York and the two Ohio cities (Columbus
and Dayton) showed similar features. Typically, a steady increase in light scattering started in the
early evening and continued until sunup. This trend is most likely attributable to surface inversions
that begin at sunset.(30} The overall light-scattering patterns also revealed slight fluctuations coin-
cident with increased automobile traffic during the morning and late afternoon rush hours. With
few exceptions, noonday and early-afternoon maxima in light scattering were absent, thus leading
to the conclusion that photochemical reactions are not the dominant processes involved in the
generation of light-scattering aerosols in these regions of the country.
The hour-by-hour changes in light scattering in these cities can be described satisfactorily by
an empirically derived time-series model which predicts light scattering on the basis of the daily
trends of the scattering coefficient and other air-quality factors, namely the concentrations of
nitrogen dioxide, carbon monoxide, total hydrocarbon and relative humidity. We have shown that
the relatively simple time-series models developed here are capable of predicting the next hourly
light-scattering average within 5-10 percent of the actual b^t average at the 50-percent-confidence
level. The model also proved useful in forecasting light scattering several hours in advance of exist-
ing air-quality data. For example, for 4-hour lead times, the model predicted light scattering within
10 and 17 percent of the actual values in Dayton and New York, respectively, at the 50-percent-
confidence level. Forecasts of light scattering in hourly increments up to 8 hours were attempted,
and the results are described in the text of the report.
It is important to note here that conclusions about causality cannot be drawn from the struc-
ture of the time-series models developed or from the variables comprising the models. The variables
were chosen because, statistically, they form a set of predictors which could best explain the
variance in light scattering. A variable such as CO, for example, is undoubtedly a "tracer" variable,
that is not one which is directly involved in the process of haze formation, but one which in some'
way adequately summarizes information about other variables which are involved in the process.
The results of the modeling are very encouraging for the prospects of developing a somewhat
more sophisticated and expanded model which would provide statistical predictions of afternoon
and evening visibility based on trends in the atmospheric conditions of the morning and the pre-
vious day. Eventually, it might also be possible to couple the stochastic model to a kinetic chemical
model to adequately explain the process of haze formation in smog.
BATTELLE — COLUMBUS
-------
10
As made evident throughout the report, there is a dramatic difference between the chemical
composition of the light-scattering aerosols (i.e., diameters <2 /urn) and the larger aerosols (diam-
eters > 2 Mm). The finding supports the idea that the sources of aerosols in the two size ranges
are decidedly different and that there is little interaction (e.g., agglomeration) between the aero-
sols in these size ranges. Coupling the rates of change observed in light scattering with the con-
clusions regarding particle interaction leads to other implications. Among them is the generaliza-
tion that light-scattering aerosols in the regions sampled are continually and rapidly being diluted
and removed by the influx of cleaner air. In general, coagulation of these aerosols which occurs
with aging and which reduces the net scattering efficiency, and scavenging mechanisms which
ultimately remove the aerosols from the air are both slow processes (half-lives of several days
are typical) relative to the rate of dilution required to balance the apparent emission and/or
formation rates of the light-scattering aerosols in these cities. (It was not uncommon to observe
light scattering increasing at hourly-average rates of 1-2 bscat units per hour, corresponding to
aerosol-mass concentration increasing at rates of perhaps 40-80 ng/m^ per hour.) Thus it appears
that the Eastern cities are blessed by a consistent breeze which is particularly prevalent during the
peak sunlight hours when solar energy is crucial to the development of photochemical aerosols.
The overall result of the local cleansing is, of course, the transport of light-scattering aerosols to
other regions of the country. And the net effect in the outlying regions is likely that of creating
an artificial background of particles whose average concentration will be dominated by dispersion
rates and scavenging rates. Superimposed on the "background" aerosol concentration will be the
diurnal variations in concentration controlled by local emissions and local meteorological conditions.
Features of Aerosol Composition
Chemical analyses of aerosol samples collected in Columbus, New York, and Pomona show
that there are vast differences in the chemistry of aerosols classified on the basis of aerodynamic
diameters above and below 2 jum. The larger particles (>2 Mm) which play little part in the
scattering of light are composed primarily of soil and sea compounds. Metallic compounds, silica,
and carbonates predominate. Nitrate and sulfate compounds comprise a minor fraction of this
matter.
In contrast, aerosols in the size range <2 Mm contain almost no metallic elements - lead
and zinc are exceptions. They consist primarily of organic compounds and sulfur compounds.
Sulfate accounted for about 20 percent of the small-particle mass, but sulfate sulfur accounted for
only one-third of the total sulfur concentration. The concentration of sulfate was highly correlated
with the ammonium concentration. Nitrate was a minor constituent of the aerosol in Columbus
and New York but a substantial constituent in Pomona. Statistical analyses of the inorganic com-
position of the light-scattering aerosols in conjunction with the quantity of haze and other air-
quality parameters revealed few statistically-significant relationships. Aside from the strong corre-
lation between sulfate and ammonium, inorganic components are unrelated to each other. Consider-
ing air-quality factors, there was a significant positive correlation between the degree of light
scattering and the percent concentration of sulfate and ammonium. A very high correlation exists
between the percent lead in the aerosols and the daily average NC"2 concentration. For the most
part, daily variations in the inorganic composition of aerosols are not large, and subtle relationships
which might exist with atmospheric variations remain hidden by the complexities of the atmo-
spheric conditions.
For organic analyses, samples of light-scattering aerosols «2 Mm) were subjected to soxhlet
extraction using methylene chloride (MeC^), and some of the samples were subsequently subjected
to a second extraction using dioxane. For 10 atmospheric aerosol samples, the weight percent
MeCl2 extractable averaged 17 percent, and ranged from 8 to 45 percent. For 7 aerosol samples
subjected to a second extraction with dioxane, the weight percent dioxane extractable averaged
BATTELLE — COLUMBUS
-------
11
19 percent and ranged from 5 to 40 percent. In cases where both solvents were sequentially
employed, total extractable matter averaged 33 percent, and ranged from 16 to 44 percent.
It was determined that solvent-extractable blanks for glass-fiber and quartz-tissue filters are
significant compared to the mass of extractable material obtained from a daily collection, and
preextraction of such filters is recommended in sampling for similar analytical purposes. The
magnitude and variability of the dioxane blank was sufficiently large that data concerning weight-
percent dioxane extractable matter should be interpreted cautiously. Owing to the uncertainties
concerning the dioxane extracts most analyses were conducted using MeCl2-extracted matter.
With the exception of the sample from the Trout farm area of Denver, which showed several
peculiarities in composition, the weight-percent-extractable and the gross-elemental (i.e., the C, H,
N, 0 distribution) compositions of the methylene chloride extracts of samples from New York,
Ohio, and California were quite similar. These extracts were fractionated into water-soluble and
water-insoluble components. The water-insoluble material was then fractionated into acid, neutral
and basic component. The distribution of material among these classes served as part of the general
characterization of the sample. In general, the MeCl2 extracts of the West Coast samples contained
a higher percentage of water-soluble matter than the East Coast samples (40 percent versus 15 per-
cent). Among the water-insoluble fractions of these extracts, the neutral fractions account for 60-
70 percent of the mass and the acid fractions account for 30-40 percent. Quantitative analyses for
total car bony I and total alcohol among the neutral fractions reveal the presence of about 5 weight
percent oxygen for each of the functionalities. No consistent trends were apparent in the carbonyl
and alcohol data, either with respect to their relative distributions or with respect to the total
oxygen content of the aerosols or the concentration of ozone during the sampling periods.
Infrared analyses of the MeCl2 extracts show the presence of a variety of carbonyl-containing
compounds and possibly the presence of peroxide and/or carbonate compounds. There are also
indications of small concentrations of organic nitrates, and the trends in the intensity of the
nitrate bands correlate well with the total nitrogen content of the extracted aerosols. Fourier-
transform-IMMR spectrometry indicates that aliphatic protons are an order-of-magnitude more
prevalent than aromatic protons.
The Denver (Trout farm) sample which was nearly one-half soluble in MeCl2 contained rela-
tively little oxygen and nitrogen, and appeared to be highly "saturated" compared to all the
samples from other regions.
Excluding the unusual Denver sample, the overall characteristics of the organic composition of
light-scattering aerosols are only moderately different for samples from different sites and are only
moderately different on days of different air-quality at the same site. There appears to be a trend
of higher percentages of total oxygenated matter on days of higher light scattering, but the more
specific analyses for oxygen-bearing functionalities show no obvious patterns with the degree of
light scattering or other air-quality parameters. Comparing the organic chemistry of the urban
aerosols with the chemistry of the aerosols collected from smog-chamber irradiations of auto ex-
hausts yielded no chemical basis by which to estimate the contribution of automobile exhaust to
the total paniculate burden. In general, the chemistry of the exhaust-related aerosols showed some
significant similarities with the West Coast samples, marginal similarities with the New York samples,
and no similarities with the Denver sample. In the end, it must be stressed that only a very limited
number of aerosol samples have been analyzed for organic constituency. It is noteworthy however
that good agreement between replicate analyses of a Pomona sample suggests that differences
revealed by the organic analytical sequence are genuine. While it is felt that the samples and the
analyses thereof are representative of the chemistry of urban atmospheric particulate, conclusion
cannot be drawn about specific sources and precursors of the condensed organic matter.
BATTELLE — COLUMBUS
-------
12
RESULTS AND DISCUSSION
Light-Scattering Measurements
As discussed in the Background Section of this report, the quantity of haze was determined
by an integrating nephelometer (with a heated inlet) which operates on the principle of light
scattering. Throughout the course of this program, nephelometer measurements were made in
New York City (Bronx, September, 1970; Welfare Island, August, 1972), Columbus (July, 1972),
and Dayton (August, 1974), and the Blue Ridge Mountains (October, 1970). Data collected
continuously at the Welfare Island and Dayton sites are of sufficient duration (nearly 1 month)
to justify making generalizations about the average or typical patterns of haze in these localities.
The averages of the hourly averages of light scattering are plotted in Figure 3 for each hour
of the day in Dayton and New York. The structure of the two diurnal light-scattering curves
are quite similar. In both cities there is an increase in light scattering beginning near 1700 hours
and continuing to a maximum in the early morning. In New York, and to a lesser extent in
Dayton, there is a secondary maxima occurring during the morning rush hours. Surprisingly,
there is a lack of a maxima during the noon or afternoon hours attributable to increased photo-
chemical aerosol formation so commonly observed on the West Coast. Even on days where
substantial photochemical smog was evident, and there were many (03 exceeded the 1-hour max-
mum standard of 0.08 ppm on 15 of 21 days in New York and 12 of 20 days in Dayton),
afternoon increases in light scattering were only slightly pronounced. In the Blue Ridge Moun-
tains where primary aerosol sources were negligible and light scattering sometimes approached
background conditions (i.e., molecular scatter), an afternoon maximum in light scattering often
occurred due to photochemical reactions involving terpenes and nitrogen oxides.
We conclude from these data that throughout an average diurnal period, light-scattering
aerosols in midwestern and eastern cities originate primarily from primary emissions and/or
dark (thermal) reactions of gaseous pollutants and secondarily from photochemical reactions.
However, during the daytime when haze is visible, the two sources of aerosol are more nearly
equal, and our conclusion regarding the principal source of these aerosols is made with caution,
surmising that if it were not for the good ventilation of these cities in the afternoon, the photo-
chemical aerosol contribution would be of greater significance.
Statistical Analysis of Light Scattering
The objective of this portion of the program was to develop an empirical model which would
mathematically account for haze (light scattering) in terms of various air quality and meteorolog-
ical data. It was desirable that the model be derived from hourly or shorter-time averages of the
data; i.e., the model was designed to describe hour-by-hour changes in light scattering rather than
long-term trends. A model developed in such a manner could perform a two-fold function:
(1) It could lead to the development of the capability for short-term forecasting
of urban visibility and aerosol concentrations (much like a weather forecast).
(2) Since the basic time frame is short enough to reflect the impact of thermal
and photochemical reactions, an empirical model of this type could yield
information relevant to the understanding of these processes. Eventually,
BATTELLE — COLUMBUS
-------
5.0
ID
t>
H
H
m
p
p
m
n
o
r
c
3
0
c
Ul
4.0
'E
•*
'O
X
^ 3.0
o
o
c
1_
O)
o
c/> 2.0
en
1.0
0
Dayton, Ohio
July, 1974
New York City (Welfare Island)
August, 1972
1
8 10 12 14
Hour of Day
16
18
20 22
24
FIGURE 3. AVERAGE DIURNAL PATTERN OF LIGHT SCATTERING IN NEW YORK CITY AND DAYTON, OHIO
-------
14
perhaps a blend of stochastic and theoretical chemical models could be devel-
oped to adequately explain the process of haze formation.
Preliminary Analyses
On the premise that the concentration of light-scattering aerosols at any given time might de-
pend on the immediate history of the atmosphere, as well as some instantaneous factors, the ap-
proach selected for modeling light scattering was based on time-series analysis. Since the computer
programs available for developing time-series models are limited to the use of three predictor
variables, it was necessary to reduce the number of possible predictor variables to be included in
our analysis. For this purpose, a multivariate empirical technique known as AID (short for Auto-
matic Interaction Detector) was employed.(31) By means of an iterative binary splitting pro-
cedure, this method shows those variables and interactions among variables which are likely to
best explain the variance of a criterion variable (in this case, light scattering). A more complete
discussion of this procedure and a detailed account of the results is presented in Appendix A.
In the diagram below, the results of applying AID to the New York data are summarized.
Of the 15 independent variables, NO2, CO, relative humidity (RH), and total hydrocarbons
(THC) were chosen as the variables to include in the time-series analysis. Since four variables
were selected, it was decided to develop four time-series models for light scattering, one for each
of the four possible combinations of three variables taken from this set of predictors.
Original Independent Variables
Global Irradiance
Wind Speed
Wind Direction
Temperature
Relative Humidity
Sulfur Dioxide
Ozone
Nitric Oxide
Nitrogen Dioxide
Nitrogen Oxides
Carbon Monoxide
Methane
Total Hydrocarbons
Ethylene
Acetylene
Selected Predictor Variables
Nitrogen Dioxide
Carbon Monoxide
Relative Humidity
Total Hydrocarbons
It should be noted that time-lagged variables were not explicitly included in the AID analysis.
However, since the variables were hourly averaged, there is some reflection of short-term time lag
relationships in the data. Since time lags of even longer duration may be inherent in the photo-
chemical processes of haze formation, it is possible that other important variables were overlooked
by this screening technique.
BATTELLE — COLUMBUS
-------
15
Time-Series Analysis
Nine days of hourly averaged data (the four predictor variables NC>2, CO, RH, and THC, and
light scattering) from August 22 to August 30, 1972, in New York were programmed for time-
series analysis. Some preliminary analyses were performed on data as originally collected, i.e., at
10-minute intervals. However, for reasons discussed below, the remainder of the analyses were
performed on hourly averages of 10-minute data.
The method used to develop the time-series models is that of Box and JenkinsJ32) Tne
method, while theoretically based, places emphasis on pragmatism and parsimony, and hence is a
practical method for time-series analysis of real data. Further discussion of the Box-Jenkins
methods and detailed results are also included in Appendix A. A general discussion of the basic
procedures follows.
Before developing the relationship between light scattering and the various predictor vari-
ables, it is first necessary to obtain univariate models for each of the variables to be included in
the equation. These models relate the observed series to an uncorrelated noise series. The noise
series may be interpreted as the effect of other variables on the time structure of the observed
series. This noise series is then due to effects unexplainable by the given observed series, and it
is the noise series that embodies the stochastic character of the observed series, i.e., the part due
to chance. After the univariate models are developed, the multivariate relationship is studied.
The multivariate relationship is incorporated in a time-series model called a transfer-function
model. It is a dynamic model, and incorporates structural relationships between the dependent
variable and predictor variables, as well as a noise series relating to those effects still unexplained
by the relationship. The univariate models and transfer-function models are capable of fitting
seasonal effects, trends, and even characteristics of time series which can change with time, e.g.,
stochastic trends. Diurnal variations were detected in the observed series for light scattering and
the chosen predictor variables, and the subsequent model development incorporated seasonal
(diurnal) factors to account for this variation.
Initially, AID analyses and time-series analyses were performed on the 10-minute data. How-
ever, difficulties were encountered in developing transfer-function models. Careful examination
revealed that part of the problem was due to the fact that changes due to "noise" were on the
same order as real changes and thoroughly intermixed, hence "masking" statistical relationships
in the data. In an attempt to reduce this problem, hourly averages of data were used. This
would tend to "smooth" variations due to noise and thus put in sharper focus basic changes oc-
curring in the data. The AID analyses for the 10-minute data and the hourly averages were sim-
ilar, but now meaningful relationships emerged in the transfer-function models. At this point,
it was decided to carry out the time-series analyses solely on hourly averages.
A total of eight transfer-function models were developed. The four models mentioned pre-
viously were developed for each of two cases: (1) with seasonal differencing of the data, and
(2) with no seasonal differencing of the data. There were two reasons for developing the second
set of four models. First, it was noted that when seasonal differencing was applied (and this
was indicated by preliminary Box-Jenkins analysis) the resulting models had a large degree of
"overfitting". That is, much of the effect of seasonal differencing was nullified by parameters
occurring in the consequent models. Hence, in interests of simplicity, it was decided to develop
the nonseasonally differenced models. A second consideration was that the Denver data, ob-
tained from EPA for purposes of attempting forecasts, covered only a short period of time, and
diurnal differencing was not appropriate. It should be noted here that seasonal parameters re-
mained incorporated in the nonseasonally differenced models.
BATTELLE — COLUMBUS
-------
16
The best of the eight models developed from the New York data, in terms of reducing the
variance of the noise factor (i.e., the unexplained effects), was the diurnally differenced model in-
volving the predictor variables N02, CO, and RH. The diurnally differenced models as a group
were superior to the nondiurnally differenced models.
Table 1 presents a comparison of the different models according to the variance of the
noise component. For a base comparison, the univariate light-scattering model and its variance
are also included. Note that the N02, CO, and RH transfer-function model reduces the variance
of the noise component by about 50 percent compared to the univariate light-scattering model.
It is important to consider that no conclusions about causality can be drawn from the struc-
ture of the time-series models developed, or from the variables included in the various models.
These variables were chosen because, statistically, they form a set of predictors which could best
reduce the unexplained effects. A variable such as CO, for example, is undoubtedly a "tracer"
variable, that is, not one which is directly involved in the process of haze formation, but one
which in some way summarizes information from other variables which are involved in this pro-
cess. In the present analysis, therefore, the models developed are best suited for predictive pur-
poses, and the tie-in between the empirical method of time-series analysis and theoretical cause-
effect models is left to future research.
Forecasting
The transfer-function models developed form the basis for the forecasting procedure. Fore-
casts from a given point in time make use of past history of the light-scattering and predictor-
variable series. The past history used for each variable depends on the structure of the particular
forecast model. In general, a time period of up to several hours preceding the forecast is used,
and if diurnal factors are included, appropriate information from preceding days are used. The
forecast is comprised of two basic steps: first, relevant observed past values are put into the fore-
cast model; second, the structure of the time-series model transforms this information into a
forecast value for a given lead time.
Forecasts were of interest for two reasons: (1) various statistical tests performed on the
forecasts could be used to attempt to validate models developed; (2) results of the forecasts
could be studied to see how effective short-term forecasts for various lead times would be. An-
other consideration in the forecasting analysis was whether the models developed in New York
City could be applied elsewhere. Data were obtained from Denver, Colorado, and Dayton, Ohio,
for this purpose.
Figure 4 displays the results of the 1-hour-in-advance (lead-1) forecasts for light scattering
over a 49-hour period from August 10 to August 14, 1972, in New York City. This period was
not used in developing the forecast model. • The transfer-function model in this case was the
seasonally differenced model with predictor variables N02, CO, and RH.
The X-axis represents time in hours, the Y-axis light scattering in conventional bscat units
(10~4 nrr1). The solid line records the observed light scattering for this time period. The asterisks
represent forecasts made 1 hour in advance, and the dotted lines designate the 50 percent confi-
dence band about the forecasts, i.e., approximately 50 percent of the time the observed values
should fall within the dotted lines. The figure was constructed in such a way as to provide im-
mediate comparison between forecast values and the "eventual" observed values, and to yield
visual evidence that the appropriate percentage of observed values fall within the confidence bands.
BATTELLE — COLUMBUS
-------
17
TABLE 1. COMPARISON OF UNIVARIATE AND TRANSFER FUNCTION
MODELS OF LIGHT SCATTERING BY RESIDUAL VARIANCE
Residual Variance in
Model Type Light Scattering (x lO^
A. Models Without Seasonal Differencing
Univariate model 0.300
Transfer function models with
predictor variables
I. NO2, CO, RH 0.155
II. IM02, CO, THC 0.169
III. N02, RH.THC 0.172
IV. CO. RH.THC 0.165
B. Models With Seasonal Differencing
Univariate model 0.282
Transfer function models with
predictor variables
I. N02, CO, RH 0.146
II. NO2, CO, THC 0.155
III. N02, RH.THC 0.182
IV. CO, RH, THC 0.157
BATTELLE — COLUMBUS
-------
7 . 0
fr . 0
5 . 0
t>
H
H
m
p
p
m
n
0
p
c
C
a
1 . 0
3 . 0
2 . 0
1 . 0
0 . 0
0.0
00
5 . C
0.0 15.0 20.0 ? 5 . 0 30.0
T imE
I HOUR FORECAST
35.0
10.0
15.0
50.0
55.0
LINE=ACTU*L B-SCM
• 51 EM SR -1 -HOUR FORECAST
DASHED LlfcE=.»0 CONFIDENCE
*YCI I -LS VS CO , RM,ND2
FIGURE 4.
-------
19
The width of the confidence band about the forecast in Figure 4 is ±0.33 b^^ units, and 24 out
of 49 observed values fall within the confidence limits. Another statistical test shows that the
1-hour-lead forecasts are uncorrelated, a property they should possess. Hence, statistically and
practically, the 1-hour lead forecasts are satisfactory.
Figures 5 through 7 show, respectively, the results of the 2-hour, 5-hour, and 8-hour lead
forecasts using this model for predictions in New York. Table 2 summarizes the results of these
forecasts (as well as those for other locations), including widths of confidence bands and per-
centage of observations falling within the 50 percent confidence bands. While perhaps still being
acceptable, the longer lead forecasts may indicate some degradation in model effectiveness.
Figure 8 is a composite of the lead-1 through lead-8 forecasts. The numbers 1 through 8 on the
graph are located at the values of the 1-hour through 8-hour-in-advance forecasts, respectively.
This figure shows the tendency of the forecasts to converge to the observed values as the lead
time becomes shorter. The convergence illustrates the adaptive character of the Box-Jenkins
forecasting model.
Figures 9 through 13 show the same information for forecasts made on data collected in
Denver, Colorado, on November 15 and 16, 1973. The figures show about the same degree of
success with the Denver data as with the New York data. However, because of the short span
of consecutive data, and because the N02 data were incomplete, the forecast model used for
Denver was nonseasonally differenced, and the predictor variables were CO, RH, and THC.
Finally, Figures 14 through 18 summarize the results of forecasts made on data collected in
Dayton, Ohio, in July and August, 1974, where the seasonally differenced model was used with
predictors N02, CO, and RH. Results are extremely good, perhaps indicating that a model de-
veloped from Dayton data would have a smaller noise variance than that developed from New
York data.
Summary
The results of forecasting light scattering in three localities for lead times up to 8 hours
show statistically acceptable results for the shorter lead times, with "reasonable" results for
longer lead times. Further, the adaptive nature of the forecasting procedure continually im-
proves forecast results as further information becomes available (and lead times become corre-
spondingly smaller). The models used to generate the forecasts must be considered extremely
simple and yet results are surprisingly effective. A possible explanation of why the forecasting
method developed in New York seems to apply equally well in Dayton and Denver is that per-
haps not enough sophistication has yet been built into the model, and that differences between
individual cities are still embedded in the noise process. Further resolution into other predictor
variables may make differences more pronounced.
One element of surprise in determining good predictor variables was that wind speed did
not seem to be of much help in predicting light scattering. It was decided to investigate the
data for a possible reason. Looking at wind speed in conjunction with wind direction showed
more promise. Hence a method was devised to incorporate a wind-velocity vector into the AID
analysis. Results, graphically displayed in Appendix A, show that wind velocity is an important
predictor of light scattering. Unfortunately, it was not possible at this time to incorporate a
vector-valued variable in the time-series analysis.
BATTELLE — COLUMBUS
-------
7 . 0
H
H
m
r
r
n
o
r
C
I
t. o
5 . 0
3 . 0
2 . 0
I . 0
0 . 0
0. 0
5 . C
0.0 15.0 20.0 25.0 30.0
T I HE
2 HOUR FORECAST
35.0
10.0
15.0
50.0
55.0
LINE-ACTUAL 6-SC AT
ASTER ISK=2-HOUR FORECAST
DASHED LIME-.50 CONFIDENCE
NVCII -LS vs CO,RH,NOJ
B »ND
FIGURE 5.
-------
7 . 0
n
o
r
C
6 . 0
5 . 0
1 . 0
3 . 0
2 . 0
1 . 0
0 . 0
0. 0
5.0 10.0 15.0 20.0 25.0 30.0
T IKE
5 HOUR FORECAST
FIGURE 6.
35.0
10.0
15.0
50.0
55.0
LINE=»CTU»L B-SC»T
BSTFR I S* =5-HOUR FORECAST
DASHED LIME-.iO CONFIDENCE
NYCI I -LS VS CO,RM,N02
-------
7 . 0
r
r
I
0
0
r
C
t. o
. 0
t . 0
3 . 0
2 . 0
1 . 0
0 . C
0.0 S.C IC.C 15.0 20.0 25.0 30.0
t I PE
8 HOUI> FORECAST
FIGURE 7.
35.0
t C . 0
15.0
50.0
L t ME MCTUAl B - S C « 1
»STER I SK =8-HOU»i F"K
DASHED ll*E-.5C CCSFICE^CE
NVCII -LS V5 CP .«•• . fcP Z
CANC
-------
23
TABLE 2. FORECAST SUMMARY
Forecast Hours in Advance
.50 Confidence Band
No. of Observations in
Forecast Band/Total
1
2
3
4
5
6
7
8
NYC II
± .33
± .51
± .60
± .68
± .75
± ,81
± .87
± .93
24/49
22/48
16/45
12/42
1
2
3
4
5
6
7
8
Denver
± .33
± .52
± .62
± .71
± .785
± .855
± .92
± .98
21/38
20/37
14/34
9/31
1
2
3
4
5
6
7
8
Dayton
+ .33
± .51
± .60
± .68
± .75
± .81
t .87
± .93
30/49
28/48
29/45
24/42
ATTELLE — COLUMBUS
-------
7 . 0
0
>
H
H
m
p
r
m
n
o
r
C
S
o
c
0
6 . 0
5 . 0
•« . 0
3 . 0
2 . 0
I . 0
7 I
I 2 3
0 . 0
0. 0
S.O 10. C 15.0 20.0 25.0 30.0
T I f\f
I TO 8 HOUR fORECASIS
35.0
1 0 . 0
15.0
50.0
5 5 C
I I WE =»CIU«l 6-SCat
•lYCII -LS VS C 0 . « H , fc P 2
ISJ
FIGURE 8.
-------
H
m
r
p
m
n
o
r
C
3
D
C
a
s . o
? . o
0 0
0 0
5 . 0
I 0 0
?0.0 ?S 0
T I OF
1 HOUR c n H p r a '. T
FIGURE 9.
TOO
3Sv/0 tO . 0
A S TF S I <;« r I -HPUH FPPFCA^T
L I » F = . S 0 rnNflOFH
-LS V'. r".Dn.THt
-------
J>
H
H
ffl
P
P
m
n
o
p
c
c
a
7 0
(- 0
•> . 0
3 . 0
? . 0
a o
o o
10.0
15.0
?0 . 0 ?5 0
T I Hf
Z HOUR FORFT•S T
300
o. / •"to.o ts.o
V
L I HE =»C TU«L 8 -
-------
J>
-\
H
m
r
P
m
I
n
0
r
C
I
0
C
0
s o
i . o
o o
o o
•>
? 0 0 7 S 0
T I P1F
S Hfl(|B
FIGURE 11.
ISO1. ,' \ 1 J» 0 / 1 '
l/l N F ?« r T j a i n -*,r»'
na'.MFn IINF= •><)
OFNVFB-1"; V^ rn
NJ
-------
0
*
H
r
p
m
0
0
p
c
3
•
c
a
0 0
0 0
I 5 0
?0 0 ?S 0
T I nF
B H n || R FPRPrAr. T
FIGURE 12.
15.0 ', • "t 0 0' X •
I I NF =»C~t U«l B -^r
• STFR I •.« =8-HOll«
no ".H F n i i N( = s o
DFNvFM-t<; »<; ro.
-------
7 ft
ft
1 M
0
J>
H
H
m
r
r
m
n
o
r
C
c
0)
1 . (i
1 ft
h 7 «
' 0
' M ;: I s s
. c
I 0 0
ISO
?0.0
I rn
10 0
ISO
10 0
- a r T '; A L ft - s r a T
FR-i*i v ^ r o . R H , T M r
FIGURE 13.
-------
in
r
r
m
0
0
r
C
2
a
c
a
\ i. o
10.0
9 . 0
8 . 0
7 . 0
6 . 0
5 . 0
. 0
3 . 0
I . 0
0.0
5 . 0
\15.0 20.0 25.0 30.0
T I PIE
1 HOUR FORECAST
FIGURE 14.
35.0
100
15.0
SO . 0
55.0
I I ME -ACTUAL § -SC »T
ASTER I Sn = I -HOUR FORECAST
DASHED LINE-.SO CONFIDENCE B»SC
OAVTOK-LS VS CO.RM.TMC
-------
H
H
m
r
r
n
o
c
0
11.0
i o .0
9 . 0
8 . 0
7 . 0
t . 0
5 . 0
M . 0
3 . 0
\V A •
2 . 0
0. 0
5 . 0
1 C
15.0 20.0
25.0 30.0
T THE
2 HOUR FORECAST
FIGURE 15.
35.0
HO.0 S5 . 0
$0.0
55.0
CO
B-SCAt
ft 51 ER1 SH = 2-HOUR FORECAST
DASHED LIKE-.50 CONFIDENCE BAND
OAYTON-LS WS CO , BH,TMC
-------
11-0,
H
-4
n
r
r
n
o
r
C
3
ID
C
n
10.0
9 . 0
8 . 0
I 7-0
5
C
A
T
6 . 0
5 . 0
1 . 0
3 . 0
2 . 0
0. 0
5 . 0
10 15.0 20.0 250 30.0
T I «E
•> HOUR FORECAST
FIGURE 16.
35.0 "(0.0
tS . 0
50 . 0
55.0
L I NF = «C1 UAl B-SCAT
»STER I SK-S-MOLR FORECAST
DASHED LINE^.SO CONFIDENCE BAND
OAVTON-lS VS CO,«H,TMC
-------
H
H
R
I T
n
o
r
C
i
11.0
10.0
•» . 0
8 . 0
7.0-
6 . 0
5 . 0
1 . 0
3 . 0
2 . 0
0.0
5 . C
0.0 15.0 20.0 25.0 30.0
T IHE
8 HOUR FORECAST
FIGURE 17.
35.0
10.0
15.0
50.0
55.0
CO
CO
LlNE=»CTl'»l 8-SCAT
»S1ERIS*=8-HOUR FORECAST
DASHED LIME-.50 CONMOENff BAND
DAYTON-IS VS CO.RM.TMC
-------
r
p
n
o
C
01
11.0
10.0
9 . 0
8 . 0
. 0
6 . 0
5 . 0
1 . 0
3. 0
2 . 0
0.
5 . 0
I .
15.0 20.0 25.0 30.0
T I HE
I TO 6 HOUR FORECASTS
35 . 0
10.0
•(5.0
90.0
55.0
LI ME ^ACTUAL t-SCAT
OAVTON-lS VS CO,AH,h02
FIGURE 18.
-------
35
Thus, although this model must be regarded as preliminary and incomplete, the results are
very encouraging. It appears that with some additional expansion and refinement of the tech-
nique, it would be possible to make accurate predictions of the concentration of light-scattering
aerosols in afternoons or evenings based on information gathered earlier in the day and from
preceding days. One obvious improvement is to make provisions for including more predictor
variables. Haze formation is a complicated process, and more than three variables are necessary
to explain its evolution. A noticeable deficiency in the present models is the lack of more vari-
ables with a potential to describe the effect of time lags inherent in photochemical and perhaps
thermal processes governing aerosol growth. Inclusion of such variables would be expected to
enhance the performance of models in forecasting over longer lead times. Secondly, interactions
among predictor variables may have profound effects on light scattering which do not show up
when the variables are treated separately. A good example of this is the interaction between
wind speed and wind direction. Taken individually, even though in succession, they do little to
explain an effect on light scattering. Treated together, as a velocity vector, a powerful influence
on light scattering is detected. Methods of identifying such interactions and incorporating them
into the time-series analysis must be devised.
Chemical Composition of Light-Scattering Aerosols
Inorganic Composition
During the second year of this program, selected aerosol samples from Columbus, Ohio,
New York City, and Pomona, California, were analyzed for their organic and inorganic composi-
tions. Analyses were performed on diurnal collections of aerosols having diameters <2 jum.
Analyses were also performed on composite samples of aerosols having diameters >2 /urn. De-
tailed analytical results were presented in the Second Year Report.^)
The average inorganic composition of the aerosols collected at the three sites is summarized
in Table 3. As indicated in Table 3, there are vast differences in the composition of aerosols in
the two size classifications implying that the sources of these aerosols are different and that there
is little interaction (e.g., agglomeration) among aerosols in these size ranges; at least on a diurnal
time scale. The chemical results tend to support the contention based on mobility measurements
of aerosol size that aerosol mass in urban atmospheres is distributed bimodally, with a saddle point
in the mass distribution near 2 pm diameter. ^2) Indeed, it was on the presumed validity of that
distribution data, as well as considerations of light-scattering properties, that particle separation
was performed for diameters <2
According to the summary in Table 3, nearly all of the metallic compounds are associated
with larger particles; of all the metals, only lead and zinc were found to have substantial fractions
in the smaller aerosol size range. As a group, metallic elements constitute the major fractional
part of the mass of larger aerosols, but <2 percent of the mass of the smaller aerosols. Soil-
derived elements (Fe, Al, Si, and to a lesser extent Na, K, Ca and Mg) are found primarily among
the larger aerosols. Elements of sea salt (Na, Cl, and to a lesser extent Mg) also predominate
in the larger size range. Inorganic carbon compounds (carbonates) constitute the other major part
of larger aerosols. Carbon in this case is assumed to be inorganic on the basis of the small
amounts of hydrogen associated with the aerosols. The assumption is consistent with the re-
sults of Mueller, et al., who measured carbonate carbon in size fractionated aerosols from Pasa-
dena. (33) Tney found that carbonate carbon was concentrated in size ranges >2 /urn, and it
constituted about 10 percent (12 /ug/m3) of the total particulate mass. If we assume that all of
the carbon in the larger aerosols is inorganic, and all the carbon in the smaller size range is
organic, then the inorganic carbon concentration in the New York area averages 9
BATTELLE — COLUMBUS
-------
36
TABLE 3. AVERAGE AEROSOL CONSTITUENTS AT THREE LOCALITIES lal
Aerosol diameter (/urn):
Avg mass concentration
Columbus
New York
Pomona
<2
54
>2
24
<2
58
>2
91
<2
43
Weight Percent According to Aerosol Size
>2
56
Constituent
Metals
Carbon
Hydrogen
Halogen
Sulfate
Nitrate
Ammonium
Total
1
17
6
0
19
1
6
50
36
17
1
1
2
2
0
59
2
26
6
1
22
4
7
68
37
16
1
6
4
3
0
69
2 NC'b)
25
5
2
14
24
9
81
(a) The composition of aerosols <2 jim diameter are averages of samples from 5 days in Columbus, 11 days in New York,
and 6 days in Pomona. Data for aerosols >2 tun are averages from 11 days in Columbus and 27 days in New York;
no large aerosols were collected in Pomona.
(b) NC: large aerosols not collected for analysis.
BATTELLE — COLUMBUS
-------
37
Results of the same Pasadena study indicated that 90 percent of the organic carbon occurred
in particles < 1 pm in diameter, and that organic carbon comprised 37 to 54 percent of the par-
ticulate mass in that size range. By pyrolitic analyses, we found somewhat lower percentages of
organic carbon in the smaller aerosols collected from Columbus, New York, and Pomona, but the
very limited Pomona samplings are not likely representative of the Los Angeles area. As dis-
cussed in the next section of this report, organic extractions by methylene chloride and
dioxane indicated 20-30 weight percent organics in the light-scattering aerosols around Columbus
and New York.
Nitrate was found in both aerosol size ranges in contrast to a report by Lee and Patterson'^)
where nitrate predominated in the submicron size range. Judging from recent evidence that
gaseous nitric acid can be captured as particulate on ordinary glass fiber filters'34), it is suspected
that discrepancies regarding the distribution and/or concentration of nitrate in aerosols may be
related to nitric acid collection. High purity quartz filters (to our knowledge used here for the
first time in atmospheric sampling), do not seem to absorb nitric acid, and the nitrate concentra-
tions we report may therefore be somewhat less than those reported previously.
Ammonium was found to be exclusively associated with the smaller size fraction of aerosols
which is consistent with most reports on the ammonium distribution. The sum of nitrogen from
nitrate and ammonium accounts for about 75 percent of the total-aerosol nitrogen. The question
is often raised whether ammonium salts are formed on filter substrates from reactions of gaseous
ammonia with acids. This is an important consideration, and our results do not provide answers
to that type question. Although we are reasonably certain that nitric acid did not "fix" ammonia
on the quartz substrate, it is possible that sulfuric acid did.
Along with organic matter, sulfate compounds constituted a major fraction of the mass of
the smaller aerosols. The exclusiveness of sulfate among small particles does not imply a dominant
source or mechanism of sulfate formation as numerous processes for both heterogeneous and
homogeneous conversion of S02 to sulfate have been postulated which result in submicron aero-
sols. Sulfate sulfur accounted for only about one-third of the total-aerosol sulfur, but no deter-
minations were made for sulfite. Novakov et al. reported that sulfite sulfur predominated over
sulfate sulfur in the <2 /urn aerosol-size range.(24) The average concentration of sulfate in light-
scattering aerosols in New York of 12 ng/m^ is near the overall 5-year (1964-1968) average of
11.4 jig/m3 of total sulfate for eastern urban sites.'35)
The halogen contribution to the mass of small aerosols was <2 percent. In New York, chlo-
ride was the predominant halogen - and its concentration is attributed primarily to sea salt.
Regression analyses were performed to determine statistical relationships which might exist
between the inorganic composition of aerosols and the air quality and meteorological data associ-
ated with the respective periods of aerosol collections. Eleven days of sampling in New York
was used for the analysis, and the air quality data were averaged on a 24-hour basis for compari-
son with the diurnal aerosol samples. The results are summarized in Table 4.
Before discussing the regressions relating aerosol composition to air quality data it is impor-
tant to note that, in several instances, the regression variable selected as most significant on the
basis of the highest correlation coefficient was only slightly better than other highly correlated
independent variables. Correlations among the independent variables are indicated in Table 5;
coefficients >0.75 appearing in bold face. Correlations among the dependent variables, i.e., the
compositional components of the aerosols, were quite poor with the exception of the correlation
between sulfate and ammonium concentrations which was R = 0.94.
BATTELLE — COLUMBUS
-------
38
TABLE 4. REGRESSION ANALYSES ON INORGANIC AEROSOL COMPOSITION
AND AVERAGE AIR QUALITY CONDITIONS
Type of regression: Linear
Variable dimension: Mass Concentration
Dependent Variable Variable'3* R F
Pb
Br
Cl
S
S04
N03
NH4
C
NOX
NO
C2H2
bscat
bscat
C2H2
bscat
WS
0.87
0.79
0.59
0.58
0.95
0.57
0.98
-0.79
28.9
14.8
4.9
4.7
75.1
5.0
76.8
15.0
Linear
Percent Concentration
Variable R F
NOX
03
temp(b)
wp
bscat
temp
bscat
THC
0.90
-0.75
-0.51
0.42
0.78
-0.49
0.83
-.50
32.9
11.4
3.2
1.9
13.9
2.9
20.2
3.0
Logarithmic
Mass Concentration
Variable R F
NOX
°3
C2H2
bscat
bscat
C2H2
bscat
WS
0.88
-.77
.59
0.56
0.92
-0.66
0.91
-0.81
32.3
13.2
4.9
4.2
46.4
6.9
42.3
17.5
(a) In several instances, the "best" independent variable had a correlation coefficient only slightly higher than
that of another independent variable. Correlations between the independent variables are listed in Table 5.
(b) Underlined variables do not meet the 95 percent confidence test.
BATTEULE — C O I. U M B U S
-------
TABLE 5. CORRELATION COEFFICIENTS AMONG THE INDEPENDENT VARIABLES SUBSEQUENTLY
REGRESSED AGAINST AEROSOL COMPOSITION
m
H
H
m
p
p
m
1
n
0
p
c
s
CD
c
01
S02
03
NO
N02
CO
CH4
THC
C2H2
NMHC
BSCAT
WS
TEMP
RH
IRRAD
1.00
-.46
.41
-.73
-.31
-.55
-.55
-.60
-.43
-.26
-.31
.25
-.25
-.41
.37
1.00
-.89
.23
-.39
.08
.27
-.03
-.27
-.27
.22
-.01
.56
.20
-.01
1.00
-.02
.61
.17
-.09
.16
.23
.29
.05
-.16
-.23
.05
-.2
1.00
.74
.89
.74
.85
.62
.41
.79
.35
.49
.73
-.73
1.00
.80
.48
.74
.60
.50
.67
-.60
.25
.56
-.74
1.00
.88
.68
.68
.17
.83
-.60
.38
.75
-.90
1.00
.43
.61
-.15
.65
-.38
.18
.71
-.79
1.00
.48
.80
.63
-.57
.32
.62
-.60
1.00
.99
.39
-.47
-.06
.41
-.59
1.00
.27 1.00
- .36 - .70 1 .00
.17 .64 -.32 1.00
.24 .78 -.33 .49 1.00
-.18 -.82 .75 .59 -.69 1.00
OJ
CO
S02 03
NO
NO2 NOX CO CH4 THC C2H2 NMHC BSCAT
WS
TEMP
RH
IRRAD
-------
40
Referring to Table 4, three types of regressions were derived for each of the dependent vari-
ables appearing in the left-hand column. In the first case (column 2), a linear model was fitted
using mass concentration units for the dependent variables. In the second case (column 3), the
aerosol composition data are expressed as the percent of the total aerosol mass concentration,
and in the final case, concentration units were used again but with an exponential regression
model. In Table 4, R denotes the correlation coefficient and the sign of the relationship, and F
denotes the degree of significance of the correlations; the significance increases with increasing F
values. An F value of about 4 corresponds to a 95 percent confidence interval, and a value of
13.5 corresponds to a 99 percent confidence band.
The results obtained where linear and logarithmic equations were regressed on the same data
are remarkably similar. In all but one case where the correlation coefficients exceed 0.75, the
independent variables having the highest correlations were the same for the linear and logarithmic
models, and the correlation coefficients were also quite similar for the two models. This implies,
of course, that the correlations tend to be linear. In the one case (Br) where the most highly
correlated variable was different for the two models, a substantial correlation exists between the
two highest correlated predictor variables (R = -0.83 for 03 versus NO).
Linear regression results where the compositional data are expressed on a percentage basis
are not much different from those where absolute concentrations were used. Overall, the corre-
lations with percent composition are lower, and in four cases (Cl, total S, N03, and total C) the
regression equations are not significant at the 95 percent level.
On the basis of these analyses, there is definitely a statistically significant correlation between
daily averaged Pb concentrations and the 24-hour average concentration of NOX. One might ex-
pect a priori that the concentration of most any of these aerosol substances would correlate best
with light scattering (bscat) since aerosol mass loading (<2 /im) increases with increasing light
scattering (R = 0.94 for b^-^ versus mass <2 jum). This result occurs in the regressions on S,
804, and NH4- The fact that Pb shows a better correlation with NOX under these circumstances,
further supports an apparently strong correlation between NOX concentration and the amount of
Pb in aerosols.
As noted, there is substantial correlation between 804 and IMH4 concentrations (total S to
a much lesser extent) and average light scattering. Significant positive correlations between
bscat and tne percent of 804 and NH4 indicate that the percentage of (1X^4)2804* in the par-
ticulate tends to increase on days of highest light scattering.
Correlations of air quality data with the concentration of particulate Cl, total S, and N03
are quite low and therefore of little meaning. The particulate carbon concentration correlates
fairly well with negative wind speed; again, an expected first-order correlation which offers little
information on the possible relationships with sources.
To search further for statistical relationships on aerosol composition, stepwise multiple re-
gression equations were derived. Statistically acceptable regression equations containing two or
more independent variables were possible for Pb, total S, 804, N03, and NH4. The results are
summarized below. The correlation value (2R) there is the cumulative coefficient (required to
increase with each succeeding step). The F values represent the significance of adding the final
term; the significance of the regression equation is always much larger. A negative sign indicates
negative correlation with the specified variable.
•The near 2/1 stoichiometric ratio of M-<4 to 804. and the high correlation between their concentrations lead us to assume
that nearly all of the NH^ and SO4 form the 2S04 compound, in agreement with other reported work where the
structure was confirmed by X-ray analyses.
BATTELLE — COLUMBUS
-------
Dependent
Variable
Pb
S
S04
N03
NH4
Second Step
Variable
-WS
-RH
-NMHC
-NO
-NO
ZR
0.93
0.78
0.97
0.76
0.98
F
7.25
5.8
6.3
4.2
24.0
41
Third Step
Variable ER
NO2 0.96 5.0
-RH 0.99 11.3
Including wind speed and IM02 in the correlation on Pb improved the overall regression co-
efficient from an initial value of 0.87 to 0.96. The negative correlations appearing in the second
regression steps on S, 864, N03, and NH4 are obviously of no positive benefit in delineating any
aerometric effects on aerosol composition. The negative correlation between NH4 concentrations
and RH tends to challenge the theory that (NH4J2S04 formation occurs predominantly in water
droplets at high humidities. The concentration of carbon, for which wind speed correlated best
in the first regression step, showed no second-order correlation that was significant.
Thus, with the exception of particulate Pb which correlated consistently with nitrogen oxide
concentrations, no outstanding statistical evidence was uncovered which would allow us to draw
conclusions regarding variations of the inorganic composition of aerosol. For the most part,
short-term (i.e., daily) changes in inorganic composition are not great, and the rather subtle rela-
tionships which may exist remain hidden by the complexities of the atmospheric conditions.
Organic Composition
Beginning the second year of this program, methods in organic analyses were developed and/or
adapted to provide information concerning the broad chemical classes and functional groups pres-
ent in the organic fraction of light-scattering aerosols. Two major factors determined the nature
of the analyses undertaken:
(1) The limited quantities of sample available for analysis
(2) The desirability of obtaining numerical (i.e., semiquantitative) data.
Because it is ultimately desirable to relate the organic composition of aerosols to diurnal trends
in atmospheric conditions, samples of aerosols available for analyses were limited to an average
mass of about 14 mg, i.e., the average mass of organic matter extracted from the <2 /jm aerosol
fractions collected from three sampling systems operated continuously at a flow of 20 cfm.
In the selection and development of procedures to be included in the organic analytical
scheme, particular attention was given to the time and cost of each operation. In view of an
ultimate goal of obtaining data on a statistically significant number of samples, the cost per
sample coutd not be excessive. Nevertheless, while not prohibitive in cost, the procedures do
require a moderate degree of analytical sophistication (refer Appendix B). During the second
and third years of the program, the analytical scheme was applied to ten samples of atmospheric
particulate and two samples of particulate corresponding specifically to automotive sources. The
samples, listed below, have been numbered to permit convenient reference in the discussion to
follow.
BATTELLE — COLUMBUS
-------
42
Sample
Reference
No. Site/Source Date
1 Columbus, Ohio July 21, 1972
2 Columbus, Ohio July 26, 1972
3 New York City August 23, 1972
4 New York City August 11, 1972
5 Pomona, California November 18, 1972
6 Pomona, California November 10-13, 1972
7 Pomona, California October 14-18, 1972
7A Replicate A
7B Replicate B
8 Denver, Colorado November 16-17, 1973
9 Rubidoux, California September 21, 1973
10 West Covina, California September 21, 1973
11 Primary Automotive Exhaust
12 Irridiated Automotive Exhaust
Where available, atmospheric data corresponding to the sample collection periods are sum-
marized in Table 6. Sample Nos. 1 and 2 from Columbus, 3 and 4 from New York, and 5 and 6
from Pomona were collected on days of substantially different air quality, based primarily on
the range of light-scattering values encountered during the sampling periods at each site. Columbus
sample No. 1 was collected during one of the most severe smog episodes in the Ohio area. New
York sample No. 3 was selected as a typically hazy day in New York; however, there were no
days of "unsatisfactory" air quality during the sampling period. Haze was evident during the col-
lection of Pomona sample No. 5, b,ut was moderate relative to most smog episodes in the Basin
area. Pomona sample No. 6 is a composite of four very similar days during which very little haze
was evident. Pomona sample No. 7 is a composite of aerosol collected by EPA which was pooled
for the purpose of obtaining an organic extraction large enough for replicate analyses performed
to assess the reproducibility of the analytical procedure. The Denver sample was of interest be-
cause of its rather unusual and uncertain origin in the outlying regions of Denver. The Rubidoux
and West Covina samples represent simultaneous collections at two points (about 50 miles apart)
along a line extending due west from Los Angeles. Primary auto exhaust particulate was collected
from a dilution tunnel in which exhausts were diluted during chassis dynamometer operations.
Sample No. 12 is a pooled sample of secondary and primary auto exhaust particulate obtained by
filtering the contents of a 17 m^ smog chamber after irradiating diluted auto exhausts (8 ppm C)
for 6 hours. The samples were collected as part of API Project EF-8 at Battelle-Columbus. De-
tails regarding generation of the automotive samples are available.(36)
The data presented here serve to illustrate the types of numerical data that can be developed
for the organic composition of particulate.. In Appendix B the procedures employed are described
in sufficient detail that other chemists may find them applicable in their work. Although the
number of samples undertaken for organic analysis during this program was not sufficient to per-
mit statistical analyses, it is hoped that further utilization of the described procedures will yield
data that can be more fully analyzed and interpreted. Indeed, it is expected that apart from
interpretations presented in this report, the data will be useful to others investigating the chem-
istry of atmospheric aerosols.
BATTELLE — COLUMBUS
-------
TABLE 6. SUMMARY OF AIR QUALITY CONDITIONS DURING ORGANIC PARTICULATE COLLECTIONS
J>
H
m
r
r
m
1
n
0
r
C
3
ID
C
m
Sample
No.
1
2
3
4
5
6
7
8
9
10
Site
Columbus
Columbus
New York
New York
Pomona'3'
Pomona'3'
Pomona
Denver
West Covina'c>
Date
July 21. 1972
July 26, 1972
August 23, 1972
August 11, 1972
November 18, 1972
November 10-13, 1972
October 4-18, 1972
November 16-17, 1973
September 21. 1973
September 21, 1973
Weather Conditions
Aerosol Mass Loading, fig/m3
Quartz. Millipore
General^1 Temp, C RH. % <2 fim A-Total B<2 pm B/A
S 30 64
S 22 -
PC 25 74
PC 21 55
PC — —
S — —
(Composite sample furnished by
— 4 40
— — —
PC 18 76
69.5 85.9 72.9 0.85
34.6 60 36.5 0.61
53.4 179.0 71.9 0.40
34.3 81.1 23.8 0.29
52.4 114.2 55.0 0.48
17.5 42.7 26.2 0.30
EPA for replication purposes only.)
_ _ _ _
_ _ _ —
— 123 — —
Pollutants
Light 24-Hr Avrj 1-Hr Max
Scattering, THC,
lO"4 m'' ppm
5.8 34
1.6 2.9
3.5 50
1.4 3.5
2.4(a) 3.2
1.4
-------
44
Solvent-Extractable Particulate Matter. As shown in Table 7, samples 1-6 were subjected to
Soxhlet extraction, first with methylene chloride (20 hours) and then with dioxane (44 hours).
Sequential Soxhlet extraction using methylene chloride followed by dioxane was used to obtain a
wider range of organics than are extractable using the more commonly applied single-solvent pro-
cedures. Dioxane is expected to dissolve the polar, relatively oxidized organics normally left behind
by benzene or chlorinated solvents. The data shown in Table 7 indicate that the dioxane extract-
ables do indeed represent a significant fraction of the total extractable organic matter.
As noted in Appendix B, an experiment was conducted to verify that dioxane does not extract
significant quantities of inorganic salt from the filters. Several salts were stirred individually with
dioxane, and the filtered solvent was lyophilized in tared flasks. The data indicate that such in-
organic salts do not correspond to >4 percent of the dioxane extractable matter, and that most
likely this value is <1 percent of the dioxane-extractable matter.
Throughout the course of the determinations, solvent and filter blanks were carried through
the reflux and concentration procedure. In calculating the values shown in Table 7, blanks due to
solvent and filter background have been subtracted. Typically, three 182-cm2 quartz-tissue filters
were extracted using 100 ml of solvent. Blanks were determined as 0.09 mg (±0.01 mg) per filter
for methylene chloride extractions, and 0.8 mg (±0.3 mg) per filter for dioxane extractions. The
magnitude and variability of the dioxane blank is such that data concerning the dioxane-extractable
matter should be interpreted cautiously. Nevertheless, it is felt that these values are not entirely
spurious, and that they demonstrate that a significant fraction of atmospheric paniculate matter
consists of relatively oxidized organic matter. This assertion is supported by data for weight per-
cent C, H, and N determined for methylene chloride and dioxane-extractable matter.
In view of the difficulties associated with the blanks where dioxane was used as an extraction
solvent, atmospheric samples 7-10 and automotive samples 11 and 12 were extracted only with
methylene chloride. Additionally, in an effort to further reduce blank values, all field sampling
after 1972 was conducted using preextracted quartz-tissue filters. In these cases, preextraction was
conducted first with methanol, then with benzene. Preextraction with methylene chloride led to
a high chloride background. The preextracted material was shown to have a methylene chloride
blank of 0.002 mg per filter and a methanol blank of 0.2 mg per filter. Thus, the reduction of
background organic matter by preextracting was significant. Samples of primary and irradiated
automotive particulate were collected on Gelman Type A glass-fiber filters (not preextracted).
The blank for the Gelman filters was determined as 0.020 mg methylene chloride extractable
matter per 81-cm2 filter.
In considering the data in Table 7, note first the good agreement between replicate samples
7A and 7B. It is emphasized that this sample was divided after the extraction step. The methy-
lene chloride extract was halved gravimetrically, after which concentration of the extract, deter-
mination of aliquot-weights (details, Appendix B), and all subsequent analytical procedures were
conducted independently on the replicates. It was felt that cutting the filters in half initially
to permit duplicate extractions would introduce an error of itself. Moreover, such division of
filter samples is not a usual step in the scheme.
For 2 of 3 sampling sites (compare Samples 3-6, Table 7), it appears that a substantially
larger fraction of dioxane extractable matter was present on days of more intense haze. Because
highly oxidized materials are extracted by dioxane, a greater fraction of such compounds may be
present on hazy days. The methylene chloride extractable fraction was fairly similar for all of
BATTELLE — COLUMBUS
-------
TABLE 7. SOLVENT EXTRACTION OF PARTICIPATE MATTER
H
H
ffl
r
r
m
n
o
r
C
2
0
C
U)
Sample
No
1
2
3
4
Site
Columbus
New York
5 Pomona
6
7 Pomona
7A Replicate A
7B Replicate B
8 Denver
9 Rubidoux
10 West Covina
11 Primary Automotive
Methylene Chloride Extraction
Date
July 21. 1972
July 26, 1972
Aug. 11, 1972
Aug. 23. 1972
Nov. 10-13, 1972
"Nov. 18, 1972
Oct. 4-18. 1972
Nov. 16-17, 1973
Sept. 21, 1973
Sept. 21, 1973
Mass
Loading,
Atg/m3
69.5
34.6
34.3
53.4
17.5
52.4
—
—
—
—
—
Total Paniculate
Taken for
Extraction, mg
128.6
63.1
55.6
89.0
82.0
60.0
465.1
—
26.56
167.5
47.64
No. of
Filter Disks
Extracted
2.5
2.5
2.25
2.25
1.5
1.5
4
,
—
2
1
1
Weight
Extractable Matter,
Corrected for
Blank, mg(a|
10.58
10.15
9.17
11.50
17.65
9.52
29.60
29.50
12.03
15.40
7.29
Weight
Percent
Extractable
8
16
16
13
22
16
13
13
45
9
15
Dioxane Extraction
Weight of
Extractable Matter. Weight
Corrected for Percent
Blank, mg Extractable
8
16
16
13
22
16
13
13
45
9
15
1U.91
17.45
5.38
20.07
13.24
24.03
11.5
12.0
—
—
—
8
28
10
23
16
40
5
5
—
—
—
(a) Sample Nos. 1-7: Circular quartz-tissue filters (1.82 x 10~2m2|, not preextracted, corrected for methylene chloride blank of 3.2 x 10'3 mg/m2.
Sample Nos. 8-10: Rectangular quartz-tissue filters (5.16 x 10"2m2), preextracted, corrected for methylene chloride blank of .067 x 10'3 mg/m2.
Sample Nos. 11-12: Circular Gelman-A glass fiber filters (8.1 x 10'3 m?), not preextracted, corrected for methylene chloride blank of 1.6 x 10'3 mg/m?.
Oi
12
Exhaust
Irradiated Automotive
Exhaust
— 27.3
— 19.7
5
26
3.78
2.26
14 — —
11 — ~
-------
46
the samples, with the outstanding exception of Denver, and showed no consistent trends with gross
aerometric conditions.
Weight Percent C, H, and N. Values of weight percent C, H, N, and [0] for the methylene
chloride and dioxane extractables are shown in Table 8. Determinations of weight percent C, H,
and N were conducted using a Perkin-Elmer Model 240 elemental analyzer with gas purification
accessory. The instrument performs automated Pregel-Dumas determinations. Values shown for
weight percent oxygen [0] were calculated by difference from the CHN data. Approximately
1 mg of extractable matter was used for each determination. As expected, weight percent [0]
is significantly higher for the dioxane extractables than for the methylene extractables. Within
these fractions, however, there is little variation on extreme days, and not much variation from
site to site, again the Denver sample may be an exception. Primary exhaust aerosol showed con-
siderably less [0] and N in the methylene chloride fraction compared to the atmospheric sam-
ples. There was insufficient material in the sample of irradiated automotive aerosol to permit
CHN analysis. As indicated in succeeding sections, however, infrared spectroscopic analysis and
functional group analysis indicate that the irradiated automotive aerosol does indeed have a
higher proportion of oxygen and nitrogen containing organics.
Infrared Spectra. Relative compositions of the methylene chloride and dioxane extractables
were considered previously. Continuing along these lines, infrared spectra are shown in Figure 19
for methylene chloride and dioxane extractables of particulate collected in Pomona on November
18. Note that the absorbance due to C-H stretching, near 2900 cm'1, is relatively strong for the
methylene chloride extractables and relatively weak for the dioxane extractables. Absorbances
near 1100 cm"1 may be assigned to C-0 stretching for a variety of compounds including ethers,
lactones, and esters. This region shows significantly stronger absorption with the dioxane ex-
tractables. Both samples show strong absorptions for carbonyl compounds in the region above
1700 cm'1. These spectra indicate that the dioxane extractables consist of the relatively oxidized
aerosol constituents compared with the less oxidized methylene chloride extractables.
Examination of the spectra compiled in Appendix C,indicates that the spectra of methylene
chloride extractable matter generally exhibit significantly more detail and fine structure. In view
of this and the difficulties associated with use of dioxane as an extraction solvent, this discussion
will be confined to consideration of spectra for the methylene chloride extractable matter.
In order to present spectroscopic results in numerical form, key bands were selected and
relative spectroscopic intensities were determined. Relative intensities were calculated by taking
the ratio of optical density for the specified band to that observed for the CH stretching vibra-
tion at 2920 cm'1; i.e.,
Relative Intensity = <0-D- at specified absorption)
(O.D. at 2920cm-1)
Such relative intensities have been calculated for the carbonyl band, the "percarbonyl" band
(discussed below) and bands tentatively assigned as corresponding to organic nitrate.
BATTELLE — COLUMBUS
-------
TABLE 8. WEIGHT PERCENT CARBON, HYDROGEN. NITROGEN, AND
OXYGEN IN SOLVENT EXTRACTABLE MATTER
o
>
H
H
m
p
r
m
1
n
Q
p
c
§
D
C
10
Sample
No.
1
2
3
4
5
6
7
7A
78
8
9
10
11
Methylene Chloride
Extractables,
weight percent
Site
Columbus
New York
Pomona
Pomona
Replicate A
Replicate B
Denver
Rubidoux
West Covina
Primary Auto-
motive Exhaust
Date
July 21, 1972
July 26, 1972
Aug. 11, 1972
Aug. 23, 1972
Nov. 10-13, 1972
Nov. 18, 1972
Oct. 4-18, 1972
Nov. 16-17, 1973
Sept. 21. 1973
Sept. 21, 1973
C
69.3
73.0
68.8
69.1
69.3
61.6
63.2
62.9
76.5
60.6
65.7
82.4
H
9.5
10.2
9.0
9.1
8.9
7.7
7.7
8.2
10.5
7.9
8.6
11.4
N
0.8
0.9
1.3
1.1
1.3
1.9
1.4
1.4
0.8
2.2
1.5
0.5
[0]
20.4
15.9
20.9
20.7
20.5
28.8
27.7
27.5
12.2
29.3
24.2
5.7
Dioxane
Extractables,
weight percent
C H N [O]
44.4 6.1 0.8 48.7
47.2 6.0 0.8 46.0
46.6 5.5 2.2 45.7
54.4 4.8 1.6 39.2
50.6 5.9 1.6 41.9
39.1 5.0 1.9 54.0
_ — — —
- - - -
- - - -
- - -
_ _ _ _
_ _ _ _
-------
03
>
H
H
m
r
p
m
n
o
r
C
2
CD
C
10
2.5
WAVELENGTH (MICRONS)
6 78
30 40
100
METHYLENE CHLORID
EXTRACTABLCS
Sample Collected Nov. 11
Pomona, California
4000
3500
3000
2500
2000
1800 1600 1400
FREQUENCY (CM1)
1200 1000 800
60U
WAVELENGTH 'MICRONS.
6 78
400 200
30 40
DIOXANE
EXTRACTABLES
Sample Collected Nov. II
Pomona. California
4000
3500
3000
2500
2000
1800 1600 1400
FREQUENCY (CM1)
1200 1000 800 600 400
200
FIGURE 19. INFRARED SPECTRA OF SOLVENT-EXTRACTABLE MATTER
-------
49
Group-frequency assignments in the range of 1700-1740 cm"1 have been made for a variety
of carbonyl-containing compound types. The spectra of the methylene chloride extractables
reveal some apparent trends in position and intensity of the carbonyl peak. A subtle variation in
the position and width of the peak appears to be related to the weight percent oxygen in the
sample. Samples with high weight percent oxygen display a subtle shift or broadening of the
carbonyl peak to slightly higher frequency. This may be seen in the spectra shown in Figure 20
{carbonyl peaks are noted with Marker B). This effect is consistent with a greater population of
diketones, keto acids, and keto aldehydes in the samples having higher oxygen incorporation. As
shown in Table 8, the November 18 sample has 28.8 weight percent oxygen, while the November
10-13 sample has 20.5 weight percent oxygen.
Table 9 includes a listing of relative intensity of the carbonyl band as well as the value for
weight percent oxygen in the methylene chloride extractable matter. Examination of these
data reveals a good correlation (R = 0.90) between oxygen incorporation and carbonyl band
intensity.
A second interesting effect concerns a shoulder near 1770 cm"1 (Marker A in Figure 20).
The shoulder may be assigned to peroxides, organic carbonates, anhydrides or lactones. It is
most probably for alkyl/aryl peroxides and carbonates. A relatively limited variety of anhydrides
or lactones would exhibit this absorption, principally five- or six-membered strained cyclic species.
Thus, this absorption is probably due to peroxides or carbonates. Since this absorption cannot
be unambiguously assigned, we will refer to it as the "percarbonyl" peak. This is meant to refer
to the relatively oxidized character of the candidate compound types, especially the peroxides
and carbonates.
•9 « fl
R-C-O-O-C-R R-0-C-O-R
Peroxide Carbonate
Regarding both the carbonyl and percarbonyl bands, good reproducibiiity is apparent for
samples 7A and 7B. There is a substantial difference in band intensities for the primary and
irradiated automotive exhaust aerosols. As expected, the irradiated sample shows greater con-
tent of carbonyl and percarbonyl compound types, and the band intensities for the irradiated
sample are of comparable magnitude to those observed for the more highly oxidized atmo-
spheric samples.
Interesting trends in peak intensities were also observed for absorptions at 1630 cm"1 and
1275 cm"1 {Figure 20, Markers C and D). According to Colthup, et al.(37) organic nitrates
(R— 0-N02) display strong bands near 1660 to 1625 cm"1 (N02 symmetric stretching). Broad
absorptions observed at 850 cm"1 are also consistent with organic nitrates. The observed bands,
however, might also be attributed to aromatic amines. Nevertheless, the group frequencies for
aromatic amines are .quite broad. For example, the band due to CN stretching could appear in
the region of 1250 to 1380 cm"1. Most primary amines have an NH2 deformation band at
1650 to 1590 cm"1, while secondary aromatic amines have an NH bending absorption near
1510 cm"'. Most aliphatic secondary amines have no significant NH bending band above 1470
cm"1. Thus, although various aromatic amines could give rise to the observed bands at 1275
cm"1 and 1630 cm"1, these absorptions are more generally assignable to organic nitrates. The
observed bands are therefore tentatively assigned to organic nitrates. Values for relative band
intensity are shown in Table 10 along with data for weight percent nitrogen. A trend in the
BATTELLE — COLUMBUS
-------
H
H
m
p
r
m
n
o
r
C
2
0
C
(0
2.S
WAVELENGTH (MICRONS)
6 78
3O 40
METHYLENE CHLORIDE
EXTRACTABLES
Sample Collected Nov. 18
Pomona, California
4000
3500
3000
2500
2000
1800 1600 1400
FREQUENCY (CM1)
WAVELENGTH (MICRONS)
6 78
1200 1000 800 60U 400 200
30 40
100
METHYLENE CHLORIDE
EXTRACTABLES
Sample Collected Nov. 10-13
Pomona. California
4000
3500
3000
2500
2000
1800 1600 1400
FREQUENCY (CM1)
1200 1000 800 600
400
200
FIGURE 20. INFRARED SPECTRA OF SOLVENT-EXTRACTABLE MATTER
-------
TABLE 9. INFRARED SPECTROSCOPIC DATA ON CARBONYL BANDS
Data presented are for methylene chloride extracts.
0
H
H
m
r
r
m
I
n
0
r
C
2
0
C
in
Sample
No.
1
2
3
4
5
6
7
7A
7B
8
9
10
11
12
Site
Columbus
New York
Pomona
Pomona
Replicate A
Replicate B
Denver
Rubidoux
West Covina
Primary Automotive
Exhaust
Irradiated Automotive
Exhaust
Date
July 21, 1972
July 26, 1972
Aug. 23, 1972
Aug. 11, 1972
Nov. 18, 1972
Nov. 10-13, 1972
Oct. 4-18, 1972
Nov. 16-17, 1973
Sept. 21, 1973
Sept. 21, 1973
Weight
percent [0]
20.4
15.9
20.7
20.9
28.8
20.5
27.7
27.5
12.2
29.3
24.2
5.7
—
Carbonyl Band
Relative Intensity
0.58
0.38
0.47
0.50
0.64
0.51
0.56
0.56
0.13
0.96
0.64
0.06
0.77
Percarbonyl Band'3*
Relative Intensity
0.16
0.07
0.14
0.13
0.06
0.03
0.05
0.05
<0.02
0.16
0.12
<0.02
0.16
CJl
(a) Discussed in text.
-------
TABLE 10. INFRARED SPECTROSCOPIC DATA ON NITRATE BANDS
Data presented are for methylene chloride extracts.
0
J>
H
H
m
r
r
m
1
n
0
r
C
2
D
C
to
Sample
No.
1
2
3
4
5
6
7
7A
7B
8
9
10
11
12
Site
Columbus
New York
Pomona
Pomona
Replicate A
Replicate B
Denver
Rubidoux
West Covina
Primary Automotive
Exhaust
Irradiated Automotive
Exhaust
Date
July 21. 1972
July 26, 1972
Aug. 11, 1972
Aug. 23, 1972
Nov. 18, 1972
Nov. 10-13, 1972
Oct. 4-18, 1972
Nov. 16-17, 1973
Sept. 21, 1973
Sept. 21, 1973
Weight
Percent N
0.8
0.9
1.3
1.1
1.9
1.3
1.4
1.4
0.8
2.2
1.5
0.5
Nitrate Band l'a> Nitrate Band ll
(1630 cm-1) (1275 cm-1)
Relative Relative
Intensity Intensity
0.17
0.15
0.25
0.19
0.43
0.29
0.22
0.21
0.04
0.47
0.33
<0.02
0.34
0.19
0.15
0.28
0.22
0.37
0.24
0.25
0.26
0.05
0.47
0.32
<0.02
0.49
Average
Intensity,
Bands 1 and II
0.18
0.15
0.27
0.21
0.40
0.27
0.24
0.24
0.05
0.47
0.33
<0.02
0.42
Ul
N)
(a) Tentative assignments.
-------
53
data is clear, and correlation coefficients for nitrate bands I and II with nitrogen content are 0.87
and 0.89, respectively.
Note in particular the enhanced intensity of nitrate peaks in the irradiated automotive aerosol
compared to the primary automotive aerosol. Nitrate band intensities for the irradiated aerosol
are comparable to those determined for atmospheric samples having the greatest degree of nitrogen
incorporation. Again, good reproducibility was obtained for band intensities in replicate samples
7A and 7B.
Aromatic/Aliphatic Ratio. Nuclear magnetic resonance spectra were obtained using a 60 MHz
Varian Associates spectrometer equipped to perform Fourier-transform spectroscopy. Fourier-
transform spectroscopy permits the rapid generation of spectra, consequently allowing the genera-
tion of a large number of spectra in a relatively short period of time. Computer time-averaging a
series of such spectra leads to enhanced signal-to-noise ratios and permits acquisition of useful
spectra not obtainable by conventional techniques. The current application required the use of
FT-NMR to obtain adequate spectra. Spectra were obtained using ~ 1-mg samples of extractable
matter. Methylene chloride extractables were analyzed in deutero-chloroform solution and dioxane
extractables were analyzed in deutero-dioxane solution. From the spectra obtained, aromatic/
aliphatic ratios were computed. The values were calculated on the basis of the integrated reso-
nances for methyl, methylene and aromatic protons; i.e.,
Aromatic/Aliphatic Ratio = (integrated resonances; aromatic protons)
(integrated resonances; methyl + methylene protons)
The results are shown in Table 11. Results for the dioxane extractable matter have been
omitted. It was observed that upon standing, a fine precipitate formed in the concentrated
,deutero-dioxane solutions leading to irregular results. Data for automotive aerosol samples have
been omitted because of sample limitations. Good reproducibility was obtained for replicate
samples 7A and 7B. Reasons for the order of magnitude difference in aromatic/aliphatic ratios
are not apparent. It is clear, however, that aliphatic protons are dominant in the methylene
chloride extractions.
Acid/Base Neutral Distribution. The methylene chloride extractable matter was fractionated
according to the scheme shown below. The material was fractionated first into water-soluble and
water-insoluble components by partitioning the methylene chloride solution against distilled water.
Water was removed by lyophilization, and the weight of the water-soluble fraction was determined.
Methylene Chloride Extractable Matter
I I
Water-Soluble Fraction Water-Insoluble Fraction
I
Acid
Fraction
Neutral
Fraction
I
Basic
Fraction
BATTELLE — COLUMBUS
-------
54
TABLE 11. FOURIER TRANSFORM-NMR ANALYSIS OF
METHYLENE CHLORIDE EXTRACTABLES
Sample No.
1
2
3
4
5
6
7
7A
7B
8
9
10
Site
Columbus
New York
Pomona
Pomona
Replicate A
Replicate B
Denver
Rubidoux
West Covina
Date
July 21, 1972
July 26, 1972
Aug. 23, 1972
Aug. 11. 1972
Nov. 18, 1972
Nov. 10-13, 1972
Oct. 14-18, 1972
Nov. 16-17, 1973
Sept. 21, 1973
Sept. 21, 1973
Aromatic/
Aliphatic
Ratio
0.10
0.11
0.12
<0.01
0.15
<0.01
0.15
0.14
0.04
<0.01
0.01
BATTELLE — COLUMBUS
-------
55
The water-insoluble material remaining in methylene chloride solution was next fractionated
into acid, basic, and neutral components. The solution was extracted using first 2N aqueous sodium
hydroxide and then 2N hydrochloric acid. Material remaining in methylene chloride solution after
extractions with both aqueous sodium hydroxide and hydrochloric acid is defined as the water-
insoluble neutral fraction.
The sodium hydroxide extract (containing organic-acid salts) was brought to pH ~0.8, and
the free acid was extracted into methylene chloride, using a continuous liquid/liquid extractor.
Similarly the hydrochloric acid extract (containing organic-base salts) was brought to pH ~13,
and the free base was extracted into methylene chloride, again using a continuous liquid/liquid
extractor. The methylene chloride solutions containing the acid, basic, and neutral fractions
were dried by refluxing the solutions over a 3 A molecular sieve. The dried solutions were con-
centrated and small aliquots of the concentrate were evaporated to dryness to determine the resi-
due weights for the various fractions. Using these data, values for the weight percent distribution
were calculated.
Details of the fractionation procedure are presented in Appendix B. The procedure repre-
sents a departure from methods most often employed in such fractionation. To minimize sample
losses to the point where meaningful distribution data could be reported for the small samples
available, special techniques including vapor-phase drying and continuous liquid/liquid extraction
were incorporated in the procedure. As refinement of the scheme progressed, however, it be-
came clear that unacceptable material losses had occurred during fractionation of some of the
earlier samples, notably samples 1, 2, 5, and 6. Thus, data for these samples has not been
reported.
Data shown in Table 12 were calculated on the basis of recovered material. Thus the distri-
bution data totals 100 percent for each sample. Overall recoveries varied from 60 to 100 percent.
Typically, about 5 mg of sample was subjected to fractionation. Wet chemical analysis of small
samples inevitably results in some random handling losses. The recovery observed here is con-
sistent with the quantity of sample used and the nature of the analytical procedure applied. The
reproducibility between replicate samples 7A and 7B is good.
Partitioning of the methylene chloride fraction of the two samples from New York indicates
similar distributions in spite of different atmospheric conditions. Samples of irradiated auto ex-
haust and those from Rubidoux and West Covina (but not Pomona) have relatively high water-
soluble fractions. Again, the Denver sample seems somewhat inconsistent with the other atmo-
spheric samples, particularly within the water-insoluble fraction where higher neutral and lower
acidic fractions are noted. Among the water-insoluble fractions (of methylene chloride extract)
the neutral fraction accounts for 60-70 percent of the mass and the acids about 30-40 percent.
Functional Group Analyses. Quantitative functional group analyses for alcohol and for
carbonyl were performed using the water-insoluble neutral fraction of the methylene chloride ex-
tractable matter. Isolation of this fraction was considered above. Details of the fractionation
and functional group procedures are presented in Appendix B.
The procedure for determination of alcohols is subject to interference from carboxylic acids,
phenols, and amines. Thus, the analysis is conducted only on the neutral fraction. Although the
carbonyl determination is not subject to the interferences that limit the alcohol analysis, the
BATTELLE — COLUMBUS
-------
56
TABLE 12. FRACTIONATION OF METHYLENE CHLORIDE EXTRACT ABLE MATTER
Mass Distribution, weight percent
methylene chloride extract
Sample
No.
3
4
7
7A
7B
8
9
10
12
Water- Insoluble Fraction Water-Soluble Fraction
Site
New York
Pomona
Replicate A
Replicate B
Denver
Rubidoux
West Covina
Irradiated Automo-
Oate
Aug. 1 1
Aug. 23
Oct. 4-18
Nov. 16-17, 1973
Sept. 21, 1973
Sept. 21, 1973
Acid
26
30
24
22
18
24
20
Neutral Base
52 4
50 5
48 <1
51 <1
66 3
33 1
36 2
Total
18
15
28
26
14
42
42
five Exhaust
12
33
54
BATTELLE — COLUMBUS
-------
57
carbonyl determination is applied to the neutral fraction in order that a comparison can be made
between alcohol and carbonyl content in the same sample fraction.
Results of functional group analyses were obtained on molar bases. This is well suited to
the current application, in which quantitative measure of specified organic functionalities is desired.
However, to permit a direct comparison between weight percent carbonyl oxygen, weight percent
alcoholic oxygen, and weight percent oxygen obtained from the CHN determination performed
on the unfractionated extract, the molar values for carbonyl and alcohol were converted to a com-
mon basis of weight percent oxygen. This was calculated as shown in the following example:
..... . , (mg [0] present as carbonyl) inn
Weight percent carbonyl oxygen = -—a .K — — x 100.
(mg sample)
The data are shown in Table 13. In considering the results of the CHN determinations, it
was noted that incorporation of oxygen into organic particulate may be a useful indicator of the
tendency for a given atmosphere to oxidize gas-phase or suspended organics. In terms of funda-
mental atmospheric reactions, incorporation of oxygen to yield carbonyl compounds might be
related to ozonolysis, while incorporation to yield alcohols might be related to free-radical
processes.
With the exceptions of the Columbus No. 1 and the Denver No. 8 samples, which were
higher and lower, respectively, the alcoholic oxygen fraction of the neutral samples were fairly
constant. The irradiated automotive sample showed a much higher content in this category. The
fractional carbonyl content was similar for Columbus and New York samples with somewhat
higher percentages occurring for Denver, Rubidoux, West Covina, and particularly irradiated auto-
mobile exhaust.
Further examination of the data does not reveal meaningful correlation between functional
group concentrations and weight percent oxygen in the methylene chloride extract or functional
group concentrations and average ozone concentration on the respective sampling days. Indeed,
the values for carbonyl and total oxygen concentrations do not correlate with each other.
Summary. Organic fractionation and functionality data are summarized in Table 14 for aero-
sol collected on days of substantial light scattering at four sites and from smog chamber irradiations
of auto exhaust. The data are presented on a weight percent basis and therefore reflect composi-
tional differences and/or similarities among the samples.
Comparing first the amounts of methylene chloride extractable material, it is apparent that
extractable fractions are quite similar in all cases except Denver. The dioxane extractable frac-
tion, not included here, probably accounts for another ~20 percent of the particulate mass.
(Note that all data in Table 14 pertain to the methylene chloride extractable fractions). Review-
ing the CHN[O] data, it appears that the Denver sample is composed of more saturated matter
(higher C-H values) than the other samples and that the West Coast and irradiated exhaust sam-
ples consist of higher percentages of [O] and N than the Denver and New York samples. Parti-
tioning the methylene chloride fraction resulted in identical water-soluble and insoluble fractions
of the West Covina and Rubidoux samples, which are somewhat similar to the analyses on the
irradiated exhaust samples and dissimilar to those on the Denver and New York samples. The
basic fraction of the water-insoluble material was small for all samples and, although some
B
-------
TABLE 13. FUNCTIONAL GROUP ANALYSIS OF METHYLENE CHLORIDE NEUTRAL FRACTION
D
H
H
m
r
r
m
1
n
0
r
C
2
n
c
10
Sample
No.
1
2
3
4
5
6
7
7A
7B
8
9
10
12
Site
Columbus
New York
Pomona.
Pomona
Replicate A
Replicate B
Denver
Rubidoux
West Covina
Irradiated Automo-
Date
July 21, 1972
July 26, 1972
Aug. 23, 1972
Aug. 11, 1972
Nov. 18, 1972
Nov. 10-13, 1972
Oct. 4-18, 1972
Nov. 16-17, 1973
Sept. 21, 1973
Sept. 21, 1973
Weight Percent
[0]
MeCI2
Extract
. 20.4
15.9
20.7
20.9
28.8
20.5
27.7
27.5
12.2
29.3
24.2
Alcoholic
Oxygen,
weight percent in
neutral fraction
19.3
5.0
3.0
4.6
4.0
4.4
1.7
1.7
0.5
5.0
3.5
Carbonyl
Oxygen,
weight percent in
neutral fraction
4.9
5.5
4.2
5.2
7.3
3.2
4.5
4.4
13.1
11.1
12.1
Total
Oxygen,
Alcoholic Plus
Carbonyl,
weight percent
24.2
10.5
7.2
9.8
11.3
7.6
Ul
CD
6.2
6.1
13.6
16.1
15.6
tive Exhaust
33
21
54
-------
TABLE 14. SUMMARY OF METHYLENE CHLORIDE FRACTIONATIONS
H
H
m
r
r
m
n
o
r
C
2
ID
C
in
Site:
Date:
lethylene Chloride Extractables, weight percent
Carbon, weight percent
Hydrogen, weight percent
Nitrogen, weight percent
[Oxygen] , weight percfent
ractionation
1 I Water-soluble fraction, weight percent
I Water-insoluble fraction, weight percent
1 lAcid fraction, weight percent
M Basic fraction, weight percent
1 Neutral fraction, weight percent
JjAlcohol fraction, weight percent [O]
ICarbonyl fraction, weight percent [0]
New York
Aug. 23, 1972
13
69.1
9.1
1.1
20.7
15
85
30
5
50
1.5
2.1
Denver
Nov. 17, 1973
45
76.5
10.5
0.8
12.2
14
86
18
3
65
0.3
8.5
Rubidoux
Sept. 21, 1973
9
60.6
7.9
2.2
29.3
42
58
24
1
33
1.7
3.6
W. Covina
Sept. 21, 1973
15
65.7
8.6
1.5
24.2
42
58
20
2
36
1.3
4.3
Irradiated
Auto Exhaust
11
61.6
7.7
1.9
28.8
54
46
12
1
33
10.9
6.9
01
CD
-------
60
similarities can be seen in the remainder of these data, no significant pattern emerges, particularly
in comparisons with the average auto-exhaust sample. In the last set of data on functional
groups, the Denver and auto exhaust samples stand out as irregular — compared to the urban col-
lections the Denver sample is low in alcohol and high in carbonyl constituents while the exhaust
sample is high in both alcohol and carbonyl fractions.
Trends in these data are consistent with those from the spectroscopic analyses. In general,
the West Coast samples (Rubidoux and West Covina) showed somewhat higher carbonyl concen-
trations. And, in accord with the total N determinations, the nitrate band intensities for
Rubidoux, West Covina, and irradiated auto exhaust were greater than for Denver and New York.
In establishing the overall characteristics of the organic composition of light-scattering aero-
sols, we were looking for gross compositional differences. We found only slight and moderate
differences among samples from different regions, with the exception of the unusual Denver
sample. There were also only moderate organic compositional differences among days of differ-
ent air quality in the same region. Among the trends is an expected one that a higher percentage
of oxygenated material is present on days of higher light scattering.
If it is desirable to arrive at some tentative, undefendable conclusion with respect to rela-
tionships between auto exhaust emissions and the organic fraction of light-scattering aerosols, one
might state that the West Coast samples showed some significant similarities to the composition
of the auto exhaust samples, the New York samples showed marginal similarities, and the rural
Denver samples showed no similarities. It must be stressed that only a very limited number of
aerosol samples have been analyzed for organic constituency. It is our hope that the knowledge
gained here, when combined with continuing work on the analysis of both smog chamber (model
systems) and atmospheric aerosols, will eventually shed more light on the chemistry of organic
aerosol formation.
BATTELUE — COLUMBUS
-------
61
REFERENCES
(1) Wilson, W. E., Schwartz, W. E., and Kinzer, G. W., "Haze Formation: Its Nature and Origin",
First Year Report by Battelle's Columbus Laboratories to the Coordinating Research Council
and the Environmental Protection Agency (1972).
(2) Miller, D. F., Schwartz, W. E., Jones, P. W., Joseph, D. W., Spicer, C. W., Riggle, C. J., and
Levy, A., "Haze Formation: Its Nature and Origin", EPA-650-3-74-002, NERC, Research
Triangle Park, N.C. (June, 1973).
(3) McNulty, R. P., Atmospheric Environment, 2, 625 (1968).
(4) Horvath, H., Atmospheric Environment, 5_, 177 (1971).
(5) Beuttelf, R. G.r and Brewer, A. W., J. Sci. lnstr.,_26, 357 (1949).
(6) Ahlquist, N. C., and Charlson, R. J., Environ. Sci. and Technol., 3_, 363 (1968).
(7) Samuels, H. J., Twiss, S., and Wong, E. W., "Visibility, Light Scattering and Mass Concentra-
tion of Paniculate Matter", Report of the California Tri-City Aerosol Sampling Project to the
State of California Air Resources Board {July, 1973).
(8) Charlson, R. J., Ahlquist, N. C., Selvidge, H., and McCready, P. B., J. Air Poll. Control Assoc.,
Jj>, 937 (19691.
(9) Buchan, W. E., and Charlson, R. J., Science, 159, 193 (1968).
(10) Covert, D. S., Charlson, R. J.f and Ahlquist, N. C., J. Appl. Meteor., ]T, 968 (1972).
(11) Lundgren, D. A., "Atmospheric Aerosol Composition and Concentration as a Function of
Particle Size and Time", paper no. 69-128 presented at the 62nd Annual Air-Poll. Control
Assoc. Meeting, New York City (1969).
(12) Whitby, K. T., Husar, R. B., and Liu, B.Y.H., J. Colloid Interface Sci., 39, 177 (1972).
(13) Hidy, G. M., "Theory of Formation and Properties of Photochemical Aerosols", paper pre-
sented at the Battelle School on Air Pollution, Seattle, Wash. (1973).
(14) Blosser, E. R., "A Study of the Nature of the Chemical Characterising of Particulates Col-
lected from Ambient Air", final report from,'Battelle's Columbus Laboratories to the National
Air Pollution Control Administration (1970).
«
(15) John, W., Kaifer, R., Rahn, K., and Wesolowski, J. J., Atmospheric Environ., 7, 107 (1973).
(16) Lee, R. E., Goranson, S. S., Enrione, R. E., and Morgan, G. B., Environ. Sci. and Technol.,
6, 102511972).
(17) Winchester, J. D., and Nifong, G. D., Water, Air, and Soil Pollution, j_, 50 (1971).
(18) Gladney, E. S., Zoller, W. H., Jones, A. G., and Gordon, G. E., Environ. Sci. and Technol., 6,
551 (1974).
BATTELLE — COLUMBUS
-------
62
(19) Lee, R. E., and Patterson, R. K., Atmospheric Environ., 3, 249 (1969).
(20) Wagrhan, J., Lee, R. E., and Axt, C. J., Atmospheric Environ., J_, 479 (1967).
(21) Miller, M. S., Friedlander, S. K., and Hidy, G. M., J. Colloid and Interface Sci., 39, 165
(1972).
(22) Friedlander, S. K., Environ. Sci. and Technol., 7, 235 (1973).
(23) Heisler, S. L, Friedlander, S. K., and Husar, R. B., Atmospheric Environ., 7, 633 (1973).
(24) Novakov, T., Mueller, P. K., Alcocer, A. E., and Otvos, J. W., J. Colloid and Interface Sci.,
39, 225 (1972).
(25) U. S. Department of Health, Education, and Welfare, "Air Quality Data, 1966", Durham,
N.C. (1968).
(26) Hauser, T. R., and Pattison, J. N., Environ. Sci. and Technol., 6, 549 (1972).
(27) Ciaccio, L. L., Rubino, R. L., and Flores, J., Environ. Sci. and Technol., 10, 935 (1974).
(28) Mader, P. P., McPhee, R. D., Lofberg, R. T., and Larson, G. P., Industrial and Engineering
Chem., 44, 1352 (1952).
(29) Gemma, J. L., and Miller, D. F., "A Model of Urban Visibility Based on Air Quality",
presented at the Annual American Chemical Society Meeting, Atlantic City, N.J. (1974).
(30) Air Pollution, Edited by A. C. Stern, Second Edition, Vol. 1, Academic Press, New York
(1968), "Meteorology and Air Pollution" (R.C. Wanta) 187-224.
(31) Sonquist, J. A., and Morgan, J. N., The Detection of Interaction Effects, SRC Monograph
No. 35, University of Michigan Institute for Social Research, Ann Arbor, Mich. (1964).
(32) Box, G.E.P., and Jenkins, G. M., Time Series Analysis; Forecasting and Control, Holden-Day,
San Francisco, Ca. (1970).
(33) Mueller, P. K., Mosley, R. W., and Pierce, L. B., J. Colloid and Interface Sci., 39, 235
(1972).
(34) Spicer, C. W., "Fate of Nitrogen Oxides in the Atmosphere", Final Report from Battelle's
Columbus Laboratories to the Coordinating Research Council and the Environmental
Protection Agency, Project CAPA-9 (1974).
(35) Altshuller, A. P., Environ. Sci. and Technol., 7_, 709 (1973).
(36) Levy, A., Miller, D. F., Hopper, D. R., Spicer, C. W., and Trayser, D. A., "Motor Fuel
Composition and Photochemical Smog", Interim Report from Battelle's Columbus Labora-
tories to the American Petroleum Institute, API Report No. CEA-4 (1973).
(37) Cothup, N. B., Daly, L. H., and Wiberly, S. E., Introduction to Infrared and Roman Spec-
troscopy. Academic Press, New York (1964).
BATTELLE — COLUMBUS
-------
APPENDIX A
STATISTICAL METHODS
BATTELUE — COLUMBUS
-------
A-1
APPENDIX A
STATISTICAL METHODS
AID Analysis
The procedure followed in AID is to divide the sample into a set of mutually exclusive and
exhaustive subgroups, through a succession of binary splits of the data. At each step, the split is
based on values of one predictor variable. The predictor variable, and its values associated with
each subgroup of the split, are chosen in such a way as to minimize the error sum of squares of
the dependent variable. The final result of applying the AID procedure is a classification tree,
which yields a sequential decision rule based on predictor variables. The result of each "path"
of the decision rule is a final subgroup. Since each final subgroup is characterized by the mean
and standard deviation of dependent variable values included in it, and since the paths leading to
final subgroups can be characterized by a sequence of sets of values of predictor variables, a quan-
titative statement can be formulated for each path and used for purposes of classification or
prediction.
AID has advantages over traditional statistical methods such as regression analysis. With
AID, no assumptions about linearity or homoscedasticity need be made. Interactions among pre-
dictor variables relative to a dependent variable do not invalidate an AID analysis. In fact, AID
helps to identify such interactions in the data.
The following is a mathematical description of the AID algorithm.
1. The total input sample is considered the first (and indeed only) group at the
start.
2. Select that unsplit sample group, group i, which has the largest total sum of
squares
/ N, y*
. N> 2 UY°)
TSSj = I, Y2 j^p , (1)
a=1 '
such that for the i'th group
TSSj > R (TSST) and NJ > M , (2)
«
where R is an arbitrary parameter (normally .01 < R < .10) and
M is an arbitrary integer (normally 20 < M < 40), and TSSf denotes
the total sum of squares for the total input sample.
The requirement (2) is made to prevent groups with little variation in them,
or small numbers of observations, or both, from being split. That group
with the largest total sum of squares (around its own mean) is selected, pro-
vided that this quantity is larger than a specified fraction of the original
total sum of squares (around the grand mean), and that this group contains
more than some minimum number of cases (so that any further splits will
BATTELLE — COLUMBUS
-------
A-2
be credible and have some sampling stability as well as reducing the error vari-
ance in the sample).
3. Find the division of the C^ classes of any single predictor X|< such that com-
bining classes to form the partition p of this group i into two nonoverlapping
subgroups on this basis provides the largest reduction in the unexplained sum
of squares. Thus, choose a partition so as to maximize the expression
p (3)
where Nj = n-| + r\2
and Yf =
for group i over all possible binary splits on all predictors, with restrictions that
(af the classes of each predictor are ordered into descending sequence, using
their means as a key and (b) observations belonging to classes which are not con-
tiguous (after sorting) are not placed together in one of the new groups to be
formed. Restriction (a) may be removed, by option, for any predictor X)<.
4. For a partition p on variable k over group i to take place after the completion
of step 3, it is required that
BSSjkp > Q (TSST) , (4)
where Q is an arbitrary parameter in the range .001 < Q < R, and TSSf is the
total sum of squares for the input sample. Otherwise group i is not capable of
being split; that is, no variable is "useful" in reducing the predictive error in
this group. The next most promising group (TSSj = maximum) is selected via
step 2 and step 3 is then applied to it, etc.
5. If there are no more unsplit groups such that requirement (2) is met, or if, for
those groups meeting it, requirement (4) is not met (i.e., there is no "useful"
predictor), or if the number of currently unsplit groups exceeds a specified
input parameter, the process terminates.
Figures A-1, A-2, and A-3 show the results obtained from AID analyses with light scattering
as dependent variable. Jn order to understand the results, a description of how to analyze the
output is included here. The criterion scale on the upper left of Figure A-1 is in units of the de-
pendent variable, light scattering. The tabular information on the right begins with descriptive
statistics about the sample. The criterion variable is given (LIGHT SCAT), there are 432 obser-
vations in the sample (18 days with 24 hourly averages of data each), and the average value of
light scattering in this sample is 2.79 with a standard deviation of 1.58. The disk containing
the number one represents this sample on the graph, and is located at the mean value (2.79) ol
the light-scattering observations in the sample.
The sample is split first by N02 into subgroups 2 and 3. On the graph, the subgroups are
located at the mean value of the light-scattering observations in each subgroup. Information
pertaining to this split is contained in the table immediately to the right of the graphical display,
Hence, subgroup 2 contains 305 observations with an average value of light scattering of 2.1 1
BATTELLE — COLUMBUS
-------
AUTOMATIC INTERACT I Ol\l DETECTOR
HAZEVECTUR AID ANALYSIS
1 .00
BINARY TREE STRUCTURE
CRITERION sroiE
2.17 3. 31 <«.52
SUMMARY TABLE
5.69
6. 86
•=]
ffl
f
p
m
I
n
e
TOTAL GROUP
CRITERION -LIGHT STAT
TOTAL GROUP N = 132
MEAN = 2.79
STD. DEV. = 1.58
PARENT 1 SPLITTING VARIABLE - N02
PEON* 2.11 S.D.= 1.10 N: 305
PREDICTOR VALUES — 01231
5 6 7 6 <> 10 11 12 13 11 U
ffEAN- 1.«2 S.0.= l.3fr fu* )?r
PREDICTOR VAI CFS --. It l» IB 19 20
21 22 23 21 25 2( 27 26 29 31 31 33
PARENT 2 SPLITTING VARIABLE - TE"P
m£ON= 1.11 5-0.= 0.65 N- 106
PREDICTOR VALUES — 01215
6 7 a 9 10 11 12 13 11 15 16 IT
16 14 20
BEA«I= 2-1' S.0.= 1.03 (•= 199
PREOirTOS VAI I'FS -- 21 22 23 2« !•>
26 2T 25 29 3C 31 3Z 33 31 35 3* 3'
PARENT 3 SPLITTING VARIABLE - TO
BEON= 1.01 S.D.- l.H N= 100
PREDICTOR VAIUES -- 2 3 1 6 t
9 10 11 12 13 11 15 16 17 16 19 20
21 22 23 21 2; 26 27 26 29 30 31 .
BEAN= 5.96 S.0.= 1.09 «= ?7
PREDirTOB VAI I'FS -- 3f 39 «C "1 "Z
qj 141 15 «6 if 50 52 5" 5S 5* 5f 59
61 63
PARENT 5 SPLITTING VARIABLE - HUMIDITY
BfoN= 1.59 5.0.= 0.51 N= 56
PREDICTOR VAIl'ES — 1 2 7 9 10
11 12 13 11 15 16 17 16 19 20 21 22
23
FINAl CROUP
I"E.A«= 1. ft 5.0.= 0.97 «1= 1-1
PREDICTOR VAI I'FS -- 21 25 2* 27 ?F '•
29 30 31 32 33 3* 35 36 37 36 39 10
PARENT 9 SPLITTING VARIABIE - f»FT«
HIAN= 2.66 S.D.- o.eo N= 121
PREDICTOR v«l UES — 01231
5 6 7 8 9 10 12 |1 15 16 17 18
19 20 21 22 23 21
*EAN? 1.01 S.p.= 1.0" "•- 2C
PREDICTOR VAI I'FS -- 25 !t 2r If ft
30 31 33 31 3T «|. i
FINAI FROUP :
PARENT 6 SPLITTING VARIABLE - TO
BEAN- 2.61 S.D.- 0.79 N- 13
91011
F1NAI GROUP
F"E«N= 1.21 S.p.= "1. 00 U= 6! i
17 16 19 20 21 22 23 21 25 2f 27 26
29 30 31 32 31 35 36
PARENT 13 SPLITTING VARIABLE - HI'fllOITY
MEAN: 3. 6» S.D.- o.n N: 11
PREDICTOR VAIUFS — 1 11 19 21 25
26 27 28 29 30 31 32 33 31 35 36
FINAL PROIIP
nFAN= 9.69 S.D = 1 02 N= 16
PREDIC'IOR VAU'FS -- 37 30 39 10 1|
12 13 11 15 16 17 19 51 52 51 62
>
CO
FIGURE A-1.
-------
HAZFVETTOR AID ANALVSIS
...CONTINUED...
I .00
2.17
CRITERION SCAI C
3.3t t.52
5 . 69
6.86
«=}
fii
f
f
ft
I
0
s
r
e
e
o
summARv CONTINUED i
PARENT 10 SPLITTING VARIABLE - UINO SP j
HEAN= 2.27 S.D.= 0.88 N= 10
PREDICTOR VALUES — 16 17 18 19 20
13 11 15 63
HEAN= 2.et S.D.= G.67 IV;
PREDICTOR VAI I'fS -- 123*
PARENT t SPL-ITTING VARIABIE - HUCIIDITY
BEAN= 1.07 S.O.- 0.31 N= 71
PREDICTOR VALUES — 0 3 6.811
25 26 27 28 29 30 31 32 33 31
FINAI GROUP
nEAN= 2.21 5.0.= I-.C9 1-
PRE01CTOR VALUES — 35 37 3f 3*
61 62
PARENT 15 SPL ITTING. VARIABLE - TEPP
REAM: 3.02 s.o.= o.«i N= 15
PREDICTOR VALUES' — 11 20 21 21 25
26 27 29
FINAL CROUP
*EAN= 5.11 s.o.= C.M N=
PREDICTOR VAll'Fl -- 30 31 32 33
35 36 37 38 39 10 i| • 1' If 1
21 22 23 2i i'. It n 2f> 3C 31
FINAL r-ROMP
PARENT 7 SPLITTING VARIABIE - TP
PEAN; ,5.3" S.o.= 0.78 N: 16
PREDICTOR VAI 1>ES — 38 39 "0 11 42
43 44 45 4( 48
FINAI FROliP
PIFAN= 6.f>l> S.P.- C.t2 '»-'
PPEPKTOR VAM'F!- -- 'C '? "• "
5f '.9 M t3
FINAI C.ROI'P
It- ',
t •
« .20:
32 ...
1 1 :
PARENT 17 SPl ITTING VARIABLE - OZONE' i
PEAN= 2.01 S.'0'.= 0.70 N: 31
PREDICTOR VOICES -- 01235
6 8 9 II 13 11 15 16 17 22 23 27
29 30 31 39'
FINAI CROUP
BEAN= 3.57 S.p.= C.67 N=
PREDICTOR VAI I'FS -- 10 «? 44 4«
FINAI RROI'P
..j
. i
FIGURE A-1. (Continue^
-------
AUTOMATIC INTERACTION DETECTOR
HAZEVECTOR AID ANALYSIS
1.10
BINARY TREE STRUCTURE
CRITERION SCAIE
2.25 3.10 ^.56
SUPIftARY TABLE
5.71
6. 86
ft
F
F
ffl
I
o
0
F
e
TOTAL
GROUP
CRITERION -LIGHT SCAT
TOTAL GROUP N = '•32
MEAN = 2.79
STD: DEV. = i .58 i
PARENT
PREDICTOR
5 6 T
PARENT
PEAN- 1
PREDICTOR
10 18 19
10 15 56
PARENT
I"IFAN= 1
PREDICTOR
12 13 11
1 SPLITTING VARIABLE - N02
.11 S.D.= 1 10 N= 305
VALUES — 01-231
8 9 10 11 12 13 11 15
REAM: 1.1; s.o.= 1.3* »- 127
PREDICTOR VAI I'FS -- 1* 17 ie 19 20
II 22 23 21 25 26 27 2P 29 31 32 33!
31 35 37 3P. 3« «C i| 12 13 "« 17 .
2 SPLITTING VARIAB1E - UIWD VEI ;
.10 S.O-- 0.62 N- lot
VALUES — 23i 5 6
59 40 61
PFAN= 2. -9 S.P.= 1 . II »= l«« :
PREDICTOR VAII'Ft -- 0 1 t « 16 ,
50 51 «t 57 j
' * SPl. ITTIHir- VARIAPI F - Hi'l"irl:TV
.It S.D.= 0.60 K- 16
VALUES — 2 6 7 10 11
15 17 18 20 21 22 23
f(CN- 2.(C S.r.- 1.0" 1- 153 :
PREOICTOS VAI I'FS -- 2« 25 26 ?7 If \
29 30 31 32 33 3- 35 3t 31 3f 3< «C:
N= 1.01 S.D.= 1.11 N= 100 rF«N= -5.«t S.p.= l.0« «= 27 ;
PREDICTOR VAU'ES — Z 3 « 6 f PRFDI"CP VAI I'FS — 3( 31 *0 11 «J |
1 10 I! 12 13 It 15 It 17 IS 19 20 13 1« -5 "* «P 50 52 5« ?c 5» 5P 5«.
21 22 23 21 25 2» 27 28 29 30 31 .; H t-3 i
I- £ r r M
SPl ITT INP VOPIPPi F -
PARENT
SPLITTINP VARIOBIE -
P .
2» •« it i
PARENT 11 SPLITTING VARIABLE - dETH
BE«N= 2.76 S.0.= 0.78 H= 109
PREDICTOR VALUES — 01231
S I 7 ( 9 10 12 11 15 16 IT 16
19 20 21 22 23 21
HEAN= 3.90 S.D-- 1
PREDICTOR VAI I'FS — 25
30 31 33 3« Jf 37 1*
FINAI C.ROI'P
01
26
N= 25
2T It 29
FIGURE A-2.
-------
HAZFVECTOR ' AID ANAtYSIS
-..CONTINUED.-.
1.10
Z .25
PR I tFRlON SfPI F
3.10 1.56
5. 71
6.86
3>
=J
•4
ff)
r
r
ra
I
n
SUMMARY CONTINUED
PARENT 13 SPLITTING VARIABLE - TO
nifH- t.ii s.o.- o.6i N= e
PREDICTOR VALUES — 23168
9 10
Final GROUP
BEAI»= 3.»6 S C. = C.PI H= T!
PREDirlOF VOM'fb -- II 12 13 1" I',
16 17 IP l« 20 21 22 23 2" 25 If 1'
28 29 30 31 32 3« 35 3*
PARENT 11 SPLITTING VARIABLE - WIND VEL 1
KM- 2.12 S.D.= 0.61 «= SI
PREDICTOR VALUES — 21 25 26 20 32
33 36 It 50
REAN= 3.10 S.P.= 0.59 N= 5!
PREDICTOR VAI CES — 0 1 f 9 1*
IT 29 10 19 56 5T '
FINAI CROl'P
PARENT 17 SPLITTING VARIABLE - TO
.
PREDICTOR VALUES — 11 12 13 11 15
16 IT 19 19 20 21 22
FINAL CROUP
!
PREDICTOR VAII'ES -- 23 21 25 it 21
28 29 30 31 32 3« 35 3( :
i
FINAI CROl'P '
PARENT 5 SPLITTING VARIABIE - TFPP
«EAN= 1.11 S.0.= 0.99 N° 67
PREDICTOR VALUES — « 9 10 11 12
13 11 15 16 17 It 19 20 21 22 23 21
26 27
FINAI CROUP
n'EAK= I.f5 S.P.= O.fT «/= 3<
PREDICTOR VAII'FS -- 30 31 32 3" 35
19 52
PARENT 1 -, SPllTr:--: VCRIAPir -
TOAN= 1.69 S.D.= 0.61 N= 12
PREDICTOR VAIUES — 0 1 II 12 11
15
FINAL GROUP
PnEAN= 2.63 S.O.* O.T3 «s «2
PREDICTOR V«l I'FS -- 16 IP 19 20 21
22 23 21 25 2T 28 29 30 32 33 38
PARENT 9 SPLITTING VARIABIE - TO
HEAN= 5.31 S.D.= 0.76 N= 16
PREDICTOR VAIUES — 3t 39 10 11 12
13 11 15 16 18
FINAI CROUP
BEAN; 6.8t S.0.= 0.82 H- II
PREDICIOR VAII'ES — 50 52 $1 55 56 ',
58 59 61 63
FINAI C.ROI'P
PARENT 25 SPLITTING VARIABIE - METH
nfAN= 1.83 5:0.= 0.51 «= 10 IPEAM 2.PP S.P.= ti.•>^ •.= 3< .
PREDICTOR VAIUES — 0123 PREDICTOR v«ll'Ei — " ' * ' *
9 12 11 17 18 19 20 21 22 2*
F1NA1 GROUP |F 1 WAI CROUP
PARENT 23 ' SPLITTING VARIABLE - UIND VEL
REAM: 1.45 S.0.= 0.41 N: 22
PREDICTOR VAIUES — 2 9 4 6 10
37 15 59 6C 61
FINAI CROUP
BEON= 2.3' S.0.= 0.96 O= If
PREDICTOR VAIUES -- 2T 31 31 11 U
11 5«
FINAI CROl'P
o>
FIGURE A-2. (Continued)
-------
HAZEVECTOR AID,ANALYSIS
...CONTINUED...
1.10
=3
4
m
r
F
m
I
n
2.25
CRITERION sroiE
3.10 1.56
5,71
6.86
SUHMARV CONTINUED
'PARENT- • 6. . SPLITTING VARIABLE - TEPP
mE«N= I.It S.D.= 0.3? IU= 30
PREDICTOR VALUES — 0 1 2 1 5
6 T 8 «' lO'll 12 13 16 IT IB 22
23 25 26 28 30 31 33
FINAL CROUP
BE 0*1= 2 03 S.n.- 0.5J Mr If
PBEDlrTOR vail'ES -- 3" 3' "1 «2 «3
4< 45 06 51 53 55
FiNai r-POi'p
FIGURE A-2. (Continued)
-------
flUTOPIATIC INTERACTION DETECTOR
HAZEVECTOR AID ANALYSIS
0.90
BINARY TREE STRUCTURE
CRITERION scate
2 . 00 3.09 <(. 19
SUPKHARY TABLE
5.28
6.38
0
J>
r
r
m
n
0
r
C
TOTAL GROUP
CRITERION -LIGHT .SCAT
TOTAL GROUP N = 132
MEAN = 2.79
STD. DEV. = 1 .58
PARENT I SPLITTING VARIABLE - THT
reAN= 2.11 S.0.= l.ll N= 300
PREDICTOR VALUES — 01231
5 6 7 8 9 10 11 12 13 14 IS 16
17 IB 19 20 21 22 23 21 25
HEAN= 4.34 S.D.= 1.31" 11- 132
PREDICTOR VAIUES -- 26 27 28 29 30
PARENT 2 SPLITTING VARIABLE - U1IIVO VEl
BEAN= 1.53 S.D.= 0.77 K= 117
PREDICTOR VALUES -- 23456
10 16 11 27 26 30 31 34 35 37 41 42
43 44 45 58 59 (0 61
HEAM= 2."8 S.D = I.I" 1- IP;
PREDICTOR VAIL'ES -- 0 1 6 « l»
24 25 26 29 32 33 36 40 -t *•) '.r, .'. 1
56 57
PARENT 3 SPLITTING VARIABLE - AfV
P1EAN= 4/02 S.D.= 1.14 H= 114
PREDICTOR VALUES -- 03 7 10 14
1721 21 28 31 35
WEAN= 6.38 1. P. = 1 -01 l*= If
PREDICTOR V01 tfES -- 3f *-2 "5 52 e.f :
63 !
FIMAI CROI'P
PARENT t SPLITTING VARIABLE - UINP VEl
P1EAH= 2.32 S.D.= 1.03 H= 1 62
PREOICTCH VALUES -- 1 f 9 24 25
26 32 31 36 40 48 SO SI 56 57
ntAI»= 3.6? S.D.= 1.26 fc= 21
PREDICTOR Vfll CEt -- 0 16 29 "9 '
FIIUAI PSOI'P
PARENT 9 SPLITTING VARIABLE - OZOIVE ;
nEAfus 2.22 &.D.= C.98 N- ISO
PREDlTTOIi VALUES -- 01234
5 6 7 8 « 10 11 12 13 |4 15 It
17 18 19 20 21 22 23 25 2t 2? 29 .
ft»n' 3.t2S.c.= c.s» t= iz:
PREClfTOR VAM-Ei -- :2 33 36 37 3r
39 43 44 !
FIfttAL T.ROt'P I
PARENT 6 SPLITTING VARIABLE - NOX
REAM? 3.27 S.O-- 0.87 N- 38
PREDICTOR VALUES -- 6 7 8 9 10
11 12 13 H IS 16
nEAN= 4."C S-D.= 1.07 tt= 76 !
PREDICTOR VALI'ES --17 18 19 20 21
22 23 24 25 26 27 28 29 3C 31 »t :-
37 38 41 "2 44 49 52 55 57
PARENT 10 SPLITTING VARIABLE - THT
KEHN- 1.75 S.0.= O.tl H- 46
PRFOKTOR VALUES -- 01345
6 7 8 9 10 II
AEAN= 2.42 S.O-- 0.98 H- 1 0<*
PREDICTOR VALUES — 12 13 1" 15 16
17 IB 19 2C 21 22 23 2" 25
>
CO
FIGURE A-3.
-------
WAZEVECTOR AID
. . .CONTINUED- - -
ANALYSIS
SCALE
0.90
2.00
3.09
1.19
5.28
6. 38
r
r
ffl
i
n
s
r
C
29 ) 28
SUmPIARV CONTINUED
PARENT 15 . SPLITTING VARIABLE - OZONE
MEAN- , 1.61 S.0.= 0.88 N= 17
PREDICTOR VALUES — 0
FINAL GROUP
nEAN= 2.58 5.0.= 0.92 N= 87.
PREDICTOR VALUES -- 1 2 3 4 5
6 7 8 1 10 11 12 13 11 15 16 IT
18 20 21 22 23 25 26 27 29 31
PARENT 13 SPLITTING VARIABLE - WIND VEL
MEAN=' 4.20 S.D.= 1.00 N= 62
PREDICTOR VALUES — 0 8 25 32 .33
10 11 18 5T ' ' •
REAN3 5.26 S.0.= 0.92 N-
PREOICTOR i/A'lUES — .16 24 19 56
FINAL GROUP
11
PARENT 17. .SPLITTING VARIABLE - NOX
B£AN= 2.32 S.D.= 0.85 H- 61
PREDICTOR VALUES — 15676
» 10 11 12 13 11 !5 16
FINAL GROUP.
MEAN- 3.21 S.0.= 0.'5 «=
PREDICTOR VALUES — 17 18 22 23
25 26 27 28 29 30 36 37 42 "8
FINAL GROUP
26
2'
PARENT 5 SPLITTING VARIABLE - OZONE
nEAN= 1.12 S.0.= 0.63 N= 110
PREDICTOR VALUES — . 0 1 2 3 1
IT 18 19 20 21 22 23 21 26 21 30 .
FIEAN= 3.27. S.D. = 0.*9 N:
PREDICTOR V»l UES — 38 39 10 <*2
16
FINAL GROUP
7
ttn
PARENT 19 SPLITTING VARIABLE - OZONE
nEAN= 3.8$ S.D.= 1.05 N- 31
PREDICTOR VALUES — 0
FINAL GROUP
nEOM= 1.62 S.D.= 0.75 K- 26
PREDICTOR VBLl'ES — 1Z315
6 9 10 11 .11 21 25 28 10 31 3« "5
63 -•
FINAI GROl'P
PARENT 22 SPLITTING VARIABLE - S02 '
BEAN= 1.23 S.D.= 0.52 N- 81
PREDICTOR VAIUES — 0 1 3 8 1
10 11 12 13 11 15 16 17 18 19 20 21
22 23 21
FINAL GROUP
nEAN= 1.95 S.D.= 0.59 N-
PfiEOinOfi VOII'FS -- 25 2f 27 28
30 31 32 38 10 »l
FINAL GROUP
29
29
PARENT 11 SPLITTING VARIABLE - UMNO VEL
HEA«I= 1.20 5.0.= 0.30 tl= 17
PREDICTOR VALUES — 25 32 10 50 56
FINAL GROUP
BEAN* 2.07 S.D.= O.d «l=
PREDICTOR VALUES -- 1 8 9 24
33 48 57
29
26
PARENT 12 SPLITTING VARIABLE - METH
nEAN= 2.96 5.0.= 0.5T N: 28
PREDICTOR VALUES — 5 6 7 61
10 12 11 15 16 17 IB 19 20 21 21
FINAL GROUP
BEON= 4.15 S.O-- 0.94 N=
PREDICTOR VA1 UES -- 25 26 27 28
32 33 37 16
FINAL GROUP
10
30
>
CD
FIGURE A-3. (Continued)
-------
o
c
HAZEVECTOR flID ANALYSIS
. . .CONTINUED. . .
H
H
m
r
r
m
n
o
r
C
0. 90
2 . 00
CRIIERION SCALE
3.09 ^
5 .
6 . 38
SUNHARY CONT INUED !
PARENT 26 SPLITTING VARIABLE - NO
nCANr 0.985.0= 0 . 1 7 N= 5
PREDICTOR VALUES 1 2 3
F INOL GROUP
PREDICTOR VAl I'ES -- 5 6 1 8 «
11 12 15 16 17 31
F 1 Nfil PfiOl'P '•
>
o
FIGURE A-3. (Continued)
-------
A-11
and a standard deviation of 1.10; the corresponding information for group 3 is 127 observations,
a mean of 4.42, and a standard deviation of 1.36. Further, the table lists the values of the pre-
dictor variable |NC>2) associated with subgroups 2 and 3. The coded values 0 through 15 are
associated with group 2, and the coded values from 16 through 63 are associated with group 3.
Asterisks at the end of a list signify that more coded values follow, but no space exists to com-
plete the list. Each predictor variable is coded initially into integer values in the range 0 to 63.
The coding is done as follows: let R denote the range of the observed values, let z denote the
minimum of the observed values, and let x denote a predictor value; the coded value c is given
by
c =
62.5 - + 0.5
R
where [a] denotes the greatest integer less than or equal to a.
In words, one may describe the first split loosely as follows: when NO2 values are in the
lower quartile of their range, the corresponding light-scattering readings have a mean of 2.11;
when the N02 values are higher, the corresponding light-scattering readings have a mean of 4.42.
The remaining splits may be interpreted in a similar fashion. The classification rules in the
tree may be verbalized for parts of the tree. For example, one can say that the average light
scattering is 5.96 given that the N02 reading is in the upper 75 percent of the N02 range and
the CO reading is in the upper 37 percent of the CO range. As a practical way of choosing im-
portant variables and interactions from an AID tree, one generally observes the variables associ-
ated with the first "several" splits. It is possible by utilizing various stopping rules, formal statis-
tical tests, and iterative procedures to refine an AID tree to a well-formed rule. However, the
purpose here was to select several variables for the time-series analysis which would explain a
portion of the variance in the light scattering, so the initial trees suffice.
Although temperature shows up as a potentially important splitter in this AID analysis, sub-
sequent attempts to develop a transfer function relating light scattering and temperature failed to
obtain a significant relationship. A possible explanation may be that temperature is only signifi-
cant in its interaction with IM02 in the analysis, and no attempt was made to construct inter-
action variables for the time-series analysis.
The wind-speed (WIND SP) variable in Figure A-1 was obtained by first computing a wind-
velocity vector. The 10-minute wind-speed and wind-direction values were formed into vectors,
and an hourly average wind-velocity vector was computed using vector addition. The wind-speed
variable used in the analysis was simply the wind-speed component of this vector. A previous
analysis had used a straight averaged wind speed. However, neither the previous analysis nor this
one showed much significance for wind speed.
Figure A-2 shows the results of an AID analysis of the sample when the wind-velocity vector
(WIND VEL) replaces wind speed. The coding scheme for wind velocity is necessarily different
from that used for the other variables. Since a wind-velocity value may be associated with a
point in the plane, a portion of the plane including all the wind-velocity values in the data was
divided into segments. This procedure is illustrated in Figure A-4. The plane is in polar co-
ordinates, with the r-coordinate representing the wind-speed component, the 0-coordinate repre-
senting the wind-direction component, and 0° and 360° representing due north. Concentric
circles are drawn in increments of 2 mph to a maximum of 16 mph. The circles are divided into
octants and the resulting partition of this portion of the plane coded as shown. As an example,
BATTELLE — COLUMBUS
-------
A-12
FIGURE/ A-4. CODING SCHEME USED TO DEVELOP WIND DIRECTION AND VELOCITY VECTORS FOR
SUBSEQUENT ANALYSIS OF WIND EFFECTS ON LIGHT SCATTERING
BATTELLE — COLUMBUS
-------
A-13
a wind-velocity vector of (4.5, 100°) is coded as 18. Since no order relationship exists in this
code, potential splits on wind velocity could occur on an arbitrary division of wind-velocity codes
into two subsets, whereas splits on other variables were restricted to being monotone [i.e., all
values of the predictor associated with one subgroup of a split must be greater than (or less than)
all values associated with the other subgroup.]
Figure A-2 shows the increased power of the new variable. Wind velocity becomes the second
most powerful splitter and also occurs elsewhere in the tree. Figure A-5 illustrates the split of
group 2 into groups 4 and 5 on wind velocity. The hatched areas are associated with group 5
and the dotted areas are associated with group 4. This figure illustrates the tendency of light
scattering to be higher when the wind-speed component is lower, but also shows that wind
direction is a factor in separating "lower" from "higher" wind speeds.
While this analysis is highly informative, and shows the power which certain interactions may
have, it was not possible to include the vector-valued wind-velocity variable in the time-series
analysis. Devising a method to incorporate such variables is left to future studies.
An effect called "masking" can sometimes occur in AID analyses. This occurs when one vari-
able is correlated with other "stronger" variables in the AID sense. While such a variable may be
useful in time-series analysis it may not show up initially as a strong variable in an AID analysis.
One such variable in this data is THC, and this is shown in Figure A-3. This run was made after
removing the variables N02, CO, RH, and temperature from the set of predictor variables.
As a result of the AID runs and some preliminary time-series analysis, it was decided to use
N02, CO, RH, and THC as predictor variables in a transfer-function development for light
scattering.
Time-Series Analysis
A time series is a sequence of observations, denoted {zt}t=i, taken at equally spaced time in-
tervals. The mathematical formulation of the Box -Jenkins univariate time-series model is given by
0Q + 0Q(BS)0q(B)at (1)
where
s denotes the length of the period of seasonality;
B is the backward shift operator defined by Bzt = Zf-i;
V, Bs, Vs, V^, v[?are operators defined in terms of B by
V= 1 - B (backward difference operator),
Bs = B(BS~1) for s > 1 where B° = 1,
Vs = 1 - Bs (backward seasonal difference operator),
far d > 1 where V° = 1,
VD= VS(VD-1) for D > 1 where V° = 1;
p, $p, ©Q, 6q are polynomials with constant terms equal unity of degrees
P, p, Q, and q in their respective arguments;
BATTELLE — COLUMBUS
-------
Group 5, mean bscot = 1.40
A-14
fl Group 4, mean bscat=2.49
FIGURE A-5. SCHEMATIC REPRESENTATION OF AID RESULTS WHERE WIND VELOCITY IS
A DIRECTIONAL VELOCITY VECTOR
BATTELLE — COLUMBUS
-------
A-15
is a constant;
at is a sequence of "white noise", that is, at is N(0,oa) and is uncorrelated
for different times.
To illustrate the meaning of these operators and polynomials, we give some examples:
(a) Vzt= (1 - B)zt = zt- zt_-|
(b) B24zt = zt_24
(c) V24zt = zt-zt_24
(d) v2Zt = V(zt-zt.1)
= zt - zt.-| - (zt.-| - zt_2)
= zt - 2zt_! + zt.2
(e) 02(B) = (1 - .758 + .5B2)
= zt - .75zt_i + .5t_2-
A shorthand notation for (1) if ®Q = 0 is (p, d, q) X (P, D, Q)s. 0g ^ 0 signifies a deter-
ministic trend. 0p(B) and $p(Bs) are called autoregressive operators; 0q(B) and 0Q
-------
A-J6
TABLE A-l. SUMMARY OF UNIVARIATE ANALYSIS (WITH SEASONAL DIFFERENCING)
Light
Scattering
N02
CO
RH
THC
Model Parameters Residual Variance
<2,1.0XO,1,2(24 01= .2031.07* .282
02= -.3381.07
61 = 1.0631 .07
' !
©2= -.1821 .06
(2,1,0)(0,1,1)24 01= -.131 1.07 .133x10-3
02 = -.290 ± .07
0-1= .9001.02
(0,1,2)(0,1,1)24 9 -| = -.255 1 .07 .116
02= .2371.07
@}= .8961.02
(2,1,OXO,1,1)24 01= .1151.07 11.198
02= -.1301.07
0,= .8521.03
(0,1,0)(0,1,1)24 ©!= .8621.025 .120
Residual Autocorrelation
X^/Degrees of Freedom
83.09/56
47.71/57
47.57/57
40.57/57
47.27/59
1 standard deviation of parameter estimate.
TABLE A-2. SUMMARY OF UNIVARIATE ANALYSIS (WITHOUT SEASONAL DIFFERENCING)
Light
Scattering
N02
CO
RH
THC
Model Parameters Residua! Variance
(2,1,OX1,0,0|24 0!= .2111.07* . .300
02 = ..244 1 .07
4^ = .0501.066
(2.1,0X1,0,0)24 0! = -.0941. 07 .140X1Q'3
02= -.2241.07
*1 = .285 1 .06
10,1,2X1,0,0)24 0-t= -.2881.07 .137
02= .1681.07
4>1 = .2031.06
(0,1,1X1,0,0)24 0T= -.240 1.13 12.300
! = .1321.07
(0,1.0X1,0,0)24 $1 = .1961.07 .129
Residual Autocorrelation
X2/Oegnees of Freedom
83.67/57
61.73/57
55.28/57
54.08/58
39.78/59
1 standard deviation of parameter estimate.
BATTELLE — COLUMBUS
-------
A-17
The coefficients of the polynomials ojj(B), ftj(Bs), 6j(B), Aj(Bs) are determined from the cross
correlations of various lags between the prewhitened input series and the transformed output
series. The prewhitened input series are the residual series of the corresponding univariate time-
series models for the input series. The transformed output series is obtained by operating on the
output series with the same formal model for each input series. The noise series is then modeled
by univariate analysis and all parameters refit by a least squares analysis.
The X^/Degrees of Freedom values listed in Tables A-1 through A-4 are obtained from a
"portmanteau" test of the hypothesis of model adequacy. This test is described by Box and
Jenkins'32). Some of the values obtained may seem higher than ones generated from white noise,
but an examination of the data shows that one or two large values obtained from differencing can
have a severe impact on the "portmanteau" X , and that trying to reduce the X* statistic by
further fitting would only distort the model.
Tables A-3 and A-4 contain the statistical information on the transfer-function models devel-
oped. The models in Table A-3 were developed with seasonal differencing; the models in Table A-4
were not seasonal differenced.
ATTELLE — COLUMBUS
-------
A-18
TABLE A-3. SUMMARY OF TRANSFER FUNCTION ANALYSIS (WITH SEASONAL DIFFERENCING)
1 Xj=N02
X2 = CO
X3= RH
II X!=N02
X2 = CO
X3 = THC
III X1=N02
X2= RH
X3 = THC
IV X^CO
X2= RH
X3 = THC
Transfer Function
CJn COl
12.67012.9* -10.98112.8
.6531.09 .1791.09
.042 1 .009 0.
12.43413.0 -8.10413.0
.5731.10 .2341.11
.2731.10 -.2531.1
19.74012.9 -6.90412.9
.053 1 .01 0.
.3031.1 -.2551.1
.7191.09 .0051.09
.037 1 .009 0.
.2721.1 -.2251.1
Cross
Correlation
X2/DF
Noise Model (X,-Noise)
co2 0t 62 6,
0. -.1531.081.1271.081.6701.05 85.15/48
.232 1 .08 Residual Variance = .146 70.25/47
0. x2/DF (Noise-Noise) = 50.92/48
0. -.256 1 .08 1.074 1 .08 |.71 1 1 .05 80.83/48
.2911.09 Residual Variance = .155 70.78/47
.0301 .10 x2/DF (Noise-Noise) = 49.03/48 55.69/47
0. -.0271.091.1471.081.6811.06 84.33/48
0. Residual Variance = .182 45.41/49
0. x2/DF (Noise-Noise) = 53.19/48 56.30/47
.2751.09 .2511.08|.118±.08|.6711.05 66.24/47
0. Residual Variance = .157 46.22/49
.0881 .1 x2/DF (Noise-Noise) = 40.60/48 52.36/47
* 1 standard deviation of parameter estimate.
TABLE A-4.
1 X! = N02
X2 = CO
X3= RH
II XT = NO2
X2 = CO
X3 = THC
III X1 = N02
X2= RH
X3 = THC
IV X^CO
X2=RH
X3 = THC
SUMMARY OF TRANSFER FUNCTION ANALYSIS (WITHOUT SEASONAL DIFFERENCING)
Transfer Function
14.23912.72* 0.
.468 1 .09 .032 ± .08
.0461.01 -.0181.01
12.12212.9 0.
.456 1 .09 .089 1 .08
.3461.09 -.2391.09*
18.89512.6 0.
.0421.01 -.0181.01
.2511.09 -.1691.09
.6141.08 -.0021.08
.0421.01 -.0041.01
.3121.09 -.1371.09
Cross
Correlation
X2/DF
Noise Model (Xj-Noise)
0. -.072 1 .07 |.101 1 .07 68.28/49
.129 1 .07 Residual Variance = .155 57.34/47
0. x2/DF (Noise-Noise) = 68.30/48 37.01/48
0. -.12310.71.1001.07 53.45/49
.2101 .08 Residual Variance = .169 55.10/47
0. x2/DF (Noise-Noise) = 82.57/48 48.61/48
0. -.033 1 .07 1.122 1 .07 63.98/49
0. Residual Variance = .172 39.52/48
0. x2/DF (Noise-Noise) = 79.35/48 47.45/48
.1421.08 -.1091.071.1361.07 60/87/47
0. Residual Variance = .165 33.80/48
0. X2/DF (Noise-Noise) = 75.49/48 39.94/48
* ± standard deviation of parameter estimate.
BATTELLE — C O L. U
BUS
-------
APPENDIX B
ORGANIC ANALYTICAL PROCEDURES
BATTELLE — COLUMBUS
-------
B-1
APPENDIX B
ORGANIC ANALYTICAL PROCEDURES
Solvent Extraction of Participate Matter
Samples were subjected to either methylene chloride soxhlet-extraction {20 hours) or to
sequential soxhlet-extraction first using methylene chloride (20 hours) and then dioxane (44
hours). High purity "distilled-in-glass" solvents were obtained from Burdick and Jackson,
Muskegon Michigan; dioxane was additionally redistilled before use.
Extractions were conducted in soxhlet apparatus of 50 ml capacity fitted with 100 ml RB
flasks; 25 mm x 80 mm thimbles were used. Thimbles were preextracted with methylene chlo-
ride (48 hours) and methanol (48 hours) before use. Up to three 6-inch-diameter filter disks or
one 8-inch x 10-inch filter could be extracted in one thimble. For each analytical extraction 100
ml of solvent was employed.
The volume of each methylene chloride extract was reduced to approximately 0.5 ml by
vacuum distillation (250 mm Hg at 35 C pot temperature). This was accomplished using 25
ml graduated concentrator tubes fitted with Kuderna-Danish columns, obtained from Kontes
Glass Company, Vineland, New Jersey. Samples were heated under vacuum using a 10-tube
Rotary-Evapomix evaporator obtained from Buchler Instruments, New York, New York; rubber
hose connections were replaced with Teflon.
The weight of methylene chloride extractable matter was determined by taking a small
aliquot of the concentrate (~5 percent) and evaporating the solvent on a light (~5 mg) alu-
minum weighing pan. Rapid evaporation of the methylene chloride led to collection of a
small quantity of moisture on the aluminum pan and residue. This was permitted to dissipate
by storage of the pans overnight in a desicator; the residue weight was then determined. Pans
were tared and reweigrted using a Cahn electrobalance, and aliquot weights were determined to
±2 HQ. The fraction represented by the aliquot was determined gravimetrically. That is, the
total weight of the concentrate (solution) was determined. Next, a volume was withdrawn in
a microsyringe (i.e., 25 n\). The syringe was then weighed before and after the solution was
dispensed onto the weighing pan. Using these data the total weight of the extractable matter
was calculated. Use of this procedure obviated the necessity of taking the entire methylene
chloride extract to dry ness.
Dioxane extracts were concentrated by lyophilization (freeze drying). An all glass appa-
ratus was used, and extracts were lyophilized in tared 25 ml RB flasks. Use of vacuum grease
(silicone) was minimized by utilization of Teflon joint tape.
Solvent and filter blanks were carried through the reflux and concentration procedures.
In calculating the weight percent solvent extractable, values due to such solvent and filter back-
ground were subtracted. Data concerning blanks are presented in the Results and Discussion
section. Values for weight percent methylene chloride extractable and weight percent dioxane
extractable were calculated as follows:
IATTELLE — COLUMBUS
-------
B-2
,.. . , . . . (Weight of extracted matter, corrected) „ inn
Weight percent solvent extractable = (Weight of total particulate) X 1°°
In order to demonstrate that the dioxane does not extract significant quantities of inorganic
salt from the filters, 100-ml portions of dioxane were stirred with ammonium sulfate, ammonium
nitrate, lead nitrate, sodium carbonate, and sodium chloride. The suspensions were filtered and
the dioxane filtrates were lyophilized in tared flasks. Salt residues were only barely visible and
did not exceed 0.3 mg.
Determination of Weight Percent CHN
Determinations of weight percent C, H, and N were conducted using a Perkin-Elmer Model
240 elemental analyzer with gas purification accessory. The instrument performs automated
Pregel-Dumas determinations. Values shown for weight percent oxygen [0] were calculated by
difference from the CHN data. Approximately 1 mg of sample was used for each determination.
Infrared Spectroscopy
Infrared spectra were obtained using a Perkin-Elmer Model 521 grating infrared spectro-
photometer. Spectra were obtained of thin films of sample on a micro sodium chloride plate
(13 mm x 4 mm). Typically ~ 1.5 mg of extractable matter was used. Films were deposited
from concentrated extract solutions, after which the last traces of solvent were removed under
vacuum. Typically samples were maintained in vacuo for 1 hour before spectra were obtained.
It was demonstrated that extended maintenance of the sample in vacuo (i.e., 3 hr) had the effect
of uniformly reducing the intensities of the IR bands. Thus, it was concluded that removal
of solvent traces under vacuum does not fractionate the sample and is an acceptable procedure
in this application.
Instrument operating conditions included a slit program setting of 1000 units to produce
a varying spectral slit width of 2-6 cm"1. The gain setting was adjusted to give a 1 percent
overshoot for a 10 percent pen deflection. The scan rate setting gave a rate of 6 cm'Vsec
above 2000 cm'1 and 3 cm'Vsec below 2000 cm'1. The attenuation was 3 sec full scale. The
total instrument was purged with dry nitrogen to eliminate any lack of compensation between
the two beams due to water vapor. Amplifier balance was set for essentially zero drift with
both beams blocked.
In practically all cases film thicknesses were adjusted so that the intensity of all measured
bands (with the exception of that for CH stretching at 2920 cm'1) fell in the desired range of
30-60 percent transmittance. In some cases it was not possible to obtain spectra having all
bands in the desired intensity range. Thus spectra were obtained to give the maximum number
of key bands in this desired range.
Reduction of spectroscopic results to numerical form was performed by calculating relative
peak intensities for specified absorptions. Values of relative peak intensity were obtained by
calculating the ratio of the optical density for the specified absorption to that observed for the
CH stretching vibration at 2920 cm'1; i.e.,
BATTELLE — COLUMBUS
-------
B-3
Nuclear Magnetic Resonance Spectroscopy
Nuclear magnetic resonance (NMR) spectroscopy was performed using a Varian Associates
HA 60-IL spectrometer (60 MHz) or a JOEL PS-100 spectrometer (100 MHz). Both instruments
were equipped to perform Fourier-Transform (FT) spectroscopy. Instrument control, data
acquisition, and reduction in the Fourier-transform mode were accomplished using a Digilab
digital computer interfaced with the spectrometers.
Spectra were obtained using the unfractionated methylene chloride extractable matter. A
portion of concentrated extract was taken containing ~ 1.5 mg of residue. The solution was taken
to dryness in a small conical tube under a gentle stream of dry nitrogen. After the sample was
no longer fluid, the tarry residue was maintained under the nitrogen stream for an additional 15-20
min. The sample was then dissolved in 40-50 ^tl of CDCI3 (Merck, silver-leaf stabilized) and trans-
ferred to a 1.8-mm ID capillary tube. The capillary tube was sealed and placed in a standard
5-mm OD x 5-inch NMR tube. Teflon spacers were used to maintain the capillary tube concentric
with the larger tube.
When samples were run on the Varian spectrometer 100 n\ of hexafluorobenzene was placed
in the outer tube annulus as a source of fluorine resonance for field stabilization. When the
JOEL instrument was employed, CDCI3 served as the field stabilization source. Tetramethylsilane
was used as the internal standard.
All spectra were obtained in the FT mode. Only the 1H-| nucleus was observed. Typical
operational parameters were a 1 /isec pulse width (where a 90-degree flip angle required 20 psec)
with a pulse-to-pulse delay time of 1.5 seconds. The bandwidth was 2000 Hz with 1096 data
points collected per pulse, to give a resolution of 0.98 Hz. typically, computer-time-averaging of
500 pulses was conducted for each sample. The spectral data were integrated both digitally and
in the analogue mode, and the integrated values checked to within ±2 percent. From the inte-
grated values, aromatic/aliphatic ratios were obtained. These were calculated on the basis of
integrated resonances for methyl and methylene protons and integrated resonances for aromatic
protons; i.e.,
Aromatic/Aliphatic Ratio = <|"tegrated resonancesiaromatic protons)
(Integrated resonances:methyl + methylene protons).
Sample Fractionation
The methylene chloride extractable matter was fractionated as noted in the scheme below.
Methylene Chloride Extractable Matter
Water-Soluble Fraction Water-Insoluble Fraction
Acid Neutral Basic
Fraction Fraction Fraction
BATTELLE — COLUMBUS
-------
B-4
Fractionation into water-soluble and water-insoluble components was accomplished by extract-
ing the methylene chloride solution (~3 ml) four times with 2-ml portions of distilled water. All
extractions were conducted in conical centrifuge tubes agitated using a Vortex mixer. This proce-
dure permits convenient centrifugation of emulsions prior to separation of phases and minimizes
sample losses. Water was removed by lyophilization, and the weight of the water-soluble fraction
was determined.
The water-insoluble material remaining in solution was next extracted four times with 2-ml
portions of 2N aqueous sodium hydroxide and twice with 2-ml portions of distilled water. The
sodium hydroxide and water washes were combined. The methylene chloride solution was then
extracted three times with 2-ml portions of 2N hydrochloric acid and twice with 2-ml portions
of distilled water. The acid and water washes were combined. Additional water washes were
made until neutral; these were discarded. Material remaining in methylene chloride solution after
extraction with both aqueous sodium hydroxide and hydrochloric acid is defined as the water-
insoluble neutral fraction. Although not conducted during this program, it is recommended that
the above acid and base aqueous extracts be backwashed with methylene chloride before neutral-
ization to prevent any slight carry-over of neutral products to the acid and/or basic fractions.
The sodium hydroxide extract (containing organic-acid salts) was brought to pH 0.1, and the
free acid was extracted into methylene chloride using a continuous liquid/liquid extractor for
96 hours. Similarly, the hydrochloric acid extract (containing organic-base salts) was brought to
pH 13, and the free base was extracted into methylene chloride, again using a continuous liquid/
liquid extractor for 96 hours. The liquid/liquid extractor employed was designed for use with
solvents denser than water. It is similar to the commercially available Hershberg/Wolfe extrac-
tion apparatus except that the apparatus employed is of smaller capacity (i.e., 70 ml methylene
chloride and 20 ml aqueous phase) and in place of a sintered glass frit, the top of the aqueous
column was agitated using a small magnetic stirring bar (i.e., the drive magnet was positioned at
the side of the apparatus).
In order to determine the acid/base/neutral distribution, the methylene chloride solutions
of these fractions were concentrated as described above, aliquot weights were determined, and
the total weight of each fraction was calculated. In preparation for alcohol determination the
methylene chloride solution of the neutral fraction was dried before the final concentration
step. Drying was accomplished by refluxing the solution over a 3-A molecular sieve contained
in the glass thimble of a small soxhlet apparatus (i.e., "micro" or Bantamware size, 30-ml flask
capacity). Drying the solution directly with anhydrous magnesium sulfate or a molecular sieve
was found to be unsatisfactory because of irreversible adsorption of organics upon the drying
agent.
Functional Group Analysis for Carbonyl
Quantitative analysis for carbonyl was conducted using the water-insoluble neutral fraction
of the methylene chloride extractable matter. The procedure employed is based on the reaction
of aldehydes and ketones with 2,4-dinitrophenylhydrazine (DNPH) to form the corresponding
colored hydrazones (A-1.A-2). The color formed is stable for several hours and the molar ab-
sorbances of the hydrazones vary so little that the method is well suited to determination of
total carbonyl in a mixture.
(A-1) Siggia. S.. Quantitative Organic Analysis. John Wiley (1962), p 124.
(A-2) Lappin, G. R.. Clark, L. C., Anal. Chem., 2^541 (1951).
BATTELLE — COLUMBUS
-------
B-5
The analyses were conducted in carbonyl-free methanol. This was prepared by refluxing
500 ml of reagent-grade methanol with 2 g of DNPH and 1 ml of concentrated hydrochloric
acid for 24 hours. The methanol was then distilled from the mixture, followed by redistillation.
Determinations of unknowns and calibration standards were conducted as follows. In a
10 ml volumetric flask were combined 1 ml of sample in carbonyl free methanol, 1 ml of
carbonyl free methanol saturated with DNPH, and one drop of concentrated hydrochloric
acid. The flask was stoppered, and heated at 50 C for 30 minutes. After cooling, the reaction
mixture was treated with 1 ml of 10 percent potassium hydroxide in 20 percent aqueous
methanol (carbonyl free). (In the procedures cited above (A-1,A-2)/ the addition of 5 ml of
potassium hydroxide solution is called for. In the current work, however, it was observed that
use of more than 1 ml leads to formation of a precipitate.) Finally, the volume of the mixture
was then made up to 10 ml with carbonyl-free methanol. The optical density of the sample
at 480 nm was determined with reference to a blank reaction mixture which incorporated 1 ml
of carbonyl free methanol in place of the sample.
Calibration experiments were carried out using 5 x 10"2 molar solutions of n-heptaldehyde
and 2-heptanone; between 1 and 25 n\ of these solutions were added to 1 ml of carbonyl-free
methyl alcohol. A linear calibration over the range 10"? to 10"6 moles of carbonyl against
optical density was thus obtained.
The concentrated methylene chloride extracts used in this determination generally have a
concentration of 2 mg/ml. Typically, 250 //I of such solutions contained carbonyl in the range
of 10~6 - 10~7 moles. Analysis of distilled-in-glass methylene chloride (obtained from Burdick
and Jackson, Muskegon Michigan) indicated that the solvent is free of appreciable concentra-
tions of carbonyl contaminants. Finally, it should be pointed out that solutions should not
exceed 10~3 molar in carbonyl in order to prevent precipitation of hydrazones.
Functional Group Analysis for Alcohol
Quantitative analysis for alcohol was conducted using the water-insoluble neutral fraction
of the methylene chloride extractable matter. The sample was thoroughly dried before analyses,
as described above. The procedure employed is based upon the reaction of alcohols with lithium
aluminum hydride to liberate hydrogen. It was adapted from a large-scale procedure in which
about 10"2 moles of sample is required, and the evolved hydrogen is collected in a gas burette
(A-3,A-4) |n tne modified procedure 10"^ moles of sample may be conveniently analyzed,
and the evolved hydrogen is measured gas chromatographically.
The apparatus employed consists of a 0.2-ml reaction vessel situated along a capillary gas
loop. The vessel is equipped with a screw cap and Teflon-lined injection septum. The apparatus
is plumbed to permit gas chromatographic carrier gas to be diverted through the gas loop.
Evolved hydrogen is then swept onto the chromatography column and is quantitated using a
thermal conductivity detector.
Typically, the analyses were conducted as follows. A saturated solution (~ 10 percent
w/w) of lithium aluminum hydride was prepared in sodium-dried tetrahydrofuran, and was
stored in a vessel equipped with a Teflon-lined septum cap. 10 n\ of reagent was withdrawn
(A-3) Siggia, S., Quantitative Organic Analysis, John Wiley (1963). p 8.
(A-4) Gaylord, N. G., Reduction with Complex Metal Hydrides, Interscience (1956).
BATTELLE — COLUMBUS
-------
B-6
by syringe and injected into the reaction vessel. The solution was repeatedly degassed, by diverting
the chromatograph carrier gas through the loop, until hydrogen was no longer present in the injec-
tion, or until a very low constant value was achieved. Commercial dry nitrogen was used as
the carrier gas, and was further dried by passing through a dry-ice trap.
The test sample was injected into the reaction vessel, and after one minute the evolved
hydrogen was swept into the chromatographic column. Longer reaction times and mechanical
agitation were without effect. A 1-ft x 2-mm ID silica gel column was used at room temperature.
The thermal conductivity detector employed was obtained from a Varian Autoprep chromato-
graph. Calibration experiments were carried out by injecting 1-10 ^l of 0.10 percent or 1.0
percent solutions of cyclohexanol in dry tetrahydrofuran. Typically four replicate analyses
were conducted for each calibration standard or unknown. Reproducible results and linear
calibrations were observed over the range of 10"^ - 10"^ moles of alcohol per injection. Typically,
a 1-2 mg sample of fractionated particulate extract contained 10~6 moles of alcohol. Thus, the
sensitivity of the method is entirely adequate. If necessary, additional sensitivity could be
achieved by using a thermal conductivity detector of lower thermal capacity.
IATTELLE — COLUMBU
-------
APPENDIX C
INFRARED SPECTRA OF ORGANIC PARTICULATE
BATTELLE — COLUMBUS
-------
2.5
WAVELENGTH (MICRONS)
5 678
9 10
15 20
TRANSMITTANCE (PERCENT)
|> 8 fe g § §
?=
\
4000
•• — i —
i
1
\ '
V
— 1—
i
_^
- -
.
1
1 - •
-
"ft
V
.., —
0 ./I*
s \
,'
•
60
4
2
1
3500 .3000
— K.
iN
i
—
1 • 1 ,
n
. :
2500
;
i
2000
\
1
JH
t
_j
i
1
-
1800
s
i
i
i
"11
/
/
V
— f—
r?>
rf_
1
it
i
\ ^
V
,
*—*-*
i
-• •"•
/
•
i
/
t
a_
1
1
P "
/
2
i
\ I
P
u
1
*^S
s
1
\
\
\
3
\
,
r!^^l
•Mra
Oq
— r •?*• 1
i
METHYLE'NE CHLORIDI
EXTRACTABLES
Sample Collected July 21
Columbus, Ohio
1 1 1 1_ 1 , 1 , 1 1 , 1 , 1 1 1 1 1 u
1600 1400 1200 1000 800 600 400
$
—
FREQUENCY (CM1
100
WAVELENGTH (MICRONS)
5 67
8 9 10 12 15 20 30 40
DIOXANE
EXTRACTABLES
Sample Collected July 21
Columbus, Ohio
4000
3500
3000
2500
2000 1800 1600 1400
FREQUENCY (CM1)
1200 1000 800 600 400 200
-------
WAVELENGTH (MICRONS)
TRANSMITTANCE (PERCENT)
-o
... I
1
-
1 1 1 ••
•
|
L
II D 1
~~^'
r
\
00
T
• i i
-1 i
T :
V
5 6
1 . i
1 \
tt-H
T
; fc
— — "1
: ;*
,.f
/
nr_±t:
; \J -
•
i : :
1
o 1
1 1
1
. 1 1
—
*
.
I
I
7 89
r
)
—
i.^\
V
v. :
—
-•-
1
V
p
— i—
—
- - -
10
i
i
-! — •
— --
12 15 20 30 4
. . 1 , . 1 .. .1. .,1, ..1 1 . 1 . 1 . 1. .... .... l-Lf-^
r
1
g
*
;
1
1
'
— _-
"•">
I
I
r
p
0
-
IV
<
~^s>
~^
s
I
\
i
\
\
\
\
\
i
j
i
>
> .
g —
IETHYLENE CHLORIDE
EXTRACT ABLES
Sample Collected July 26
Columbus, Ohio
0
— ^
•
—
4000 3500 3000 2500 2000 1800 1600 1400
FREQUENCY (CM1)
1200 1000 800 600
400 200
O
NJ
WAVELENGTH (MICRONS)
2
»
50
X)
10
JO
0
40
j
r- •
-^v
3 4
\
\
s
\
_^S
*~~<
. —
j
L/J
p
L*-'"
'
1C
5
••—f —
A J
V
\i
- •
_ . 1
•i
,
- - 1 •
I. .
.
00 3500 3000 2500
-
-
2000
t •
"""N
6 78
"T
ti
\
\\
:\l
j
; '
j
1800
E
/
j**
£T5
T\
\
1
:
=
,
1
—
0
w*^>
5
55
*
9
w
\/
1600 1400 1200
10 12 15 20 3Q 40
i . . i i > i • .1 I • 14 •
/
i/
r,/
•
-H"*1
/ 1
| ,
t
J
2
1
/
1
i
~\
'
V
\
I
\
\
\
i
i
i
1
1 .
1
DIOXANE
EXTRACTABLES
Sample Collected July 26
Columbus. Ohio
I •
1000 800 600 400 21
10
FREQUENCY (CM1)
-------
WAVELENGTH (MICRONS)
6 78
100
METHYLENE CHLORIDE
EXTRACTABLES
Sample Collected Aug. 11
New York City
4000
3500
3000
2500
2000
1800 1600 1400
FREQUENCY (CM1)
1200 1000 800 600 400 200
2.5
WAVELENGTH (MICRONS)
6 78
9 10 12 15 20
30 40
.
i— w^.
•N
\
N
X
v
v
\
™
1
1
•""^
—^ — 1
-
•S~*
-
, up . ,
'" *~^*~^
\S'^^'-
4
to
i /
.
•! • :
»
" - -^
1
V
^*V
\
\
ii
V F
\ Vr
- f
.
! i
v i
I '
00 3500 3000 2500
[
«
' I _J i
2000
i
TJ
/
^
{
4
4
1
i
*V-
\
,v
K
!
0
. ,
:-
— x
"^— v
V
^^v
*
v
-y-
f-f
i
?
i
1800 1600 1400 1200
i
^£~*
i
,3
i
1000
1
S3
u
s~
1 .
p
\
L_
X,
\
\
1
i
1
|
4
j
i
B
|
|
DIOXANE
EXTRACTABLES
Sample Collected Aug. 1 1
New York City
—
—
—
800 600 400 2
100
80
20
FREQUENCY (CM1)
n
CO
-------
WAVELENGTH (MICRONS)
6 7
30 40
6
4000
3500
.3000
2500
2000
1800 1600 1400
FREQUENCY (CM')
200
O
2.5
P
IU
WAVELENGTH (MICRONS)
6 789
10 12
15
20 30 40
Ml
BO
40
20
0
40
. .
i — '
i
•»v-
.
1 !
K t ' ~^\
\ r" \
\w v^
j ftfl
.
. : .
.
: :
i
i ' i
• I ' !
9 1
^
/*•"
f
"
*
t
•
i\(M-
P
D
-
U^r— «
\ .
• ; -
- \
-
N-
*• — -\
\
- —
—
a
X^^ 1
/
1
J
.
s~
Ju
|
J
3
,
Sar
DO 3500 3000 2500 2000 1800 1600 1400 1200 1000 800
s
J^
DIOX
EXTRAC
nple Code
New Yo
Al
T;
Ct€
rk
J
1
j
p
p
yiE
VBLES
id Aug. 23
City
1 1 ' 1 ' 1
:
600 400 2(1
FREQUENCY (CM1)
-------
2.5
100
WAVELENGTH (MICRONS)
8 9 10 12 15 20 30 40
METHYLENE CHLORIDE _
EXTRACT ABLES
Sample Collected Nov. 10-13
Pomona, California
4000 3500 3000 2500 2000 1800 1600 1400
FREQUENCY (CM1)
1200 1000 800
600
400 200
o
tn
2.5
8
WAVELENGTH (MICRONS)
5 6 789
2000 1800 1600 1400
FREQUENCY (CM1)
-------
2.5
WAVEIENGTH (MICRONS)
6 78
100
9 10 12 15 20 30 40
M
.. . T
2000 1800 1600 1400 1200 1000
4000
3500
METHYLENE CHLORIDE
EXTRACTABLES
Sample Collected
September 21, 1973
Rubidoux, California
3000
FREQUENCY (CM1)
O
WAVELENGTH (MICRONS)
2
!
"~~pn
Of
8i ...
y
3
aj
?
<
P 20
40
5
30
"S,
\
r
\
t
i
3500
3
-S^
\
i***
i
30
I
\l
DO
(
/
1
-
*
If
3
A
4
7
4
|
L^"*
i
i
p
2500
i
._.
20
>
^
•L
•i
00
i
3
" \ ;
q
i
n
i
a
;\
V
j \ ;
V
1800
6 7 8 9 10 12 15
.so eo
1 rs^**i
Jv \ /"^
/ / -^"- F
/ / ' f j^ V \ 1 i, 1
X : I / \ • f
\ 1 1/ V \ 7**^
\ « S T 40 j
! t
i, '
j . . .
1 ' j ¥
; i
| ;
- " ' i | 1 » '
1600 1400 1200 1000 800 600
20 30 40
METHYLENE CHLORIDE
EXTRACTABLES
Sample Collected
September 21, 1973
ina, California
600 400
200
FREQUENCY (CM1)
-------
2.5
WAVELENGTH (MICRONS)
6 78
METHYLENE CHLORIDE
EXTRACTABLES
Sample Collected
November 16-17, 1973
Denver, Colorado
4000
3500
3000
2500
2000 1800 1600 1400 1200 1000
FREQUENCY (CM1)
800
6OO
400
200
-------
2.5
WAVELENGTH (MICRONS)
6 78
MX)
P80 -
20 30 40
3500
METHYLENE CHLORIDE
-4- EXTRACTABLES
Replicate A
Sample Collected
October 14-18, 1972
3000
2500
2000 1800 1600 1400 1200 1000
FREQUENCY (CM1)
800 600 400 200
2.5
100
WAVELENGTH (MICRONS)
6 78
9 10 12 15 20 30 40
4000
^ METHYLENE CHLORIDE
EXTRACTABLES
Replicate B
Sample Collected
October 14-18, 1972
3500
3000
2500
2000
1800 1600 1400 1200 1000
FREQUENCY (CM1)
800
600 400
200
O
00
-------
2.5
WAVELENGTH (MICRONS)
6 78
9 10 12 15 20 30 40
IVU
TRANSMITTANCE (PERCENT)
> 8 fe g. 8
40
1
_
_
1
00
. j
1
3500
"" 11
X
"1
\
1
1
3000
,
JT1
™
_
t
—
> _
i
i
L^_
E3 ^f
_,_
|
— T"
1 i ' : ' 1 1 ' 1
V— K^ Ar-^^
V ! * 1 / I/ «r U
! H/ •
. 1 1
1
- H
, 1
+
\! . 1 1 . j 1 . j
-T . . . i : ; j
METHYLENECH
1^4 -I— ! -
i
- 1 -
1
CAInMWIAD
» [ . *— i Primary Autom
Aerosol
• ' i 9 ' i ! i i
2500 2000 1800 1600 1400 1200 1000 800 600 400 »
FREQUENCY (CM1)
LORIDE
LES
otive
»
WAVELENGTH ;MICRONS)
2
TRANSMTTfANCE (PEftCENT)
8 fe ,S 8 8
M
0-
40)
5 3
— ;-
_
11>
r
y
V
\
N
. '
™j
]f
\
30 3900
r^
4 5
•,
i
\
II
A
i
L
5
i
I
t
3000
i^
/^
i
,
H
r*^
r '
^_
1
I
— i —
4
f
-I
i
6 7 8 9 10 12 15 20 30 40
._J | :
• , MCTUVI C
-
+ •
2500 2000
EXTR^
j/HA r-^v ij
/ -1 ^\ • SA, f\ Irradiate
m //f "^nr "
FtA! i-v-u -H/^vv \j
V
\
\
-|- • -f • • ! 4 ; \ EE
1 9 — i -* -V -+- — »
i • • \
"i 1 _ j__j_ i
' " !
i i . — | • ...... j . -. 4 .. .)
; — $ . (__ 4 . 8 _i .. . ..«
NE CHLORIDE
NOTABLES
d Automotive
erosol
1800 1600 1400 1200 1000 800 600 400 200
FREQUENCY (CM1)
------- |