EPA-650/2-74-105
MAY 1974
Environmental Protection Technology Series
V.V.V.'.V.V.V.V.*
1
!>
55
o
5SSZ
\
UJ
-------
EPA-650/2-74-105
ANALYSIS OF
THE
COMPOSITION
OF THE ATMOSPHERE
IN THE LOS ANGELES BASIN
by
Dr. John R. Kh renlrld
Walclen Research Division of Abcor, Jnc.
201 Vassar Street
Cambridge, Massachusetts 02139
Contract No. 68-02-02IH
Program Element No. A1I008
EPA Project Officer: Dr. J . J . Bululini
i
Chemistry and Physics Laboratory
National Environmental Research Center
Research Triangle Park, North Carolina 27711
Prepared for
OFFICE OF RESEARCH .AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
May 1974
-------
TABLE OF CONTENTS
Section
I
II
III
IV
V
VI
VII
VIII
Number
1
2
3
4
5
6a
6b
7
8
9
Title Page
CONCLUSIONS 1
RECOMMENDATIONS 3
INTRODUCTION 4
THEORETICAL APPROACH 9
DEVELOPMENT OF SELF CONSISTENT COEFFICIENTS FOR
GASOLINE BASED SOURCES 21
RESULTS OF MATERIAL BALANCE INVERSION 26
DIRECT STATISTICAL ANALYSIS OF ATMOSPHERIC DATA... 61
CITED REFERENCES 73
LIST OF TABLES
Caption Page
Reactivity of Hydrocarbons Grouped According to
Relative Hydrocarbon Consumption Reactivity 13
Composition of Natural Gas 20
List of Compounds used in DOLA Analysis 27
Stoichiometric Coefficients Used in the DOLA
Inversion Runs 28
Table of Residuals 31
Unnormal ized Exhaust Source Strengths 46
Normalized Source Strengths 46
Analysis of Variance Results 63
Codes Used in DATA Computerization DOLA Data 68
Coded Compounds - Scott DATA 71
-------
LIST OF FIGURES
Number Caption Page
1 Computer Output 29
2 Apparent Source Strength - Exhaust 35
3 Apparent Source Strength - Spi 11 age 36
4 Apparent Source Strength - Evaporation 37
5 Apparent Source Strength - Natural Gas 38
6 Apparent Source Strength - Methane 39
7 Apparent Source Strength - Propane 40
8 Normalized Source Strength - Exhaust 43
9 Normalized Source Strength - Spillage 44
10 Normalized Source Strength - Evaporation 45
11 Normalized Source Strength - Combined Spillage and
Evaporati on 48
12 Relative Contribution to Non-Methane Hydrocarbons-
Exhaust 49
13 Spillage - Contribution to Non-Methane Hydrocarbon 50
14 Evaporation Contribution to Non-Methane Hydro-
Carbons 51
15 Combined Spillage and Evaporation - Contribution
to Non-Methane Hydrocarbons 52
16 Components of Non-Methane Hydrocarbons - Weekdays. 53
17 Components of Non-Methane Hydrocarbons - Weekends. 54
18 Traffic Versus Time of Day 56
19 Percent Ethylene and Propylene Reacted on
Individual Days 58
20 Averaged Percent Loss of Ethylene and Propylene... 60
21 DOLA Data - Pairwise Correlation Coefficient
Matrix 67
22 Pairwise Correlation Matrix - Scott Data 70
11 /
lltiden,
-------
SECTION I
CONCLUSIONS
The principal conclusions that can be drawn from this study are as
follows:
1. The material balance model and inversion technique appear to
produce reasonable estimates of the contributions of different emission
sources to the atmosphere, and, can with improvements provide a means
of deducing source behavior from atmospheric data directly, without
application of reactive simulation dispersion models.
2. The selection of potential sources, exhaust, spillage, evapor-
ation, natural gas seepage, special methane and special propane, appears
reasonable. Excellent fits to the atmospheric data are achieved for all
24 conservative hydrocarbon species with only six independent variables.
The special methane source appears to be a consequence of natural methane
background in the air mass entering the Basin. The apparent source of
propane can not be so related to a natural or man-made source.
3. The different sources exhibit very characteristic diurnal and
daily patterns. Exhaust is higher on weekdays than weekends by almost
an order of magnitude, as would be expected in relation to traffic den-
sity patterns. The weekday source strength peaks about 0830 and falls
off more slowly than does the traffic density, indicating a build-up of
hydrocarbons in the morning when atmospheric mixing is limited. The
week-end pattern does not show a similar peaking.
Maiden l
-------
4. Estimates of loss of reactive species can be made. Estimates of
the loss of ethylene and propylene differ by the same order of magnitude as
do the reactivities of the two species in laboratory experiments. Lack of
independent data on gasoline and exhaust composition limited the reactive
loss estimates to only a few species.
5. The contribution of the auto exhaust to the non-methane hydrocarbon
content of the atmosphere is the dominant element during the peak weekday
traffic period. The other sources also appear to be significant, and are of
the same order of magnitude or even greater at nonpeak times.
6. The absolute values of the derived source strengths agree reasonably
with estimates of emissions derived from automobile density and average
emission factors.
7. The normalizing technique used to adjust data from varying times in
which all sources were referred to the natural gas apparent strength, appears
to reduce considerably the variability of the data due to atmospheric randomness,
-------
SECTION II
RECOMMENDATIONS
The following recommendations for future considerations are made:
1. In future sampling programs, data on gasoline composition should
be gathered at the same time the program is run.
2. Additional work be done to develop a better understanding of the
correlative relationships connecting the observed data and derived source
parameters with atmospheric parameters and source characteristics.
/UhUem
-------
SECTION III
INTRODUCTION
GENERAL
There is a considerable body of work concerned with elucidating the
mechanisms and processes involved in the formation of photochemical smog.
These studies may be broadly grouped into three main areas of interest:
1. Chemistry of photochemical reaction involving hydrocarbons,
2. Models for describing pollutants behavior, particularly in
this case, in the Los Angeles Basin, and
3. Source characteristics of the Basin.
The objectives of this study are primarily directed toward the third topic,
i.e., to identify and characterize the sources contributing to the measured
composition at the downtown Los Angeles and Huntington Park sampling sites,
and to determine the relative contribution of each source type. In addition,
relationships connecting the important pollutant species to the sources and
to the state of the atmosphere (meteorology) are to be sought.
One approach to develop these relationships would be to develop a
deterministic model containing a mathematical analog to the physical and
chemical process occurring. Such models have been developed for the Los
Angeles Basin (1, 2, 3). This study is not such a development. We have sought
instead to extract the statistical information contained in the compositional
data available for analysis and utilize it to derive empirical relationships.
llUaUen/
-------
SOURCES OF DATA
Detailed hydrocarbon and other pollutant gaseous species data have
been collected in Los Angeles over several years. One such set of data
was collected in the fall of 1968. The objective of this project was to
examine the composition data and concommitant meterological data in order
to develop a description of the atmosphere of the Los Angeles Basin and
of its reactive character. Diurnal patterns were to be identified.
The data were collected at two sites. One site was located in down-
town Los Angeles (DOLA), at which the EPA was responsible for data collection
and analysis. The second site was located at Huntington Park. Scott Research
Laboratories was responsible for the data collection and analysis at that
site.
The EPA mobile laboratory data include chemical analyses for about 60
chemical species on approximately 330 air samples. They include a variety
of paraffinic, olefinic, aromatic and acetylenic hydrocarbons, nitrogen
dioxide, nitric oxide, and carbon monoxide. Samples were collected over a
50-minute period, eight samples per day (5:00 a.m. - 3:00 p.m.) for 41 days
between September and November 1968. Approximately 10 additional samples
were collected along roadways at a later date (1970) to provide a basis for
estimating the composition of auto exhaust emitted into the Los Angeles
atmosphere. In the 1968 series other data were obtained at DOLA by the
Los Angeles Air Pollution Control District including analyses for carbon
monoxide, total hydrocarbon, nitric oxide and nitrogen dioxide. In addition,
ultraviolet light dosage, oxidant, temperature, total hydrocarbon, and
nitrogen oxides have been averaged for four-hour intervals between 7:00 a.m.
and 7:00 p.m. The Scott Research Laboratories has reported analyses of
approximately 103 air samples taken at the Huntington Park, California site
at various times between October and November 1968 (4). The hydrocarbon
analyses of Scott, however, were obtained by instantaneous sampling of
the atmosphere.
lUtiden,
-------
OBSERVED SOURCE BEHAVIOR
As noted, a principal set of objectives is related to the identification
of sources. Automobiles contribute most of the photochemically active species
to the Basin (5). This source alone cannot, however, account for all of the
hydrocarbon species found in atmospheric samples collected by a number of
investigators. Neligan found that the low boiling hydrocarbons could not
be accounted for by automotive exhaust alone and suggested natural gas as a
probable additional significant source (6). He also suggested contributions
by blowby and evaporative losses from carburetor and the gasoline tank.
Stephens and Burleson note further that natural gas alone cannot explain
the excessive ratio of propane to ethane and methane, and suggest that evapora-
tion of the light ends from gasoline may be the source (7). In a later study,
Gordon, et.al., note the same sort of patterns as Stephens and Burleson (8).
They also note that the ratios of exhaust hydrocarbons measured in the
atmosphere do not agree with those obtained on tests based on the California
test cycle.
Lonneman, et.al., in a study of aromatic hydrocarbons in the L.A.
atmosphere found that toluene levels were too high to be associated only
with automotive emissions and suggest solvent evaporation as a possible
source (9). Scott in their analysis of the data to be used in this study
also note that auto exhaust (based on carbon monoxide levels) is the major
source reflected in their measurements (4). They find, as in the earlier
works referred to above, that the light hydrocarbons (C-]-C3) must come from
other sources.
These studies plus reference to the source inventory of the L.A. Basin
(5) suggest that initially all the following potential sources should be
included in our analysis:
UJMen
-------
1. Automobile exhaust
2. Automobile evaporation and blowby
3. Solvent evaporation
4. Natural gas
5. Stationary combustion (NO)
6. Process emissions (CO)
Other sources of hydrocarbons such as refinery emissions and aircraft
exhaust do not appear, a priori, to be significant.
As will be shown in detail below, this analysis was based principally
on hydrocarbon emitters, so that the last two sources were not relevant.
Further solvent evaporation was not included for the following two reasons.
First, the magnitudes of solvent emission in Los Angeles, although not
negligible, is small compared to the emission associated with gasoline(B).
Secondly, solvents represent such a varied composition spectrum, that it
would have been very difficult to define a representative hydrocarbon
species spectrum. The approach is deficient in this regard, and refinements
in the future should include consideration of solvents. Six potential
source sub-groups were postulated. These were:
1. Automobile exhaust
2. Gasoline evaporation
3. Gasoline spillage
4. Natural gas seepage
5. Methane source
6. Propane source
The first four are, in essence, included in the a priori list above. It
appeared important to separate gasoline loss into two components, evaporation
and spillage reflecting the different compositions of the two. In evaporation
losses (carburetor hot soak, etc.)» the vapor would be enriched in the more
volatile species, (10) whereas spillage would result in a source resembling
the full compositional spectrum of gasoline (11).
-------
The latter two sources were added to the list during the study to treat
/hat appeared to be an anomalous behavior pattern of methane and propane
nth respect to the first four sources. Assuming a realistic methane and
)ropane composition for the first four sources, the initial analysis gen-
erated results that did not appear reasonable. The composition data in-
h'cate the existence of a quantity of methane and of propane that must arise
:rom sources other than the four included herein. It was expedient to
segregate the uncertainty as to the nature of this unknown source factor into
i separate term for propane and methane.
-------
SECTION IV
THEORETICAL APPROACH
MATERIAL BALANCE MODEL
An understanding of the behavior of the sources of hydrocarbons is,
per se, important in understanding and controlling photochemical smog.
Since photochemical smog formation is produced by a dynamic set of processes,
relative to the general dispersive mechanisms simultaneously taking place,
detailed understanding of diurnal and day-to-day patterns of behavior is
necessary in statistical analysis of observed oxidant levels and in pro-
viding realistic inputs to dynamic, reactive models.
Such dynamic models have been the subject of intensive development,
and hold great promise in being able to simulate conditions which have been
observed leading to verification of theory, and in predicting smog levels for
hypothetical cases. These models, in essence, represent a transformation of
the emissions of various hydrocarbon species, through kinetic and dispersive
processes, into ambient pollutant'concentration perceived by measurement.
Previous analyses of observed data have been carried out by applying
such models or by using fairly insensitive techniques based on ratios of
species to key or conserved components. The first approach has been un-
successful in isolating detailed source behavior, largely due to the com-
plexities of applications of reactive models (3). The second approach has
provided more general understanding to source behavior, but is limited by
the uncertainty in the measurements and by the scatter introduced by dispersive
randomness (4, 8, 12).
/Uhldeni
-------
Our approach, outlined below and applied as shown in later sections, is
an attempt to find an intermediate technique, which falls between the simple
ratio-analysis and complex simulation modeling. The basic model on which
this whole analysis is set is that the measured concentration of certain
non-reactive (conserved) hydrocarbon species can be related in a determin-
istic, integral way to the sources of these species. The relationship
takes the form of material balances, assuming that the entire basin or at
least the air parcel being sampled can be modeled as a variable volume
chemical reactor, as has been discussed by Friedlander and Seinfeld. (13)
The general form of the material balances equation is:
eQ 1-1.2....M (1)
j-l
where C.. = measured atmospheric concentration of the i hydrocarbon
species
V = the volume in which the source emissions have been mixed
Q. = emission source strength of the j source
J th
e.. . = a stoichiometric coefficient for the i species from the
1J .th
j source
Appropriate units for these quantities are introduced in the detailed
discussion below. Emission source strength is a measure of the total quantity
of hydrocarbon added to the vo1ume( being sampled, as for example, gallons of
gasoline, or kilograms of natural gas, etc. The stoichiometric coefficient
gives the fraction of a particular species in the source, for example, as
ppm or mole fraction butane in natural gas or exhaust.
The size of the characteristic volume cannot be deduced directly from
the data, so that the working formula is modified by redefining the source
strength according to the following equations:
10
1 Maiden,
-------
0=1
Qj - Qj/V (3)
Now, if the concentrations and stoichiometric factors are known, it should
be possible to invert the system of material balance equations to determine
the unknown source terms, Q!. If there are fewer conservative components
\J .
than sources (N < M), the equations cannot be solved. If the two are equal,
then a single algebraic solution exists. If there are more components than
sources (N > M), then the equations do not have a unique algebraic solution,
but can be solved by statistical techniques to find the best-fitted source
values.
There appear to be some 10 to 11 species in the Scott and DOLA field
data which have zero or very low reactivity, and another 10 to 20 of low
reactivity, so that there may be up to 30 such conservative equations.
Important sources number only six (6) and are as noted earlier (automotive
exhaust, evaporative losses, spillage losses, natural gas leakage, and
"unknown" methane and propane sources). The essential feature of the system
of material balance equations (relative to a ratioing technique) is that a
solution for the Q1. terms utilizes all of the information contained in the
25 to 30 species measurements and compresses it into a new set of only a few
variables. Because this transformation of variables is performed in a least-
square sense, the effect of individual measurement errors is minimized. Con-
sequently the new variable should be statistically more reliable than the
basic concentration values, provided the basic model is realistic.
Ulalden
-------
The derived source strength can be used to estimate the degree of
reaction for the active species. Using the derived source strengths, a
similar set of material balances would give predictions of the concentra-
tion of the reactive species assuming an absence of reactions, viz.:
J-1
where £? = unreacted concentration of reactive species.
There are some 20 to 30 compounds that would be considered as reactive species,
Thus, a comparison of the predicted concentrations with the measured values
provides a direct estimate of the photochemical reaction of each species.
The fraction, fk, represents the degree of reaction, as
fk = %6 *
* ^k
where £. = the measured concentration of the k species.
Plots of averaged (over different days) values of f^ versus time of
day should show the significance of reactivity groupings, in the sequence
illustrated below in Table 1 (14) or according to some other set (15).
Difficulty in estimating the composition of the sources, particularly
for the reactive species, has limited the analysis of reactive species to
two compounds, ethylene and propylene.
The material balance approach differs from other studies (4, 8, 12)
which attempt to follow the reaction by examining ratios of a single re-
active species to a non-reactive species. Since, in general, more than
one source contributed to the atmosphere over the time that reaction is
occurring, the ratio may lead to erroneous estimates of reaction, if the
12 Ulaldeni
-------
TABLE 1.
Group
REACTIVITY OF HYDROCARBONS,GROUPED ACCORDING
TO RELATIVE HYDROCARBON CONSUMPTION REACTIVITY
Classes of Compounds
Specific Species
O
4.
5.
17
C-i-Cr paraffins, acetylene.
Benzene
+ paraffins
Toluene, other monoalkyl'
benzenes
Ethylene
Dialkyl- and Trialkylbenzenes
1-alkenes
Internally double-bonded
alkenes
Methane, ethane, acetylene,
propane, isobutane, butane,
pentane, isopentane, cyclo-
pentane, benzene (10+1 (CO) =
11 species).
Hexane, heptane...undecane;
substituted butanes and pen-
tanes...19 species.
Toluene, ethyl benzene, o,m,
p-xylenes, propylbenzenes,
2,3 and 4-ethyltoluenes, etc.
(14 species).
Ethylene
1,3,5 a'nd 1,2,4-trimethyl-
benzenes (2 species).
Propylenes, butenes, 1-pen-
tene, cyclopentene, 1-hexene,
etc. (7 species).
Cis and trans 2-pentene,
2-methyl 2 butene, cis and
trans 3-hexene, cis and
trans 2-hexene (7 species).
13
-------
relative strengths of the sources vary over the period (as, is noted later,
appears to be the case). (4, 12) After inverting the material balances
relating the composition for each data set to the sources identified above,
further analysis is performed using suitably averaged sets of the source
strengths, not on averaged sets of the compositional data, per se, as in
Eschenroeder and Martinez (3, 12).
Prior to the analysis the basic equations (Eq. 1) were transformed to be
consistent with the units of concentration and composition. The atmospheric
composition data are expressed as ppb, the components of exhaust are expressed
as ppm, as is customary, while components of gasoline, both spillage and
evaporation, and natural gas are expressed as mole fractions. It is convenient
to consider the volume of air over a one-square meter area. The material
balance, equivalent to Equation 1, is:
r[ PL Y ,n-9l _
clv
+
1000R,, 1000R.
. c + <~
MWy cv MWg cs
16 /1000Rx){£v) x 10"6 +
MW
1000 Rng
MWng
1000Rm
MWm cm
Eng
1000R
' MWP
EP
(6)
where C = concentration in ppb
P = atmospheric pressure in atm
L = mixing height in meters
T = ambient temperature in °K
R = gas constant (87.06 x 10"6
RV Rs' Rnq' Rx5 Rm' Rp ^°"r^e_strengtl1 inJ^A^yFor gasoline
evaporative, spillage, natural gas, exhaust, methane and
propane sources, respectively.
ev' es' ena' em' eo = stolcni°metric coefficient for evaporative,
spillage natural gas, methane and propane sources respectively,
in mole fraction
14
lUhUeni
-------
e = stoichiometric coefficient in exhaust in ppm.
X
MWy, MW , MW , MW , MW , MW = molecular weight of vaporization,
spillage, external gas, exhaust methane and propane sources,
respectively.
The factor of 16 in the exhaust term represents the combined gasoline
plus combustion air at the typical air/fuel ratio of 15. Molecular weight
for the spillage and evaporative sources are essentially equal to the value
for gasoline; (a"yaTue~of 100 ha^^bejrTusecU ("Exhaust molecular weight is
—--. Q...
the value forqiatural gas is about \&.j The factor 10 on
the left hand side converts ppb to volume per volume, as does the factor
10" in the exhaust term. If we further define dimensionless source strengths,
R', by the following equation:
R1 = R x -ft- x 1010 (7)
We obtain the equations that were used in computer runs. The output of the
computer runs provided values of R', which we denote apparent source strength,
for the several sources. The fifth term on the right, that for methane, only
appears in the material balance for methane, i.e., the stoichiometric coefficient
is unity for methane and zero for all other components. The propane source
term, similarly appears only in the propane material balance. The form of the
equation (there will be one such equation for each conservative component) is:
r-cR'-f-c-R'+e- R' ' + 5 79 Y ID
*» ~ fc. ."V. t-~"»_ T •'£•..».'*•«•» a» /1. A lu
The inversion procedure used is based on the method of least squares as
is described by Whittaker and Robinson (16). The overdetermined set of
equations can be reduced to a consistent set of linear equations which minimize
the sum of weighted residual square errors. Solution of this set of equations
(termed the normal equations) for the source strengths (Q's) is a standard
problem in matrix algebra. A version of the BIOMED statistical analysis pro-
gram, available in the CDC Cybernet, was used for the numerical analysis.
15
I Maiden I
-------
The BIOMED software package consists of an extensive group of statistical
analysis computer programs originally developed at UCLA for medical research
purposes. The functions performed by these programs were well suited to the
processing requirements for analysis of the Los Angeles Basin data and could
be applied without developing alternative analytic programs. The programs
accept information directly from punched cards and can be selectively called
into execution by a simplified control card procedure.
The programs are arranged in the six main groups of which two were
employed as listed below. Not all of the separate programs were used.
Class R - Regression Analysis
1. Simple Linear Regression
2. Stepwise Regression
3. Multiple Regression with Case Combinations
4. Periodic Regression and Harmonic Analysis
5. Polynomial Regression
6. Asymptotic Regression
Class V - Variance Analysis
1. Analysis of Variance for One-Way Design
2. Analysis of Variance for Factorial Design
3. Analysis of Covariance for Factorial Design
4. Analysis of Co/variance with Multiple Covariates
5. General Linear Hypothesis
6. General Linear Hypotheses with Contrasts
7. Multiple Range Tests
UbUen,
-------
NON-REACTIVE CONSERVATIVE SPECIES
Using a seven-group scale suggested by Altshuller (14), based on
hydrocarbon consumption, and used in the recent analysis of the 1968 field
data by Eschenroeder and Martinez (12), the hydrocarbons found in the DOLA
and Scott studies can be classified as in Table 1. Carbon monoxide (CO) is
essentially non-reactive and is included in class 1. Other reactivity
groupings have been suggested. The relative reactivity within the seven
group scale, based on hydrocarbon consumption, is 0, 1, 3, 4, 8, 17 and
100 (15).
If we assume that both group 1 and 2 are non-reactive, we have some
30 equations to treat. Group 1 and 2 species do exhibit some reactivity (17,
18), but the relative amount compared to the higher groups should be small.
Further, the most significant contribution tqjjeak oxidant levels appears to
be early morning source emissions (19), so that the inversion^ of_the early
hour balance is perhaps^most important. Little reaction of the group 2
species wouljJJje expected in two to three hours. In practice some hydro-
carbons species are missing in a large number of samples at DOLA and
particularly at Huntington Park. In order to obtain a good set of statistical
data, the number of hydrocarbon species in the analysis was limited to 24 for
the DOLA data and 18 to 20 for the Scott data.
AUTOMOTIVE EXHAUST COMPOSITION
The validity and utility of the approach based on material balance
depends, in part, on the representativeness of the compositional description
of each source (the Q's in Equation 3). The selection of a proper set of
coefficients for automotive exhaust is difficult as the composition varies
considerably depending on the driving conditions and gasoline mix (6).
Discrepancies between the ratio of hydrocarbons or acetylene to nitric oxide
as measured according to the cycle and as measured in the atmosphere have been
17
llttJdeni
-------
noted (8). Simulation runs by Eschenroeder and Martinez were greatly im-
proved when the ratio of acetylene to nitric oxide was increased from a
value characteristic of the driving cycle, i.e. 0.2, to a value more in-
dicative of atmospheric levels, i.e., 0.8. (3)
Data were collected by the California Air Resources Board in a vehicular
tunnel during November, 1970 (20). The objectives of the study were to
collect freshly-emitted exhaust samples representative of the total auto-
mobile and gasoline mix in the Los Angeles Basis. It was assumed that
tunnel samples would provide a convenient and concentrated source of such
samples. These data were provided by EPA, as part of the inputs to this
program as a possible source of exhaust composition data. Our analysis
which is presented below, indicated that these data are not representative of
exhaust, andj»how evidence of considerable atmospheric dilution even in the
tunnel.
The collection of representative compositional data for automobile
exhaust and for full-spectrum gasoline proved to be a major and incompleted
effort in the project. Detailed data were not found. A self-consistent
method to generate_aggrgpriate composj^tipn parameters was developed, and is
described more full'
GASOLINE COMPOSITION
The composition of gasoline used in the basin will determine the com-
position of spillage losses and vehicle evaporative losses. Assuming^pm-
plete volatinizati5j[L^f_^p±llaje_,the atmospheric concentration will reflect
the gasoTTne full spectrj^cpjrip^sjjti^n_djrectly. In the initial analyses a
composition reported by Maynard and Sanders for a typical premium gasoline
was used (21). Premium gasoline represents about 75% of sales in the L.A.
basin (20,22). Maynard and Sanders report data on a regular grade as well
(21). Trying to typify the detailed composition of average Basin mix gasoline
was an extremely difficult task. Data were obtained on average mix composition
18
-------
based on Reid Vapor Pressure (RVP) or on overall proportions of saturates,
olefins and aromatics (22), but not in terms of detailed hydrocarbon structure.
Ultimately the search for actual composition data was abandoned in favor of a
selfeonsistent approach developed below. This represents a serious shortcoming
in the present work, but can^be^ avoided in the future by gathering samples of
gasoline jjmuVtaneously with the atmospheric sampling program. A considerable
effort was made to obtain detailed hydrocarbon data from individual oil com-
panies servicing the region, the American Petroleum Institute and several
agencies.
EVAPORATIVE LOSS COMPOSITION
Initially the evaporative loss composition was calculated using a modified
batch distillation approach, similar to that developed by a number of workers,
for example, by Koehl (10). The exact composition is a function of temperature,
degree of fullness of gasoline tank and other variables. The composition,
according to Koehl's work, does not change greatly in the operating range that
appears typical in the Los Angeles Basin. As above, this approach was also
dropped in favor of the self-consistent technique.
NATURAL GAS COMPOSITION
It was believed that the major source of natural gas components in the
atmosphere would result from seepage from the gas fields beneath the Basin,
rather than from leakage in the distribution system supplying gas to the area,
which gas contains a portion piped in from outside the Basin area. The natural
gas composition used in the inversion computations is shown in Table 2 (23).
Utideni
-------
TABLE 2.
COMPOSITION OF NATURAL GAS
Species Mol Percent
CH4 81.1%
CH 9.7
C3H8 ^ 3.5
i-C4H1Q 0.19
n-C4H]0 0.24
Inert 5.3
20
/Maiden/
-------
SECTION V
DEVELOPMENT OF SELF CONSISTENT
COEFFICIENTS FOR GASOLINE BASED SOURCES
The failure to obtain detailed, directly-measured data on gasoline
and exhaust composition was a major obstacle encountered in the study.
Such data are absolutely essential to the whole analytic approach. As
noted above a considerable effort was expended in a search for such data
but was finally abandoned in favor of an indirect method, which, although
limited in terms of the original anticipations, would, however, provide a
basis for applying the material balance technique. The 1970 tunnel data
which had been expected to provide a realistic estimate for exhaust com-
position, also provided disappointing results as noted by the following
analysis.
Perusal of certain compounds in the 1970 tunnel series indicated that
the tunnel data do not provide a reasonable measure of auto exhaust. The
ratio of i-pentane to n-pentane in the tunnel averages about 1.3, whereas
the ratio in the atmospheric data are clustered about a value of almost 2.
The differential can be partially accounted for by the higher relative vol-
atility of i^pentane, but not nearly sufficiently to explain the large
discrepancy. Other discrepancies were also noted. It is possible that
changes in the average gasoline composition over the two years between the
1968 and 1970 tests were responsible for the apparent mismatch.
WUem
-------
An analysis was made to determine if the apparent inconsistencies
were substantially due to the dilution of exhaust in the tunnel by out-
side ambient air. It was assumed the atmosphere in the tunnel contains
not only auto exhaust but also some amount of outside atmosphere, in the
ratio of exhaust-bearing tunnel air to outside air, 'P1. Values of P can
be estimated from ratios of tunnel concentrations to atmosphere concentra-
tions of individual hydrocarbon components. Inconsistent values were
found for P in this way. For example, the very low methane level in the
tunnel implied that very little outside air was drawn into the tunnel, i.e.,
P should be large. But, on the other hand, the ratio of n-pentane to acetylene
in the tunnel is not much different from that in the atmosphere indicating
that the tunnel sample was not very different from the atmosphere (i.e., P
should be considerably less than one).
After trying several values, it was found that the value of 0.5 for P
gave reasonable values for exhaust composition coefficients. Such a set of
coefficients was used to invert the material balances for a sample set of
days. The results of this inversion showed a considerably poorer fit for
the sample days than that of the set of self-consistent coefficients as
developed below. This adds weight to our conclusion that the tunnel data
are not a realistic base for estimating the composition of automobile exhaust.
The failure of those data to provide a base for the exhaust coefficients
produced a series of problems in finding a representative set as noted above.
No single detailed gasoline composition reported in the literature seems
representative, nor do the exhaust compositions reported by several authors
(6, 24, 25). . , ^
A set of coefficients can be derived from the atmospheric data, per se,
by making certain assumptions, first, that the atmospheric data can be
typified by some mean set of parameters. The calculational procedure begins
by taking the ratio of each conservative species to acetylene (assumed to_be
derived only from exhaust).
Maiden
-------
Starting with the basic material balance (Equation 8), we include only
exhaust, evaporation and spillage, as:
C = s. 72 x 10"5 R'e
v
A /V
Re
vv
Now, divide both sides by the equation for acetylene (Equation 10), (only
the exhaust term exists) to obtain Equation 11.
Ca = 5.72 x 10"5 R'ea (10)
Cl A G
. - - + - -
a a 5.72 x 10"' R'e 5.72 x 10"5 R'e
x a x a
A number of papers show that C,- and higher paraffins appear inthe
exhaust in more or less constant proportion to their^concentration in the
gasoline (6^, 24). Further, the concentration in the vaporized part (assuming
a simple equilibrium distillation model) is related to the concentration in
the whole gasoline through the volatility (10). These relationships can be
expressed by the following two equations:
ey = k es (13)
where k-j = proportionality con^h^t relating exhaust composition to
gasoline composi t i ons
k = volatility
23
/UbUenl
-------
Substituting Equation 12 and 13 into Equation 11 and rearranging we get:
r \ ^v f K R\/^i Rc i
M = S U.72 x 10'5 + -JJ- + _|_L
'a/ 5.72 x 10 M x x
(14)
Values of k^ and k were obtained from the references indicated above (10, 24).
Now if the acetylene content of the exhaust, ^the ratio of gasoline burned
(R /R ), and the ratio of gasoline spilled to gasoline burned (R /R ) are known
V A J /\
(on the average), the unknown, e , e , e.. can be determined
A S V
Spillage losses arise from liquid spill during transfer of gasoline to an
automobile tank and from residual dripping from the filling nozzle. A report
by Scott Research Laboratories on refueling losses provides a basis for
estimating R_/RV- By combining data on liquid spill and nozzle drip losses
j X
from that study the amount of gasoline spilled is 0.3% or R_/RV equals 0.003.
o A
About 6% of the gasoVmeJ)urned endsjjp as hydrocarbonsjin_the__exhaust.
Workers at the Bureau of Mines have shown that evaporative and exhaust con-
tributions are of comparable magnitudes at ambient temperature above 60°F (26).
That study indicates that exhaust and evaporative losses at an ambient tem-
perature of 70°F are in the ratio 7/3, so that the percent of gasoline evaporated
is about 2.7%, or R /R equals 0.027-
V J\
All that remains, then is to obtain an estimate of the average acetylene
*nt of exhaust.^Several approaches were use<
One is based on measurements of carbon monoxide.
i i iw y
content of exhaust.^Several approaches were used to arrive at that figure.
lUaUeni
-------
The CO content of exhaust is assumed to be 2.73% (by volume) of the
total exhaust. That value was reported by G. W. Dickinson, et.al., (27)
from a 500 car emission survey in L. A. in 1968 and corresponds reasonably
to other survey data on samples considered representative of the automobile
population in L. A. Multiplying that value by the mean of the ratio of
acetylene of CO for all the samples taken at DOLA, the mean acetylene content
of exhaust is 222 ppm. This value, which was used to obtain the coefficients
appears to be consistent with levels measured by others or inferred from their
measurements (24, 25, 27).
With these values and the mean ratio of each component to acetylene,
a self-consistent set of "e's" can be derived. The resultant values are
reported in the next section with the discussion of the inversion data.
25
Maiden
-------
SECTION VI
RESULTS OF MATERIAL BALANCE INVERSION
COMPUTER PROCEDURES
The material balance equations were inverted for each set of DOLA
samples, which consisted of hourly samples at 5-6 a.m., 6-7 a.m., 7-8 a.m.,
8-9 a.m., 10-11 a.m., 11-12 a.m., 1-2 p.m., and 2-3 p.m. Weekday samples were
taken Tuesday through Friday. Weekend samples were taken on Sunday, but
did not include the first and last hour periods. Input to the computer
was a set of measured concentrations of the conservative species. Twenty-
five compounds including carbon monoxide (CO), were considered to be non-
reactive. These compounds are listed in Table 3, with the numerical code
used in the computer analysis. Also input was a set of stoichiometric
coefficients, calculated by the self-consistent method. The set used in
the DOLA runs is given in Table 4. The output from the multiple regression
and program contains the values for the six apparent source strengths, and
a variety of statistical quantities representing the closeness of the fit
to the observations. In terms of the notation in the computer output, as
shown in Figure 1, each component is given as:
A + 5.72 x ID'5 E9>jR' + E,OJllj0 + E,,^, + 6.Z5E12JRfe
6'25E13,jRi3 + 3-125E14,jR14
26
IllkUenl
-------
TABLE 3.
LIST OF^COMPOUNDS USED IN DOLA ANALYSIS
COMPOUND CODE COMPOUND NAME
2 Ethane
4 Propane
5 Acetylene
6 Isobutane
7 n-butane
11 Isopentane
16 n-pentane
22 Cyclopentane/2-methylpentane
23 3-methylpentane
25 Hexane
27 2,2-Dimethylpentane
28 2,4-Dimethylpentane
34 3,3-Dimethylpentane
32 Cyclohexane
33 2-Methylhexane
34 2-3 Dimethylpentane
35 3-methylhexane
36 l-cis-3 Dimethylcyclopentane
37 2,2,4-Trimethylpentane
38 n-Heptane
39 Methylcyclohexane
41 n-Nonane
47 n-Decane
57 Methane
58 Carbon Monoxide
27
IlllaUenl
-------
TABLE 4
STOICHIOMETRIC COEFFICIENTS*
USED IN THE
DOLA INVERSION RUNS
Compound
No.
2
4
5
6
7
11
16
22
23
25
27
28
31
32
33
34
35
36
37
38
39
41
47
57
58
Exhaust
28.3
222.
7.8
43.5
60.4
32.0
37.4
17.6
23.9
1.3
22.5
.34
4.3
9.9
6.3
12.3
6.1
18.9
19.6
17.0
11.9
19.7
250.0
27300.
Evaporation
.0029
.0597
.281
.194
.0806
.0437
.0186
.0212
.0008
.0134
.0001
.0025
.0004
.0027
.0049 '
.0024
.0058
.0062
.0049
.0005
.0003
Natural
Spillage Gas Methane Propane
.097
.0002 .0350 1.0
.0092 .0019
.0596 .0024
.0944
.0500
.0585
.0275
.0374
.0020
.0351
.0005
.0067
.0155
.0098
.019
.0095
.0295
.0305
.0265
.0186
.0308
.811 1.0
Exhaust coefficients in ppm (volume) others in mole fraction
28
IllhUenl
-------
SELtCTluN NO.
9- 4
SAMKL.E .SHE i«.Pl .53069
1U .O3l
-------
where A = intercept of the multiple regression equation
E = stoichiometric coefficient
R' = apparent source strength
Subscripts 9, 10, 11, 12, 13, 14 = exhaust, spillage, evaporation,
natural gas, methane and propane respectively.
The apparent source strengths appear in Figure 1, under the column
labeled (Reg)ression (Coeff)icients. In the example shown, the exhaust
apparent source strength is 3540. The intercept is shown as the A value,
here equal to -0.19871. An estimate of the standard error between the
measured concentration and the predicted value is given. The value in this
case was 1.864 ppb, which means on the average, the material balances will
estimate the concentration of each species to within ±1.86 ppb. A second
part of the output for each run, is a list of predicted concentrations,
shown in Table 5. The compounds are listed in the same order as shown in
Table 3. Note that methane (observation number 24) and propane (observation
number 2) are fit exactly, in accordance with the inclusion of special sources
for these two compounds. The example shown here does not include CO. Dup-
licate runs including CO were made for most samples. The only formal dif-
ference would be the addition of another line to the table of residuals
(Table 5). As noted below, the summaries of the individual runs were made
on the runs without CO.
GENERAL RESULTS
In some runs, negative values of source strength for evaporation (and in
a few cases for spillage) were found. This represents a physically impossible
case and indicates that:
1. The method is inapplicable and the linear material balances do
not represent the atmosphere, or
30
-------
TABLE 5.
TABLE
OBStKVATION
1
2
3
4
5
6
7
B
9
10
U
12
1 J
1*
15
16
17
1»
19 - - —
20
21
22
23
24
Y VALUE
3u.6oooo
44.«UOOO
] 6 • 4 () 0 0 0
60.47000
51.V2QOO
- -?h»44000
16.02000
«.77ooo
10-7SOOO
.*0000
v.iooon
.33000
i.7rtcoo
.} « 3 o 0 U 0
2 . 6 } .0 o 0
*. 1*000
1 .2bOOO
-- -4. y^ooo
b.bbOOO
*.U?000
2-V7000
4.05000
33QO.OOOOO
Y ESTATE
.VB«(iC9J3 "'
J8.6JOOO
4.4. /«>ob6
14.2iol7
«)3.ol3('b
b0.7lfi?7
o.Ab .
18.1JO&4
• B.0'"59<>
10'3M07
.29674
b.6.)34a
-.U-^lSV'
1.4*479
3.JS79B
2.U7H.»3
4.16511
1.^7456
- " 6.239B8
6.5-OSJO
b.U3o*6
3*3i]bb
5.b3695
3300.00000
RESIDUAL
.020(33
.00000
•0403*
2.15933
-2.b4375
1.20173
" 5«4V114 • •"'
•3. 11064
.67^04
•3^893
.Ib32b
,49fib2
. 4 1329
.29521
.00202
.53177
-.02511
-.71456
"*2«149HH " '"•
-.84=30
-1.46QB8
••54166
-1.4bA95
-.00000
TES] OF EXTREME RESIDUALS
HATIO Of RANGES FOR THE SMALLEST KESIDUAL. . . . .
HATlO OK RANGES FOR TH£- LAkGEiT KESIUUAL. . . .^y
CWIItCAL VAuUE OF THE RiTlO AT ALPHA = .10 ...-
-?564-
.367
31
lUtiden,
-------
2. The natural inherent uncertainty in the measurements and the
input parameters can result formally in an apparent physical paradox.
This will be seen most in the smallest source as a consequence of the
best-fit algorithm, or
3. The actual reaction and disappearance of some of the paraffins,
which are assumed to be conserved in the analysis, lead to negative values.
We believe that the latter two explanations for the observed results, not
the first, are more likely the basis. The fit, using the self-consistent
set of coefficients is, in general, quite good. The 24 individual hydro-
carbon species are fit on the average to a few ppb over concentrations
ranging from almost zero to several hundred ppb.
The most significant influence on the results of the material balance
inversion procedure is thewide spread encountered in the input data. Of
the 25 so-called non-reactive compounds including CO, all but three are
generally present in the atmosphere at levels of under 100 ppb. Ethane,
however, usually appears somewhere between 100 and 200 ppb and CO and methane
are usually measured at around 3000 ppb or higher. Since the inversion pro-
cedure uses a least-square linear regression analysis to fit these data,
it tends to give undue weight to fitting those data points which are furthest
from the mean values. As a result, these three compounds (methane, ethane,
and CO) can be expected to dominate the fit. In the case of methane, the
addition of an independent source of methane frees this compound from in-
fluencing the regression fit. Thus, with methane free, ethane and CO (CO
in particular) become the dominant compounds in fitting the data.
32
I Maiden I
-------
This excessive weighting of two compounds is very significant. In
the case of CO, the result is that, since CO appears in auto exhaust
alone, the exhaust component will be fit almost entirely by the measured
CO levels; other exhaust components will have negligible effect on the
exhaust fit. In the case of ethane, a similar process happens with respect
to the natural gas source. With methane floating, the fit for this source is
dominated by the ethane levels.
Now with the exhaust and natural gas fit determined, there will exist
a fixed residual level for all remaining compounds:
ri =ci -ei,9R9-£i,12R12 <16>
where r.. is the residual level of compound i (ppb)
c. is the measured concentration of compound i (ppb)
ei 9R9' ei 12^12 are t'ie source composition and source strength
coefficients for exhaust and natural gas, respectively.
These residual levels represent the set of data against which gasoline
evaporation and gasoline spillage sources are fit. Since no particular
compounds dominate the fit to this set of data, the evaporation and spillage
sources are estimated using the information available in the entire set of
residual data.
The importance of this understanding of the regression procedure is that
it points out the sensitivity of this model to choices of source composition
coefficients. If CO is included, the precision and accuracy of the exhaust
prediction is directly equivalent to that 6f the CO source coefficient, while
the precision and accuracy of the natural gas prediction is almost equivalent
to that of the ethane source coefficient. Precision of the spillage and
evaporation prediction, however, are more complex. These predictions are
based on source coefficients for all hydrocarbon species as well as on the
Utideni
-------
residuals. Since these residuals in turn are based on the exhaust and
natural gas predictions, it follows that the accuracy of prediction of
spillage and evaporation sources is limited by the accuracy of the CO and
ethane coefficients. Additional lack of precision results from the im-
precision in the .other hydrocarbon coefficients.
As an alternative procedure, CO can be omitted so that the exhaust is
fitted more by the hydrocarbon components. In this case, the next most
dominant species, acetylene, will have the largest effect, but far less
than CO. This should improve the overall method. The results presented
hereunder were obtained with CO excluded. From the runs with CO included
it was noted that the computed regression coefficients (source strengths)
were comparable, indicating that the CO/acetylene ratio used to characterize
exhaust composition is probably quite realistic.
Inversion of the Scott data from Huntington Park was considerably more
difficult than that for the DOLA data. Grab samples were taken at random
times so that no consistent temporal patterns could be developed. Samples
for the light hydrocarbons (Cg - C,) were taken separately and at different
times than those for the heavier hydrocarbons (Cg - CIQ). Attempts were made
to associate contemporary or almost contemporary samples in the inversion
procedure. The results were disappointing. The data showed considerable
scatter and appearance of negative source strengths. As a result, only the
DOLA data have been completely analyzed, and are discussed in the sections
below.
APPARENT SOURCE STRENGTH RESULTS
The results of the inversion computations for the DOLA data are dis-
played in Figures 2 through 7. The^averaged values^farL.w£eJLJJavs and
weekends are shown as functions of^ime_o,tLdax._ .^There-Mere 28 weekdays^
and 10 weekend (Sunday) samples. The samples were taken at 0530, 0630,
~ 1430 on weekdays. On Sundays, the first
34
Maiden
-------
co
tn
1000..
Figure 2. Apparent Source Strength-Exhaust
X Weekdays
• Weekends
-------
CO
cr>
Weekdays
Weekends
10
Figure 3. Apparent Source Strength-Spillage
11 12
Time
-------
CO
90 J.
80
7°1
60 4
50
40
30
20
10
x Weekday
• Weekends
8
10 11
Time
Figure 4. Apparent Source Strength-Evaporation
-------
GO
00
"12
Weekdays
Weekends
10
-+•
6
8
10 11 12
Time
Figure 5. Apparent Source Strength-Natural Gas
-------
CO
«JD
R13
340-
330-
320-
310.
300-
290-
280-
270.-
260-
250
240
230
• Weekend
X Weekday
4-
8
10
11 12
Time
Figure 6. Apparent Source Strength-Methane
-------
• Weekend
X Weekday
R14
51
4-L
-1 1-
56789
10 11
Time
_, 1 h-
12 1 23
Figure 7. Apparent Source Strength-Propane
-------
and the last samples were omitted. The.Jjjdividual apparent source strength
valties are qjn^e_scattered_aboujbthe meajri. The coefficient of yaHations,
i.e., sta^n^rdjde^Qations jjlvided by the mean. r^s__between 50% and 100%
for the gasolines-derived sources, with the highest degree of scatter in the
morning samples. The unknown methane source (R'no) showed surprising little
" ' " • •- -— L_M . |» i, m-.. i —
scatter compared to the other sources.
If we assume that the basic emission sources, such as automobile density
or natural-gas leakage, do not change significantly from day to day, then
the variability in the apparent strength may be largely attributed to atmos-
pheric variability from hour-to-hour and day-to-day. In particular the
i
inversion layer, that so typically limits mixing in the early morning, lifts
during the day, allowing upward diffusion and a consequent apparent reduction
in source strength. Note that the actual strength should be proportional to
mixing height.
Examination of these curves shows, clearly, a larger exhaust emission
onweekdays, as would be expected. Spillage and evaporative emissions appear
to be somewhat larger on weekdays, but not greatly so. Natural gas and the
methane and propane sources do not show any significant differences between
days and weekends. The striking similarity in the weekday and weekend curves
for the unknown methane source (R'-j3)» and the consistently lower scatter
suggests a natural source. The equivalent concentration of methane corresponding
to the source is obtained by multiplying R',, by 6.25 (as shown in Equation 15)
and ranges from about 2000 ppb in the early morning to about 1500 ppb in the
afternoon. These values correspond to the range of methane in the global back-
ground, and it seems reasonable to hypothesize that this source represents the
methane present in the air entering the basin.
41
I flakier, I
-------
The downward trend during the day may be due to the replacement of
the initial (early morning) air mass which may have been largely composed
of parcels having terrestrial origin and consequently, the higher methane
background, by parcels of marine origin with lower methane content. If
this hypotheses is correct, then the large variation in observed levels of
methane could be due to the trapping of natural gas in the basin under
widely-varying mixing characteristics.
NORMALIZED SOURCE STRENGTHS
Although the average source strength curves displayed above appear
to show diurnal patterns that seem reasonable, there remains, as noted,
considerable scatter in the data. It is possible to reduce the scatter
in the data by referring the apparent source strengths (R1) to the
apparent source strength for natural gas. If one assumes that the rate
of emission Qf_natural gas into the atmosphere in the basin is constant,
unaffected by diurnal or dajMy cycles, then variations in the apparent
values, R1, as computed, should be the result of atmospheric variability.
The effect of that_vari-abjj-rfey_ should be^removed or least reduced by
dividing the_apparent_sou,rce strengths for exhaust, evaporative and
spillage by the corresponding natural gas value, prior to computing the
averages.
The results of such a normalization procedure are displayed in
Figures 8 through 10, showing the diurnal patterns for normalized source
strength, S. The degree of scatter is markedly reduced as shown in the
following set of data (Table 6) from two separate Sundays. As shown, the
derived exhaust strengths were almost an order of magnitude apart but were
reduced to approximately equal values after normalization. The atmospheric
concentrations of hydrocarbons observed on those two days were also almost
an order of magnitude apart. The weekend to weekday differences also become
more apparent. The coefficient of variation ranges from 20-50%, considerably
less than in the corresponding apparent source strengths.
42
IllhUenl
-------
90,.
.£»
CO
X Weekdays
• Weekends
Traffic Density Relative
6 7 8 9 10
Figure 8. Normalized Source Strength-Exhaust
11 12
Time
-------
2 4.
'10
1 4-
-i h
, Weekend
x Weekday
8
10
11 T. 12
Time
Figure 9. Normalized Source Strength-Spillage
-------
tn
* Weekday
• Weekend
Figure 10. Normalized Source Strength-Evaporation
-------
TABLE 6.
A. UNNORMALIZED EXHAUST SOURCE STRENGTHS
6-7
7-8
8-9
10-11
11-12
1-2
Sunday 9/29
Sunday 10/27
358
403
4920
422
5600
479
5810
479
4000
2535
B. NORMALIZED SOURCE STRENGTHS
6-7
7-8
8-9
10-11
11-12
1-2
Sunday 9/29
Sunday 10/27
16
21
19
20
20
26
25
18
15
30
46
lUhldem
-------
The value of the normalized source strength is a measure of the
quantity of gasoline emitted through the three proposed routes, exhaust,
spillage and evaporation. As noted earlier, some of the values for
evaporation or occasionally spillage regression coefficients were negative.
Thus, the separate quantities are subject to considerable uncertainty.
The sum of the two showed less scatter, and has been plotted in Figure 11.
The peak in the evaporative and spillage curve appears later in the day
than does the peak^of the exhaust curve. This may bedue to the_ggneration
of emj^sjoin^_by_carburetor_hot soalk^ after the cars have_i)_een parked in
the morning, and also may follow the general heating during the morning.
Superimposed in Figure 8 is a curve for the weekday traffic density
pattern typical of Los Angeles freeway traffic (28). The general shape
of the source strength curve fits a priori expectations. The early morning
buildup follows automotive traffic density growth. Since the source
strength represents the integrated source contributions to the air parcel
samples, the magnitude of this quantity should remain more or less constant
at the peak value until advection and inversion lifting begin to clear out
the stagnant air during mid-morning hours. For most of the days sampled,
a sharp decrease in unnormalized source strength was found to occur in the
mid-morning period, which behavior pattern would appear to be due to air
movement.
The values of the normalized source strength are proportional to the
mass of fuel consumed in the three separate emission modes. The relative
contributions of each of the three gasoline-derived modes to atmospheric
non-methane hydrocarbon are shown in Figures 12 to 15. As above a graph
of the combined contribution of evaporation and spillage is included. The
same data are presented in a cumulative percentage diagram in Figures 16 and
17. for the weekdays and weekend samples, respectively. The^ dominance of
exhaust is during the week is e^ndent in Figure 16 a1though_the sum of_
evaporative and spillage losses appear more important in the weekend pattern.
47
Maiden,
-------
-pi
00
3 -r
and 2 .
1 ..
Weekdays
Weekends
4-
10 11
Time
12
Figure 11. Normalized Source Strength-(Combined Spillage and Evaporation
-------
60-
50-
40
30 +
20--
10--
-4-
X Weekday
• Weekend
8
10
11 12
Time
Figure 12. Relative Contribution To Non-Methane Hydrocarbons - Exhaust
-------
in
o
40"
30--
% S
10
20 —
10-1-
-H-
8
10
11
12
Time
Weekday
Weekend
Figure 13. Spillage - Contribution to Non-Methane Hydrocarbon
-------
40 +
30 +
% S
11
20 +
10 +
X Weekday
• Weekend
4-
4-
8
10
11 12
Time
Figure 14. Evaporation Contribution to Non-Methane Hydrocarbons
-------
Ul
PO
aio
and
50 +
40 4-
30-4-
20
10
•t-
7
8
10 11 12
Time
^Weekday
» Weekend
Figure 15. Combined Spillage and Evaporation - Contribution to Non-Methane Hydrocarbons
-------
en
CO
100
90
80
70
60
3 50
CD
0.
Q) 40
| 3°
5 20
10
Other Sources
„•— »—
Evaporation
JL
8
9 10
Time
11
12
Figure 16. Components of Non-Methane Hydrocarbons - Weekdays
-------
CJ1
100
90
80
70
60
0)
50
+J
a 40
30
20
10
03
03
O
Other
Spillage
Exhaust
8 9 10 11 12
Time
Figure 17. Components of Non-Methane Hydrocarbons - Weekends
-------
If these results represent properly the different sources of atmos-
pheric hydrocarbons, it follows that reducing evaporative and spillage
losses is as important as reducing exhaust hydrocarbons.
1 —^
It is possible to compare the estimate of source strength to other
estimates of emissions. The actual source strength is related to the
inversion regression value (R1) by the relationship in Equation 7 above.
Substituting appropriate values for p, T, R , the equation becomes:
y
R(kg/m2) = 4.0 x 10"9 R'L (17)
where L is the mixing height in meters. During the fall, the mixing height
is less than 150 meters about half the time. The source strength of hydro-
carbons from exhaust and evaporation at 0900 is about 350, assuming that
about 5% of the hydrocarbons consumed in the vehicle appears in the exhaust.
This number is equivalent to about 540 kilograms per square mile.
The traffic density in the area near the DOLA sampling site was esti-
mated by Roberts, et.al., (28) to be approximately 200,000 vehicle-miles
per square mile per day. In urban driving, an emission factor of about 8.3
grams per mile appears typical. Using this value and assuming conditions
in 1968 were about the same, the daily emissions per square mile would be
about 1660 kilograms per mile per day.
Based on freeway traffic measurements (see Figure 18) the early
morning peak accounts for about 30% of the total traffic for the day (29),
or about 500 kilograms per square mile of the total daily output. This
value agrees with the estimate very closely, and lends credence to the
whole approach.
55
-------
10
9
f
8--
en
«o
•M
o
s_
3 O
o 3
1--
M
\.
/—\
K \
8
10 Noon 2 4
Time of Day
Harbor-Santa Monica Freeway Interchange
V
9
v
8
10
"I
M
Figure 18. Traffic Versus Time of Day
-------
LOSS OF REACTIVE COMPOUNDS
Having estimated source strengths by inverting the material balance
equations for the non-reactive compounds, it is now possible to reverse
the process for the reactive compounds and to use the computed source
strengths to estimate their concentrations assuming that no reaction took
place. The equation here is:
where the R.'s are the source strengths, e. .'s are the source composition
coefficients, and C'. is the predicted concentration of the jtn reactive
J
compound assuming no reaction took place.
The difference between C'. and the actual concentration measured for
the jth compound at that time is then an estimate of the amount of the jtn
species that has reacted. This difference, as a percentage of C1. is an
J
unbiased estimate of the percent of compound j that has reacted for a
given air parcel at a given time.
This procedure has been carried out for two of the reactive compounds
(ethyl ene and propylene) on the sample set of days. Estimates of the two
species in unreacted exhaust were drawn from the literature (6, 24). Figure 19
shows some typical daily patterns of percent propylene and percent ethyl ene
reacted. The initial morning levels are not zero, and represent the accumula-
tion of unreacted material in the air masses being sampled. As expected, the
percent reacted shows a rise from very low levels in the morning to a peak that
occurs usually after about 10:00 a.m. Since each daily cycle of measurements
is made on the composition in a changing sample of air, it is expected that
this curve will not always continue to rise with time. New emissions of un-
reacted material are accumulating at the same time that reaction is depleting
the material. The percent reacted thus reflects a balance between additional
emissions and continuing reactions. The peaks of those curves may indicate the
point in the day when increasing auto emissions balance the depletion rate due
to reaction.
lUhlden,
-------
75-
50-
25--
-I—
8 9
Date:
567
Time of Day
H 1 h-
10 11 12
10/30/68
Propylene
o
-------
The average loss curve_s_fpr the_entire _set_qf_days are plotted in
Figure 20. The curves appear to show a difference between weekday and
weekend curves^ The difference is not statistically significant r.— The
propylene ajic^ ethyl ene loss data showed a remarkably low degree of scatter
about the mean value. The coefficient of variation was generally below
Since the basic data are much more highly scattered, the narrowness
of the spread of the derived quantity supports the reasonableness of the
approach. The differBnce__in_J_oss between the two spjecies is of the same
degree as is the reactivity asjneasured in the 1 aboratory .
Unfortunately, we could not apply this technique to other reactive
species. The lack of data on composition of exhaust and gasoline precluded
reliable estimates of the stoichiometric coefficients of the higher molecular
weight species. Ethyl ene and propylene are emitted essentially only in
exhaust. Given reliable data on all of the reactive species, this technique
appears to be capable of estimating the degree to which the compounds have
been consumed.
59
jlllaldenl
-------
O)
u
O)
o.
70..
60.,
50"
40
30--
20 •-
10 --
Propylene
Ethylene
, Weekend
X Weekday
-I »-
-I 1-
-I t-
-I h
8
10
11 12
Time
Figure 20. Averaged Percent Loss of Ethylene and Propylene
-------
SECTION VII
DIRECT STATISTICAL ANALYSIS OF ATMOSPHERIC DATA
ANALYSIS OF VARIANCE OF ATMOSPHERIC DATA
An analysis of the variance in the atmospheric data was performed
initially in order to classify these data prior to the inversion analysis.
The aim of this classification was to reduce the variance in each set of
data to be analyzed by dividing the daily observations into more or less
homogeneous classes. Tentative hypotheses to be tested in the analysis
were that:
1. The automotive exhaust, the gas spillage and the evaporation
sources are about the same from weekday to weekday, but will show con-
siderable weekend-to-weekday variation.
2. Within each day there is considerable hour-to-hour variation
in the exhaust, spillage and evaporation sources.
3. The day-to-day variation in the natural gas source is insigni-
ficant but from hour-to-hour within a given day the variation may be
significant.
These hypotheses were tested as follows, using an analysis of var-
iance program in the BIOMED statistical package. The analysis of variance
was carried out on the composition data without transformation. It was
believed that the robust characteristics of the standard analysis of
various methodology would provide valid answers on such a large set of
data without transformation. On that base screening for normality was '
not done.
61
Maiden
-------
To examine auto exhaust emissions, CO was assumed to be the key
compound. CO was classified first by day of the week (for weekdays only)
and the within- and between-day variances were calculated. The results
showed that there was no significant difference between days (weekday
only) at 1% significance level test (see Table 7a).
The same CO data were then classified by weekday vs. weekend. In
this case, there were three data groups - Tuesday, Wednesday, and Sunday
observations. The day-to-day variance in this case was significant at the
1% significance level (see Table 7b).
The CO data were then classified by hour of day into eight groups.
The hour-to-hour vairation was thus shown to be significant at the 1%
significance level (see Table 7c).
To examine emissions from spillage and evaporation sources, i-pentane
was selected as the key compound. A similar series of tests were run and
the variance between weekdays was shown to be insignificant at the 1% level.
Weekday-to-weekend variance, as well as hour-to-hour variance within a given
day were shown to be significant at the 1% level (see Tables 7d, 7e, and 7f).
To examine the natural gas emission source, methane was selected as the
key compound. A similar series of tests were made and the weekday-to-weekday
as well as weekday-to-weekend variance was shown to be insignificant at the
1% significance level. The variance in methane levels from hour-to-hour
within a given day, however, was shown to be significant at the 1% level
(see Table 7g, 7h, and 7i).
The main conclusions derived from this set of analyses are, first, that
in the analysis of the automotive exhaust source, and of the spillage and
evaporation sources, data should be broken into a two-way classification -
hour-by-hour and weekday-by-weekend - in order to reduce the intrinsic vari-
ability in the data sets. In the case of the natural gas source, data need
be classified only by time of day to reduce variability.
62
Ulalden
-------
TABLE 7
ANALYSIS OF VARIANCE RESULTS
7a. Analysis of Variance on CO
Source of variance d.f.
among weekdays.
Sum of Square
Mean Square
between days
within days
4
206
"4,206 = 1.9
116.7
3169.8
29.2
15.4
7b. Analysis of Variance on CO
Source of Variance d.f.
weekday vs. weekend.
Sum of Square
Mean Square
between days
within days
2
168
"2,168 = 7.1
126.5
1506.6
63.25
9.0
7c. Analysis of Variance on CO
Source of Variance d.f.
h7,203 = 16.6
among hours.
Sum of Squares
Mean Square
between hours
within hours
7
203
1198.2
2088.2
171.2
10.3
63
UJalden
-------
7d. Analysis of Variance on i-pentane - among weekdays.
Source of Variance d.f.
between days
within days
4
213
1 05286
2556268
263?"
1 2000
4,213 = 2.2
7e. Analysis of Variance on i-pentane - weekday vs. weekend.
Source of Variance d.f. Sum of Squares Mean Square
between days
within days
2
168
r2,168 = 4.7
80283
1432961
40142
8530
7f. Analysis of Variance on i-pentane - among hours
Source of Variance d.f^ Sum of Squares
r7,268 = 8.8
7g_ Analysis of variance on methane - among weekdays,
Source of Variance d_.f. Sum of Squares
Mean Square
between hours
within hours
7
268
614140
2676628
87734
999
between days
within days
4
206
4.3
93.96
1.07
.46
4,206 = 2.36
64
UJalden
-------
7h. Analysis of variance on methane - weekday vs..weekend
Source of Variance d.f. Sum of Squares Mean Square
between days
within days
2
168
.668
78.6
..33
.47
r2,168 = .71
7i. Analysis of Variance on methane - among hours
Source of Variance d.f. Sum of Squares
Mean Squares
between hours
within hours
7
203
F7,203 = 11.04
27.1
71.2
3.9
.35
65
IllJaldenl
-------
It was assumed that the weekday-to-weekend variability in exhaust,
spillage and evaporation data was due to variability in source strengths
while the hour-to-hour variabil-ity in these data was due to both source
variations and diurnal meteorological patterns. The hour-to-hour variations
in natural gas data were assumed to result only from diurnal meteorological
patterns. Hour-to-hour variations in exhaust, spillage and evaporation
source strengths can thus be amplified by taking the ratio of these source
strengths to the natural gas source strength as discussed above.
Thus, the analysis of variance study was the basis of the treatment
of the data in the inversion process. The difference in exhaust source
strength from hour-to-hour and weekend-to-weekday was clearly evident as
shown earlier. Other source components followed the patterns predicted
according to the statistical analysis on the raw data set.
PAIRWISE CORRELATIONS
DOLA Data
The DOLA and Scott data were put through a multiple correlation
analysis. The program used for the DOLA data was limited to a 20 x 20
matrix so that three overlapping runs were required to assemble the
entire array. Incomplete records where any compound was missing were
eliminated. Out of the 304 hour/day combinations existing, about 160
have completerecords. By partitioning the whole set, larger samples could
be included in the correlation analysis. For this number ofsamples, a
siqn^i'can^c^r£e1ation^coefficient is about P-70. Figure 21 shows the
pairwise correlation matrix. The compound codes correspond to those used
in the EPA coding scheme, and are noted in Table 8.
lUtiden,
-------
LEGEND
= 0.70 - 0.79
= 0.80 - 0.89
= 0.90 - 0.99
Values for
Correlation
Coefficient
15 20 25
Compound Code
Figure 21. DOLA Data - Pairwise Correlation Coefficient Matrix
67
-------
TABLE 8
CODES USED IN DATA COMPUTERIZATION
DOLA DATA
00
1. Not Used
2. Ethane
3. Ethylene
4. Propane
5. Acetylene
6. Isobutane
7. n-Butane
8. Propylene
9. Propadiene
10. Neopentane
11. Isopentane
12. 1-Butene +
iso-Butylene
13. trans-Butene-2 +
Methyl acetylene
14. cis-Butene-2
15. Butadiene-1,3
16. n-Pentane
17. Pentene-1
18. 2-Methylbutene-l
19. trans-Pentene-2
20. cis-Pentene-2
21. 2-Methybutene-2
22. Cyclopentane +
2-Methylpentane
23. 3-Methylpentane
24. 4-Methylpentene-2
25. Hexane
26. Hexene-1
27. 2,2-Dimethylpentane
28. 2,4-Dimethylpentane
29. 2-Methylpentene
30. cis-2-Hexene
31. 3,3-Dimethylpentane
32. Cyclohexane
33. 2-Methylhexane
34. 2,3-Dimethylpentane
35. 3-Methylhexane
36. 1,cis,3-Dimethylcyclopentane
37. 2,2,4-Trimethylpentane
38. n-Heptane
39. Methylcyclohexane
40. Toluene
41. Nonane
42. Ethyl benzene
43. p-Xylene
44. m-Xylene
45. o-Xylene
46. Isopropylbenzene + Styrene
47. n-Decane
48. n-Propylbenzene
49. m+p-Ethyltoluene
50. 1,3,5-Trimethylbenzene
51. tert-Butylbenzene +
o-Ethyltoluene
52. sec-Butyl benzene +
1,2,4-Trimethylbenzene
53. Unknown
54. 1,2,3-Trimethyl benzene
55. n-Butylbenzene +
p-Diethylbenzene
56. Non-Methane HC (FIA)
57. Methane (FIA)
58. Carbon Monoxide
59. NOX
60. N0n
-------
Some correlation was expected. The principal exhaust components CO
and Cgh^, are very highly correlated. The overall pattern can be attributed
to common ancestrage for the compounds and to external factors such as
temperature, stability, or mixing depth. At this point in the analysis,
there was insufficient information to make the necessary distinction.
The apparent lack of correlation generally along the rows between
45-55 may be due to analytic problems. An examination of the raw data
shows many zero values that do not seem consistent with nearby data.
These zeros will reduce the apparent correlation. Propane (Compound No. 4) (
shows a correlation only with isobutane (Compound No. 6). This lack or
correlation with gasoline or natural gas species supports the concept of
a separate, as yet not understood, source of propane and explains the
improvement in the inversion results after such a source was added.
Scott Data
The treatment of the Scott data required a much larger effort in the
preparative stages, and a special program was written to match up data on
the various parameters, taken at somewhat different times. Atmospheric
quantities (such as temperature and humidity) and non-hydrocarbon pollutants
(CO, oxidants, total hydrocarbons) were reported as hour or half-hour
averages. Light hydrocarbons (C^ - C^) were sampled separately from the
heavier fractions (C5 - C,Q). A simple pairwise correlation routine was
included in the data preparation program. The results are presented in
Figure 22. Atmospheric parameters have been included as well as the
individual hydrocarbon species. The identification code for the variables
is given in Table 9.
The results agree generally with those for the DOLA data set. CO and
acetylene are well correlated with other exhaust gas hydrocarbons. It is
interesting to note that nitrogen dioxide is not correlated with any other
variable, but nitric oxide is, on the other hand, correlated (although
weakly) with many of the exhaust components.
69
IllUknl
-------
LEGEND
X = 0.90 - 0.99
/ = 0.80 - 0.89
• = 0.70 - 0.79
Values for
Correlation
Coefficient
Correlation Coefficient CO
7--SX7>_ • V^^/px«V ?./ 2.
S-jytS* *»/
-------
TABLE 9
CODED COMPOUNDS - SCOTT DATA
1. Carbon,Monoxide
2. Oxidants
3. Blank
4. Total Hydrocarbons
5. Wind Speed
6. Temperature
7. Relative humidity
8. Wind direction
9. UV flux
10. Methane
11. Ethane
12. Ethylene
13. Acetylene
14. Propane
15. Propylene
16. Isobutane
17. Butane
18. Butanes
19. Isopentane
20. 1-Pentene +
2-Methyl-l-Butene
21. n-Pentane
22. trans-2-Pentene
23. cis-2-Pentene
24. 2-Methyl-2-butene
25. 2,2-Dimethylbutane
26. Cyclopentene
27. Cyclopentane
28. 2,3-Dimethylbutane
29. 2-Methylpentane
30. 3-Methylpentane
31. 1-Hexene
32. n-Hexane
33. trans-3-Hexene
34. trans-2-Hexene
35. cis-3-Hexene
36. cis-2-Hexene
37. Methylcyclopentane
38. 2,4-Dimethylpentane
39. Benzene + Cyclohexane
40. 2-Methylhexane +
2f 3-Dimethylpentane
41. 3-Methylhexane
42. 2,2,4-TrimethyIpentane
43. 1-Heptene + ?
44. n-Heptane
45. Methylcyclohexane
46. 2,4-Dimethylhexane
47. 2,5-Dimethylhexane
48. 2,3,4-Trimethylpentane
49. Toluene
5O. n-Octane
51. Ethylbenzene
52. m + p-Xylene
53. o-Xylene
54. n-Nonane
55. Isopropylbenzene
56. n-Propylbenzene
57. 4-Ethy1toluene
58. 3-Ethyltoluene
59. 1,3,5-Trimethylbenzene
60. 2-Ethyltoluene
61. tert-Butylbenzene
62. 1,2,4-TrimethyIbenzene
63. n-Decane
64. sec-Butylbenzene
65. Isobutylbenzene
66. n-Butylbenzene
67. n-Undecane
68. Nitrogen Dioxide
69. Nitric Oxide
-------
The apparent lack of correlation between a number of hydrocarbon
species (Codes Nos. 19, 22-26, 33-36, 60-61) can be partly attributed
to a low number of samples for these components.
llUakkn,
72
-------
SECTION VIII
CITED REFERENCES
1. Wayne, L., Danchick, R., Weisburd, M., Kokin, A., and Stein, A.,
"Modeling Photochemical Smog on a Computer for Decision Making",
Presented at Annual Meeting, APCA, St. Louis, June 14-19 (1970).
2. Lamb, R.G., An Air Pollution Model of Los Angeles, M.S. Thesis,
UCLA (1968).
3. Eschenroeder, A., and Martinez, J., A Modeling Study to Characterize
Photochemical Atmospheric Reactions to the Los Angeles Basin Area,
General Research Corporation CR-1-152 (November 1969).
4. Scott Research Laboratories, Final Report on Phase I, Atmospheric
Reaction Studies in the Los Angeles Basin. Vol. I and II (1969).
5. Lemke, E.E., Thomas, G., and Zwiacher, W.E. (eds.), "Profile of
Air Pollution Control in Los Angeles County", County of Los Angeles
Air Pollution Control District (1969).
6. Neligan, R.E., Mader, P.P., and Chambers, L.A., "Exhaust Composition
in Relation to Fuel Chemistry", J. APCA, J]_, 178 (1961).
7. Stephens, E.R., and Burleson, F.R., "Analysis of the Atmosphere for
Light Hydrocarbons", J. APCA. 17., 147 (1967).
8. Gordon, R., Mayrsohn, H., and Ingels, R., "C2 - CB Hydrocarbons in
the Los Angeles Atmosphere", Envir. Sci. Tech.. 2^ 1117 (1968).
9. Lonneman, W.A., Bellar, T.A., and Altshuller, A., "Aromatic Hydro-
carbons in the Atmosphere of the Los Angeles Basin", Envir. Sci.
Tech.. 2., 1017 (1968). ^
10. Koehl, W.J., Jr., "Mathematical Models for Prediction of Fuel Tank j ./'
and Carbureator Evaporative Losses", SAE Paper 690506 (May 1969). \
lUlaUen,
73
-------
11. Scott Research Laboratories, "Investigation of Passenger Car Refuel-
ing Losses" (March 1970).
12. Eschenroeder, A., and Martinez, J., Analysis of Los Angeles Atmos-
pheric Reaction Data from 1968 and 1969. Draft Report (July 1970).
13. Friedlander, S.K., and Seinfeld, O.H., "A Dynamic Model of Photo-
chemical Smog", Envir. Sci. Tech., 3., 1175 (1969).
14. Altshuller, A.P., "An Evaluation of Techniques for the Determination
of the Photochemical Reactivity of Organic Emissions", J. APCA, 16,
No. 5, 257 (1966).
-7
15. Altshuller, A.P., and Bufalini, J.J., "Photochemical Aspects of Air
Pollution: A Review", Envir. Sci. Tech.. 5_, 29 (1971). I
— *
16. Whittaker, E., and Robinson, G., The Calculus of Observations, 4th
Edition, Dover Publications, New York (1967).
17. Altshuller, A.P., Kopczynski, S.L.., Wilson, D., Lonneman, W., and
Sutterfield, F.D., "Photochemical Reactivities of N-Butane and
other Paraffinic Hydrocarbons", J. APCA, 1_9, 787 (1969).
18. Kopczynski, S.L., Lonneman, W.A., Sutterfield, E.D., and Darley,
P.E., "Photochemical Reactivity of the Los Angeles Atmosphere",
Presented at 62nd Annual Meeting, APCA, New York, June 22-26, 1969.
19. Schuck, E.A., Altshuller, A.P., Barth, D.S., and Morgan, 6.B.,
"Relationship of Hydrocarbons to Oxidants in Ambient Atmospheres",
J. APCA, 20, 297 (1970).
20. Mayrsohn, H., etal_., "Second Street Tunnel Study", California Air
Resources Board (January 1971).
21. Maynard, J., and Sanders, W., "Determination of the Detailed Hydro-
carbon Composition and Potential Atmospheric Reactivity of Full-Range
Motor Casones", J. APCA, 19, 505 (1969).
22. Nelson, E.E., "Hydrocarbon Control for Los Angeles by Reeucing Gasoline
Volatility", SAE Paper 690087 (January 1969).
23. Danielson, J.A., Edit., Air Pollution Engineering Manual. EPA,
May 1973.
74
IllhUail
-------
24. Papa, L.J., "Gas Chromatography -Measuring Exhaust Hydrocarbons
Down to Parts per Billion", SAE Paper 670494 (May 1967).
25. Bonamassa, F., and Wong-Woo, H., "Composition and Reactivity of
Exhaust Hydrocarbons from 1966 California Cars" (September 1966).
26. Eccleston, B.H., Noble, B.F., and Hum, R.W., "Influence of Volatile
Fuel Components on Vehicle Emissions, U.S. Bureau of Mines, R.I.7291
(1970).
27. Dickinson, 6.W., et.al., "Tune-Up Inspection - A Continuing Emission
Control", SAE Paper 690141 (January 1969).
28. Roberts, P.J.W., Roth, P,M., and Nelson, C.M., Report 71 SAI-6,
Systems Applications, Inc., for the Air Pollution Control Office,
Environmental Protection Administration, under Contract CPA 70-148
(1971).
29. Colucci, J.M., and Begeman, C.R., APCA Paper Mo. 67-182, presented
at annual meeting, Air Pollution Control Association, Cleveland, Ohio
(June 1967).
75
/UhUenl
------- |