United States Air and Radiation EPA420-R-01-024
Environmental Protection April 2001
Agency M6.EVP009
vvEPA Evaporative Emissions of
Gross Liquid Leakers in
MOBILES
> Printed on Recycled Paper
-------
EPA420-R-01-024
April 2001
In
M6.EVP.009
Larry C. Landman
Assessment and Standards Division
Office of Transportation and Air Quality
U.S. Environmental Protection Agency
NOTICE
This technical report does not necessarily represent final EPA decisions or positions.
It is intended, to present technical analysis of issues using data which are currently available.
The purpose in the release of such reports is to facilitate the exchange of
technical information and to inform the public of technical developments which
may form the basis for a final EPA decision, position, or regulatory action.
-------
ABSTRACT
In six parallel documents (M6.EVP.001, M6.EVP.002,
M6.EVP.004, M6.EVP.005, M6.EVP.006, and M6.EVP.008), EPA noted
that a potentially significant portion of evaporative emissions
(from the in-use fleet) may be the result of a small number of
vehicles leaking liquid gasoline (rather than gasoline vapors).
This document describes EPA's approach (in MOBILE6) to estimating
both the frequency of occurrence vehicles with these significant
leaks of liquid gasoline and the magnitude of the emissions
resulting from those leaks.
This report was originally released (as a draft) in June
1999. This current version is the final revision of that draft.
This final revision incorporates suggestions and comments
received from stakeholders during the 60-day review period and
from peer reviewers.
-------
TABLE OF CONTENTS
Page Number
1.0 Introduction 1
2.0 Characterizing GLLs 1
2 .1 GLLs on RTD Test 3
2.2 GLLs on Hot Soak Test 8
2.3 GLLs on Running Loss Test 12
2.4 Summary of Magnitudes of Evaporative Emissions 15
3.0 Frequency of Occurrence of GLLs 16
3.1 First Approach to Estimating Frequency .... 17
3.1.1 On the RTD Test 17
3.1.2 On the Running Loss Test 19
3.1.3 On the Hot Soak Test 21
3.2 Second Approach to Estimating Frequency. ... 23
3.3 Selection of Approach to Estimating Frequency. 25
3.4 Overall Occurrence of GLLs
in the In-Use Fleet 28
4.0 References 30
APPENDICES
A. RTD Emissions of 11 Vehicles with Liquid Leaks ... 31
B. Hot Soak Emissions of 14 Vehicles with Liquid Leaks 32
C. Running Loss Emissions of 10 Vehicles with
Liquid Leaks 33
D. Predicted Frequency of Occurrence of
GLLs 34
E. Peer Review Comments from Sandeep Kishan 35
F. Comments from Stakeholders 43
11
-------
Evaporative Emissions of
Gross Liquid Leakers in MOBILE6
Report Number M6.EVP.009
Larry C. Landman
U.S. EPA Assessment and Standards Division
1.0 INTRODUCTION
In four parallel reports [1,2,3,4] * the US Environmental
Protection Agency (EPA) noted that for a small number of
vehicles, the primary mechanism of evaporative emissions was the
substantial** leakage of liquid gasoline (as opposed to simply
vapor leaks). In each of those reports, such vehicles were
referred to as "Gross Liquid Leakers" (GLLs). One consistent
feature of these vehicles is that their evaporative emissions far
exceed the evaporative emissions of the vehicles that were not
gross liquid leakers (non-GLLs). In this report, EPA:
• develops a set of criteria to define GLLs,
• determines the evaporative emissions produced by these
GLLs, and
• determines the occurrence (i.e., frequency) of these GLLs
as a function of vehicle age.
2.0 CHARACTERIZING "GROSS LIQUID LEAKERS" (GLLs)
The term "gross liquid leaker" (GLL) identifies vehicles
having substantial leaks of liquid gasoline, as opposed to simply
vapor leaks. But, this term has been used in different contexts
and it is, therefore, likely that some vehicles that behave as
GLLs based on one type of evaporative emissions test might not
behave as GLLs on another type of test. In this analysis, EPA
makes use of four different types of testing programs to identify
* The numbers in brackets refer to the references in Section 4 (page 31).
** Throughout this report, we use adjectives such as "substantial" and
"severe" to describe the leaks that produce GLLs. Quantitative estimates
of that type of leak can be obtained using the emissions (in grams per
hour) from Table 2-1 (page 17).
-------
-2-
those vehicles with substantial liquid leaks:
• a real-time diurnal (RTD) test [1,2] in which evaporative
emissions are measured for stabilized test vehicles that
are enclosed in a sealed housing with the temperatures
cycling over a 24-hour period to simulate the pressure-
driven evaporative HC emissions that result from the daily
increase in ambient temperature,
• a hot soak test [3] in which evaporative emissions are
measured for one hour following a driving cycle for test
vehicles that are enclosed in a sealed housing,
• a running loss test [4] in which evaporative emissions are
measured during a driving cycle for test vehicles that are
enclosed in a sealed housing, and
• a visual inspection [5].
In this report, EPA first estimates the mean evaporative
emissions of these GLLs for each type of test (Section 2), and
then estimates the likelihood of those types of leaks occurring
(Section 3).
Generally, when EPA predicts evaporative emissions (either
resting loss, diurnal, hot soak, or running loss*) these two
variables are critical:
1) the ambient temperature and
2) the fuel volatility as measured by the Reid vapor pressure
(RVP) of the test fuel.
However, for vehicles that are classified as GLLs, most (but, not
necessarily all) of the evaporative emissions are the result of
the leak of liquid gasoline. Since it is unlikely the rate of
leakage is a function of either the temperature or the fuel
volatility, EPA will treat (in MOBILE6) the evaporative emissions
of these vehicles as independent of ambient temperature and RVP.
An additional source of data was a 1998 test program
conducted for the Coordinating Research Council (CRC) in which 50
late-model year vehicles (1992 through 1997, with a mean age of
4.5 years) were tested using the hot soak, running loss, and RTD
tests.[6] However, none of those 50 vehicles had detected liquid
leaks. Thus, the results from these tests were not used in the
analyses in Section 2. The observation that no GLLs were
identified among this sample of 50 vehicles will be considered in
the analysis in Section 3.
MOBILES will not consider GLLs in its estimates of evaporative emissions
from crankcase losses or refueling. The methodology for estimating these
emissions has not changed from that in MOBILES.
-------
-3-
2.1 GLLs on the RTD Test
The category of vehicles identified as GLLs was first
discussed in a report dealing with evaporative emissions during
resting losses and diurnals. [1] In that report, the term "gross
liquid leaker" was used to refer to vehicles which had resting
loss emissions of at least 2.0 grams per hour. Those analyses
were performed on 119 vehicles tested in various EPA programs
plus 151 vehicles tested for the Coordinating Research Council
(CRC).
The analyses in that report were based on tests in which the
ambient temperature cycled over 24 hours to simulate (in real-
time) a full day's temperature pattern. The results of those
real-time diurnal (RTD) tests were used to estimate both resting
loss and diurnal emissions. In that analysis, the diurnal
emissions were calculated by subtracting the resting loss
emissions from the total RTD test results.
Since the 151 vehicles in the CRC program were randomly
recruited (within each of three model year ranges), EPA will use
that random sample to estimate the means of the resting loss and
diurnal emissions of vehicles that had liquid leaks of gasoline.
The mechanics who inspected the test vehicles identified 32 of
those vehicles as having evidence of some fuel leakage (from damp
hoses and connectors to visible leaks).
Since our intention is to only estimate the mean of the
emissions of the vehicles having only substantial leaks (i.e.,
GLLs), we first limited our sample to vehicles:
1.) whose resting loss emissions (i.e., the mean emissions
during the last six hours of the 24-hour RTD test) were at
least 0.25 grams per hour and
2.) whose total RTD emissions were at least 30 grams per day.
These limitations produced a set of vehicles whose gasoline leaks
had an observable effect on the evaporative emissions (even if
that effect was not sufficient to create a GLL). Eleven such
vehicles were found among the 32 having identified liquid leaks.
The emissions from those 11 vehicles are given in Appendix A. It
is important to note that while all of these vehicles leaked
liquid gasoline, less than half of them were eventually
classified as GLLs (i.e., having resting loss emissions of at
least 2.0 grams per hour). All of these 11 vehicles are
carbureted. In the absence of evidence to the contrary, EPA will
treat fuel injected and carbureted vehicles with liquid leaks the
same for the purposes of resting loss and diurnal emissions.
The usual approach that EPA has followed in estimating
emission levels is to simply calculate the mean of the sample of
applicable test results. However, the number of vehicles
-------
-4-
identified as GLLs (i.e., having resting loss emissions of at
least 2.0 grams per hour) is relatively small, and the range of
their emissions is relatively large. From a statistical
standpoint, the combination of these two conditions may lead to a
high degree of uncertainty in the calculated mean. An alternate
approach is to fit an assumed type of distribution curve to those
limited number of observations. The type of distribution that
has historically been used for emissions is the lognormal
distribution [7] (i.e., the logarithms of the emissions, rather
than the emissions themselves, are assumed to be normally
distributed). EPA will use this approach in MOBILE6.
Prior to modeling the estimated diurnal emissions, we
reexamined the data in Appendix A. Since our intent was to model
the distribution of diurnal emissions from vehicles with the
severest leaks, we dropped from the analysis the results of
vehicle number 9042 due to its relatively low diurnal emissions
(suggesting that it was not a GLL relative to its diurnal
emissions). Additionally, we assumed that if a valid estimate of
the diurnal emissions from vehicle 9129 had been obtained*, then
that estimated diurnal would have been less than the emissions
from the two highest emitting vehicles but higher than the
emissions from the remaining eight vehicles. Using these two
assumptions, we ranked the diurnal emissions and assigned a
percentile to each. The plot of those percentiles versus the
corresponding diurnal emissions is given in Figure 2-1, on the
following page. The solid line in that figure is the graph of
the cumulative distribution obtained by assuming that the
logarithms of the emissions are normally distributed. (The mean
of the logarithms of the emissions is 3.812; the corresponding
standard deviation is 1.075.) (Distributions other than the
lognormal were examined, but none came as close to approximating
the observed distribution.) We then used that lognormal
distribution to estimate the frequency associated with each
possible diurnal emission level.
In Reference [1], EPA noted that the hourly diurnal emissions from vehicle
number 9129 suggest that the leak actually developed around the tenth hour
of the test. Hence, that vehicle was a GLL for only the second half of
the RTD test. Trying to precisely estimate the emissions during the first
half of the RTD test, assuming the vehicle had been a GLL for the entire
test, is questionable. However, based on the vehicle's emissions for the
last 14 hours of the RTD, it appears that its 24-hour RTD emissions would
have fallen between the emissions of vehicles number 9054 and 9087.
-------
-5-
Figure 2-1
Cumulative Distribution of Estimated Diurnal Emissions
For Vehicles Exhibiting Liquid Fuel Leaks
With Diurnal Emissions Over 15 grams per day
100%
75%
50%
25%
0%
0 100 200 300
Diurnal Emissions (grams / day)
400
Although the lognormal distribution predicts that a small
number of vehicles would have impossibly high diurnal emissions,
EPA chose to limit the maximum emissions based on the assumption
that a truly severe leak would result in the quick repair of the
vehicle. Since one (real world) test vehicle (in our sample) had
diurnal emissions of almost 400 grams per day, EPA assumed that
the limit of the maximum emissions should be higher than that
value. EPA will use 1,000 grams per day as the maximum for the
purpose of estimating fleet averages.
The lognormal distribution also predicts that some leaking
vehicles will have diurnal emissions of close to zero. To
separate the GLLs from vehicles having only minor or moderate
leaks, we again examined the estimated diurnal emissions in
Appendix A. A visual inspection of those data indicated a
relatively large discontinuity (i.e., a break) between 24.86 and
62.64 grams per day. Based on that observation, EPA will use 25
grams per day as the minimum value. For a group of leaking
vehicles whose diurnal emissions were between 25 and 1,000 grams
per day, the lognormal distribution predicts that the mean
diurnal emissions of that group of leakers would be 104.36 grams
per day. (Doubling the maximum possible diurnal to 2,000 grams
-------
-6-
per day would result in increasing the estimated group average
only to 107.41 grams daily.)*
EPA will use 104.36 grams per day as the average full-day's
diurnal emissions from GLLs over a day for which the maximum
daily temperature is exactly 24°F above the daily low
temperature. (See report number M6.EVP.002 to use temperature
cycles with ranges other than 24°F.) Earlier versions of MOBILE
limited diurnal emissions to times when the ambient temperature
was at least 40°F. However, we suspect that, at temperatures
below 40°F, the diurnal emissions would still continue. However,
at those low temperatures, the likelihood of ozone exceedences
would be small.
Figure 2-2
Cumulative Distribution of Resting Loss Emissions
For 11 Vehicles Exhibiting Liquid Fuel Leaks
And Having Resting Loss Emissions Over 0.25 grams / hour
100%
75%
50%
25%
0%
0 5 10 15
Resting Loss Emissions (grams / hour)
20
The preceding approach was repeated (using the data in
Appendix A) for resting loss emissions. The resting loss
emissions from the 11 vehicles in Appendix A are plotted in
Figure 2-2.
The more traditional approach would have been simply to average the
diurnal emissions of the four vehicles in Appendix A having RTD emissions
of at least 100 grams with the diurnal emissions of two other leakers from
the EPA testing programs. The mean of those six diurnals is 100.29 grams
per day, which corresponds to using the lognormal distribution with the
maximum diurnal emissions set to 675 grams per day.
-------
-7-
As with the previous figure (Figure 2-1), the solid line in
Figure 2-2 is the graph of the cumulative distribution obtained
by assuming that the logarithms of the resting loss emissions are
normally distributed. (The mean of the logarithms of the resting
loss emissions is 0.841; the corresponding standard deviation is
1.528.) A visual inspection of that figure suggests that the
lognormal model does not fit the resting loss emissions of
leaking vehicles as well as it fit the diurnal emissions. In
fact, a straight line (i.e., a "uniform" distribution) is the
curve that best fits the resting loss emissions for vehicles
having at least 1.0 grams per hour (therefore, covering all
GLLs). However, it also predicts that forty percent of the
vehicles with leaks have zero resting loss emissions.
In previous analyses (see M6.EVP.001), EPA determined that
the lower bound of the resting loss emissions of the GLLs would
be 2.0 grams per hour. Since one (real world) test vehicle (in
our sample) had resting loss emissions of about 16 grams per
hour, EPA assumed that the limit of the maximum emissions should
be higher than that value. EPA will use 50 grams per hour as the
maximum for the purpose of estimating fleet averages. For a
group of leaking vehicles whose hourly resting loss emissions
were between 2.0 and 50 grams, the lognormal distribution
predicts that the mean resting loss emissions of that group of
leakers would be 9.163 grams per hour.* (Doubling the maximum
possible resting loss to 100 grams per hour would result in
increasing the estimated group average only to 10.875 grams
hourly.) The linear fit (i.e., uniform distribution) predicts
the mean of the resting losses from vehicles emitting at least
2.0 grams per hour would be 10.518 grams per hour. Thus, all of
those approaches produce similar estimates of the average hourly
resting loss emissions from GLLs.
Although the uniform distribution produces a superior
estimate of the observed data compared to the lognormal
distribution, both approaches produce similar estimates of the
mean resting loss emissions. Therefore, EPA will use the
lognormal distribution for consistency among the various
evaporative models in this report. EPA will use the estimate
based on the lognormal model (i.e., 9.16 grams per hour) as the
average hourly resting loss emissions from GLLs. Since the
mechanism responsible for the vast majority of the resting loss
emissions from these vehicles is the fuel leaking out of the
vehicle, and since this process is not dependent upon the ambient
temperature or fuel volatility, EPA had proposed (reference [1])
The more traditional approach would have been to simply average the
resting loss emissions of the five vehicles in Appendix A having resting
loss emissions of at least 2.0 grams per hour with the resting loss
emissions of two other leakers from the EPA testing programs. The mean of
those seven resting losses is 8.84 grams per hour, which corresponds to
using the lognormal distribution with the maximum hourly resting loss
emissions set to 45.2 grams per hour.
-------
-8-
considering resting loss emissions from GLLs as independent of
fuel volatility and temperature.
2.2 GLLs on the Hot Soak Test
The category of vehicles identified as GLLs based on
evaporative emissions during a hot soak, are discussed in a
report prepared for EPA by one of its contractors [8]. In that
report, the term GLLs was used to refer to "vehicles which
produce abnormally high evaporative emissions as a result of a
fuel leak and which have hot soak emissions of over 10 grams per
test." Since the hot soak test is one hour in duration, "grams
per test" is equivalent to "grams per hour" for the hot soak.
(See reference [8] to calculate hot soak emissions for time
periods less than an hour.) Since the hot soak test measures
total evaporative emissions during that hour, the results also
include resting loss emissions which must be subtracted to obtain
the (net) hot soak emissions.
In the analyses for that report, hot soak test results on
493 vehicles were used. Of those 493 vehicles, the mechanics
identified 14 as having evidence of some fuel leakage (from damp
hoses and connectors to visible leaks). Those 14 vehicles (along
with their hot soak test results) are listed in Appendix B. The
hot soak emissions of those 14 leaking vehicles ranged from 2.00
to 88.57 grams per test (averaging 22.47 grams). For the
remaining 479 vehicles that did not have liquid leaks detected,
their hot soak emissions ranged from 0.04 to 88.35 grams per test
(averaging 1.77 grams).
A quick inspection of the emissions listed in Appendix B
suggests that the port fuel injected (PFI) vehicles that have
leaks exhibit higher hot soak emissions than the carbureted
(CARB) vehicles that have leaks. Since the fuel delivery systems
in the PFI vehicles operate at a higher pressure than do the
systems in the carbureted vehicles, a hole in the fuel system of
a PFI vehicle will leak more fuel than a hole of the same size in
a carbureted vehicle.* Therefore, the observation that the PFIs
with liquid leaks have (on average) higher hot soak emissions
than the corresponding carbureted vehicles is reasonable. There
was an insufficient sample of leaking vehicles with throttle body
injection (TBI) systems to analyze. Therefore, the hot soak
emissions from this technology grouping will be estimated using a
theoretical rather than statistical approach.
Bernoulli's equation indicates that the leak rate will be proportional to
the square root of the ratio of operating pressures.
-------
-9-
In Figure 2-3 (below), we plotted the hot soak emissions (in
grams per test) of the six carbureted vehicles (from Appendix B)
versus the corresponding percentiles. The solid line in that
figure is the graph of the cumulative distribution obtained by
assuming that the logarithms of the emissions are normally
distributed. (The mean of the logarithms of the hot soak
emissions is 1.9644; the corresponding standard deviation is
0.6963.)
Figure 2-3
Cumulative Distribution of Hot Soak Emissions
For 6 Carbureted Vehicles Exhibiting Liquid Fuel Leaks
100%
g 75%
n
Q 50%
0)
3
3
O
25%
0%
5 10 15
H ot S oak E mis s ions (grams /T es t)
20
As was done in Section 2.1 with diurnal emissions, that lognormal
distribution was used to estimate the frequency associated with
each possible hot soak emission level. Although the lognormal
distribution predicts that a small number of carbureted vehicles
would have impossibly high hot soak emissions, EPA chose to limit
the maximum emissions based on the assumption that a truly severe
leak would result in the vehicle being quickly repaired. In
Appendix B, we can see that one owner tolerated a vehicle having
hot soak emissions of almost 90 grams per test. Based on that
observation, EPA will assume that, for the purpose of estimating
the mean hot soak emissions, the hot soak emissions of the GLLs
range between 10 and 300 grams per test.
Using the lognormal distribution in Figure 2-3, we can
predict the mean hot soak emissions for the GLL carbureted
-------
-10-
vehicles assuming hot soak emissions ranging between 10 and 300
grams per test. The mean hot soak emissions of that group of
leakers would be 16.9549 grams per test (or per hour). (That
average emission level was not very sensitive to the assumption
of the emissions of the highest possible leaker. Lowering the
assumed level of the highest emitting carbureted vehicle to 50
grams reduced the average only to 16.5503. Similarly, raising
the assumed level of the highest emitting vehicle to 1,000 grams
increased the average only to 16.9550.) EPA, therefore, will use
16.95 grams per test as the estimate of hot soak emissions from
GLL carbureted vehicles.
To estimate the mean of the hot soak emissions from the PFI
vehicles that had liquid leaks, we proceeded in the same fashion
that we employed for the carbureted vehicles. In Figure 2-4 (on
the following page), we plotted the hot soak emissions (in grams
per test) of the seven PFI vehicles (from Appendix B) versus the
corresponding percentiles.
The solid line in Figure 2-4 is the graph of the cumulative
distribution obtained by assuming that the logarithms of the
emissions are normally distributed. (The mean of the logarithms
of the hot soak emissions is 2.8830; the corresponding standard
deviation is 1.5822.)
Figure 2-4
Cumulative Distribution of Hot Soak Emissions
For 7 PFI Vehicles Exhibiting Liquid Fuel Leaks
100%
20 40 60 80
Hot Soak Emissions (grams /Test)
100
-------
-11-
A visual inspection of that figure suggests that the
lognormal model does not fit the hot soak emissions of leaking
PFI vehicles as well as it fit the carbureted vehicle. In fact,
a straight line (i.e., a "uniform" distribution) provides almost
as good a fit to the hot soak emissions for the six PFI vehicles
having at least 2.25 grams per test. (We are considering the
lognormal distribution to be a better fit because the sum of the
squares of the residuals is lower than for the linear fit.) EPA
will use the lognormal distribution because it is the better fit
and for consistency among the various evaporative models in this
report.
Using the lognormal distribution in Figure 2-4, we can
predict the mean hot soak emissions for the GLL PFI vehicles
assuming hot soak emissions ranging between 10 and 300 grams per
test. The mean hot soak emissions of that group of leakers would
be 57.1425 grams per test (or per hour). (That average emission
level is only slightly sensitive to the assumption of the
emissions of the highest possible leaker. Lowering the assumed
level of the highest emitting carbureted vehicle to 250 grams
reduces the average to 53.3468. Similarly, raising the assumed
level of the highest emitting vehicle to 400 grams increases the
average only to 63.0990.) The linear fit (i.e., uniform
distribution) predicts the mean of the hot soak emissions for PFI
vehicles emitting at least 10 grams per test would be 52.2481
grams per test. Thus, all of those approaches produce similar
estimates of the mean hourly resting loss emissions from GLLs.
EPA, therefore, will use 57.14 grams per test as the estimate of
hot soak emissions from GLL PFI vehicles.
Due to a lack of data (see Appendix B), we were not able to
perform a similar analysis for the TBI vehicles. This situation
was addressed in the report on hot soak emissions (M6.EVP.004),
in which the author stated:
"While there is no data on TBI liquid leakers in the
data sets, Bernoulli's equation indicates that the leak
rate for TBI systems would be about one half that for
PFI systems (the square root of the ratio of operating
pressures). Therefore, without further data, the
author suggests assuming that TBI liquid leakers might
emit approximately half the emissions of PFI systems."
EPA assumes (in MOBILE6) that the frequency of having a hole
of a given size is the same for both the TBI and PFI vehicles.
Based on that assumption, Bernoulli's equation predicts that at
each frequency in the cumulative distribution curve for PFIs
(i.e., Figure 2-4), the corresponding TBI curve would predict
only one-half the hot soak emissions. Thus, since the median
(i.e., the 50 percentile point) corresponds to a PFI vehicle with
a hot soak test of 17.868 grams, the median hot soak test result
for a TBI vehicle would be one-half of that (8.9339 grams).
Pictorially, the effect would be to maintain the distribution
-------
-12-
curve (in Figure 2-4) while changing the horizontal scale from
zero to 100 to a scale from zero to 50. That transformation is
performed in the following graph (Figure 2-5). Also, in that
figure, we plotted the single result for TBI vehicles in our data
base (from Appendix B).
Using the lognormal distribution in Figure 2-5, we can
predict the mean hot soak emissions for the GLL TBI vehicles
assuming hot soak emissions ranging between 10 and 300 grams per
test. The mean hot soak emissions of that group of leakers would
be 44.9990 grams per test. Therefore, EPA will use 45.00 grams
per test (or grams per hour) as the estimate of hot soak
emissions from GLL TBI vehicles. (It is encouraging, but not
statistically significant, that the actual test result of 8.28
from Appendix B is quite similar to the predicted median hot soak
test value of 8.9339 grams per test.)
Figure 2-5
Estimated Cumulative Distribution of Hot Soak Emissions
For TBI Vehicles Exhibiting Liquid Fuel Leaks
100%
10 20 30 40
Hot Soak Emissions (grams /Test)
50
2.3 GLLs on the Running Loss Test
In 1997, running loss tests were performed on 150 vehicles
as part of a testing program conducted for the Coordinating
Research Council (CRC). The mechanics who inspected those test
vehicles identified 40 of those vehicles as having evidence of
-------
-13-
some fuel leakage (from damp hoses and connectors to visible
liquid leaks). The running loss emissions from these vehicles
were measured over a single LA-4 driving cycle, using tank fuel
(RVP about 6.8 psi), and ambient temperature about 95 degrees
Fahrenheit. [9]
Since our intention is to estimate the mean of the emissions
of the vehicles having only substantial leaks, we first limited
our sample to leaking vehicles whose running loss emissions were
at least 5.0 grams per mile over the single LA-4 driving cycle.
(Five grams per mile appears to be a reasonable break point since
the next highest running loss emissions for a leaking vehicle was
only 3.52 grams per mile.) Ten such vehicles were found among
those 40 having identified liquid leaks. The emissions from
those 10 vehicles (reported as grams per mile, grams per test,
and grams per hour) are given in Appendix C. It is important to
note that while all of these vehicles leaked liquid gasoline, not
all of them are classified as GLLs (using the criteria developed
in this section). All of these 10 vehicles are carbureted. (Two
of the original 40 leaking vehicles were fuel injected; however,
their running loss emissions were each less than 0.4 grams per
mile.)
The approach used in the preceding sections (for diurnal,
resting loss, and hot soak) was repeated for running loss
emissions (using the data in Appendix C). The running loss
emissions from the 10 vehicles in Appendix C are plotted in
Figure 2-6. As with the previous figures, the solid line is the
graph of the cumulative distribution obtained by assuming that
the logarithms of the emissions are normally distributed. (The
mean of the logarithms of the emissions is 4.2; the corresponding
standard deviation is 0.88.)
-------
-14-
Figure 2-6
Cumulative Distribution of Running Loss Emissions
For Vehicles Exhibiting Liquid Fuel Leaks
With Running Loss Emissions Over 5 grams per mile
100%
75% - -
50% - -
25% -.
0%
0 10 20 30
Running Loss Emissions (grams / mile)
40
50
To determine the appropriate range of running loss emissions
for these GLLs, we reexamined the running loss test results on
all 150 vehicles. All of the vehicles that did not have an
identified liquid leak had running loss emissions (for the single
LA-4 cycle) of less than 4.2 grams per mile. EPA selected 7.0
grams per mile as the value that distinguished between vehicles
that have liquid leaks and those defined as GLLs.* Since one
(real world) test vehicle (in the CRC sample) had emissions on
the running loss test of about almost 43 grams per mile, EPA
assumed that the limit of the maximum emissions should be higher
than that value. EPA will use 200 grams per hour as the maximum
for the purpose of estimating fleet averages. For a group of
leaking vehicles whose running loss emissions were between 7.0
and 200 grams per mile, the lognormal distribution predicts that
the mean running loss emissions of that group of leakers would be
17.649 grams per mile. (As with the emissions on the hot soak
and diurnal tests, that average emission level was not very
sensitive to the assumption of the emissions of the highest
possible leaker. Lowering the assumed level of the highest
emitting carbureted vehicle to 90 grams/mile reduced the average
This 7.0 grams per mile test result over a 19.6 mile per hour driving
cycle is equivalent to 137.2 grams per hour (which includes resting loss
emissions).
-------
-15-
only to 17.181. Similarly, raising the assumed level of the
highest emitting vehicle to 500 grams/ mile increased the average
only to 17.696.) As previously stated, this analysis of running
loss emissions of GLLs is based solely on carbureted vehicles.
Using the logic (and Bernoulli's equation) from Section 2.2, it
could be argued that the running loss emissions from PFI GLLs
would be four times that amount. However, it does not seem
reasonable to assume such a high emissions rate based on no data.
Therefore, in the absence of evidence to the contrary, (for the
purposes of running loss emissions of GLLs) EPA will treat fuel
injected and carbureted vehicles the same.
Thus, EPA will use 17.65 grams per mile as the estimate of
the emissions from a running loss test from ALL GLLs over a
single LA-4 driving cycle. Since all of those GLLs were tested
over only that single cycle, an approach needed to be found to
estimate running loss emissions over different cycles (i.e.,
speed correction factors were needed). EPA assumed (for MOBILE6)
that the magnitude of the leaks were essentially independent of
speed. Thus, the 17.65 grams per mile (at 19.6 miles per hour)
results in a running loss (test) rate of 345.94 grams per hour
which includes resting loss emissions of 9.16 grams per hour
(from Section 2.1, page 8).
Therefore, the running loss emissions (in MOBILE6) were
obtained by subtracting the mean resting loss (hourly) emissions
from the total mean running loss (hourly) test emissions to
obtain the rate of 336.78 grams per hour.
2.4 Summary of Magnitudes of Evaporative Emissions
For the full-day diurnal emissions (based on the
temperatures cycling over a 24 degree Fahrenheit range) of GLLs,
EPA will use 104.36 grams per day. (See report number M6.EVP.002
to use other temperature cycles or to estimate hourly diurnal
emissions.)
For the resting loss emissions of all GLLs, EPA will use
9.16 grams per hour.
To estimate the result of a hot soak test on GLLs:
• EPA will use 16.95 grams per test for carbureted vehicles,
• EPA will use 45.00 grams per test for TBI vehicles, and
• determine the occurrence (i.e., frequency) of these GLLs
as a function of vehicle age.
To calculate the actual hot soak emissions per hour, the resting
loss emissions must be subtracted from the hot soak test
emissions.
-------
-16-
To estimate the result of a running loss emissions on all
GLLs, EPA will use 336.78 grams per hour. The resting loss
emissions have already been subtracted to obtain this value.
These average (mean) emissions as well as the minimum
(threshold) values are summarized in the Table 2-1 (on the
following page).
Table 2-1
Summary of Emissions from Vehicles with Gross Liquid Leaks
Type of Emissions
Hot Soak (grams per test)
• Carbureted Vehicles
• TBI Vehicles
• PFI Vehicles
Resting Loss (grams per hour)
Diurnal (grams per day)
Running Loss (grams per hour)
Emissions by Test Type
Minimum
10.0*
10.0*
10.0*
2.0
25.0
Average
16.95*
45.00*
57.14*
9.16
104.36
137.2** 336.78
* The Hot Soak test emissions (both Minimum and Average) include
resting loss emissions which must be subtracted.
** The Minimum Running Loss test emissions include resting loss
emissions which must be subtracted.
3.0 FREQUENCY OF OCCURRENCE OF "GROSS LIQUID LEAKERS"
In Section 2, the magnitude of each type of evaporative
emissions from liquid leakers was estimated independently using
lognormal distributions. Also, EPA believes the data can be
linked when estimating the frequency of the GLLs. However, due
to the lack of data on the occurrence of GLLS on the hot soak
test for vehicles over the age of 10, EPA made the following two
basic assumptions in predicting the frequency of GLLs:
1.) For each test of evaporative emissions (i.e., RTD, hot
soak, and running loss tests), the frequency of GLLs
increases as a function of only age. This model of the
frequency is based on the assumption that modern
technology vehicles will show the same tendency toward
developing these severe liquid leaks as do the older
-------
-17-
technology vehicles at the same age.* EPA modified this
assumption (in reference number [10]) for the 1996 and
newer vehicles certified to the new enhanced evaporative
standard.
2.) The vehicles classified as GLLs on the hot soak test are
the same vehicles identified as GLLs on either the running
loss or RTD tests. (That is, the set of vehicles
classified as GLLs on the hot soak test is the union of
the set of vehicles classified as GLLs on the RTD test
with the set of vehicles classified as GLLs on the running
loss test.) Therefore, the rate of GLLs as identified on
the hot soak test would be the sum of the two rates for
the RTD testing and the running loss of the two rates for
the RTD testing and the running loss testing minus the
number of double counted vehicles (i.e., the product of
those two rates assuming these two categories are
independent of each other).
Implicit in this assumption is EPA's belief that these
three tests of evaporative emissions do not identify the
same vehicles as being GLLs. For example, if there were a
leak in the fuel line of a vehicle, that leak may be
severe when the fuel system is under pressure (i.e., when
the engine is on). Thus, a running loss or a hot soak
test would identify the vehicle as a GLL, but the RTD test
might not (since the engine would be off).
EPA considered the following two different approaches to
predicting the occurrence of GLLs. (See footnote on page 22.)
3.1 First Approach to Estimate Frequency
The first approach involved two basic steps:
1.) Find two logistic growth functions that separately predict
the rate of GLLs on the RTD test and on the running loss
test, respectively.
2.) Verify that the union of those two functions approximate
the results observed on the hot soak test.
3.1.1 First Approach Estimating Frequency of GLLs on the RTD Test
In the report dealing with evaporative emissions measured
during the RTD tests (M6.EVP.001), EPA used the results from a
An alternative approach that EPA is not proposing (due to lack of data)
assumes that the modern technology vehicles exhibit a lower tendency to
leak (due to the more stringent demands imposed by the new evaporative
emissions certification procedure as well as heightened attention to
safety, such as, fuel tank protection and elimination of fuel line leaks).
This approach would result in replacing each single logistic growth
function with a family of two or more curves.
-------
-18-
test fleet of 270 vehicles (i.e., the combined EPA and CRC
samples) to estimate the occurrence of GLLs within each of the
three model year ranges used in the recruitment process (the pre-
1980, 1980-85, and 1986-95 vehicles). The estimated rate of
occurrence of the GLLs is reproduced in the following table
(Table 3-1). The large confidence intervals are the result of
the relatively small sample sizes.
Table 3-1
Frequency of Gross Liquid Leakers
Based on RTD Testing
Vehicle
Age (years)*
6.12
13.00
21.79
Sample
Size
85
50
51
Frequency
0.20%
2.00%
7.84%
Standard
Deviation
1.41%
1 .98%
3.76%
90% Confide
Lower
0.00%
0.00%
1 .65%
snce Interval
Upper
2.52%
5.26%
14.03%
* "Vehicle Age" was calculated by subtracting the model year
from the test year and then adding one-half to simulate
the rate as of January first.
In one of the parallel reports (M6.EVP.001), EPA derived a
logistic growth curve that exactly fit those three data points
(from Table 3-1). The equation of that function is given below:
Rate of Gross Liquid Leakers
Based on RTD/Resting Loss Testing
0.08902
1 + 414.613*exp[-0.3684* AGE]
The predicted occurrences of GLLs based on this equation are
given in Appendix D. The frequencies from Table 3-1 are plotted
in the following figure (Figure 3-1). Also graphed in that
figure are the 90 percent confidence intervals (as dotted lines)
from Table 3-1 and the predicted frequencies (as the solid line)
from Appendix D (or from the preceding equation).
After EPA had created the preceding equation, additional
test data were provided by CRC (project number E-41).
Specifically, a test program run during 1998 found no GLLs on the
RTD test in a sample of 50 late-model year vehicles (1992 through
1997, with a mean age of 4.5 years) . (See reference [6].) Those
results are consistent with that preceding equation.
-------
-19-
Figure 3-1
Predicted Frequency of Gross Liquid Leakers
With Observed Frequencies and 90 Percent Confidence Intervals
Based on RTD Testing
15%
10%-
o
£ 5%
0%
Observed Frequencies
•90% Confidence Interval
•Predicted Frequencies
10 20
Vehicle Age (years)
30
3.1.2 First Approach Estimating Frequency of GLLs on the Running Loss
Test
For the 150 vehicles in the CRC running loss testing
program, the occurrence of GLLs (i.e., the six vehicles in
Appendix B whose running loss emissions exceeded 7.0 grams/mile),
the occurrence of GLLs was calculated within each of the three
model year ranges used in the recruitment process (the same model
year ranges used in the RTD testing). Those estimated rates of
occurrence of the GLLs appear in the following table (Table 3-2).
The large confidence intervals are again the result of the
relatively small sample sizes.
Table 3-2
Frequency of Gross Liquid Leakers
Based on Running Loss Testing
Vehicle
Age (years)
8.84
14.24
22.48
Sample
Size
50
39
61
Frequency
2.00%
5.13%
4.92%
Standard
Deviation
1 .98%
3.53%
2.77%
90% Confide
Lower
0.00%
0.00%
0.36%
mce Interval
Upper
5.26%
10.94%
9.47%
-------
-20-
It was not possible to exactly fit the frequencies in Table 3-2
with an increasing function (since the observed frequencies seem
to drop after age 14.24 years). EPA derived a logistic growth
curve that best fit those three data points. The equation of
that function is:
Rate of Gross Liquid Leakers
Based on Running Loss Testing
0.06
120*exp[-0.4*AGE]
The predicted occurrences of GLLs based on that equation are
also given in Appendix D. The frequencies from Table 3-2 are
plotted below in Figure 3-2. Also graphed in that figure are the
90 percent confidence intervals (as dotted lines) from Table 3-2
and the predicted frequencies (as the solid line) from Appendix D
(or from the preceding equation).
Figure 3-2
Predicted Frequency of Gross Liquid Leakers
With Observed Frequencies and 90 Percent Confidence Intervals
Based on Running Loss Testing
15%
10%
Observed
Frequencies
•90% Confidence
Interval
•Predicted
Frequencies
10
20
30
Vehicle Age (years)
Again, the newly acquired data (noted at the end of Section
3.1.1) in which no GLLs were found during running loss testing in
a sample of 50 late-model year vehicles (mean age of 4.5 years)
are consistent with that preceding equation.
-------
-21-
3.1.3 First Approach Estimating Frequency of GLLs on the Hot Soak Test
To estimate the rate of occurrence of GLLs on the hot soak
test, we first referred to the second assumption on page 18,
which states that the collection of vehicles that are GLLs on the
hot soak test is the union of the collection of vehicles
identified as GLLs on the running loss test with the collection
of vehicles identified as GLLs on the RTD test. Thus, we were
able to estimate the rate of GLLs on the hot soak test based
solely on the rates of GLLs on the running loss and RTD tests.
In the last column of Appendix D, the rate of GLLs on the hot
soak was calculated by adding the two preceding columns and then
subtracting the product of those two columns. (As stated at the
beginning of Section 3.0, due to the lack of data at ages over 10
years, we were not able to use the same approach to predict GLLs
on the hot soak as we did on the other two tests.)
To test the reasonableness of the results of the above
assumption, we identified the six vehicles (in the hot soak
testing program of 300 vehicles conducted for Auto Oil) that had
hot soak test emissions in excess of 10 grams per test. In this
testing program, the test fleet was again stratified into three
model year ranges, but they were different groupings (1983-85,
1986-90, and 1991-93). This resulted in a sample of newer
vehicles than were used in the RTD or running loss testing
programs.* Those estimated rates of occurrence of the GLLs
within each of the three new model year ranges appear below in
Table 3-3. The large confidence intervals are again the result
of the relatively small sample sizes. We then compared those
observed rates (in Table 3-3) with the predicted rates in
Appendix D.
Table 3-3
Frequency of Gross Liquid Leakers
Based on Hot Soak Testing
Vehicle
Age (years)
1.98
5.55
9.38
Sample
Size
66
166
64
Frequency
1 .04%
1 .20%
6.25%
Standard
Deviation
1 .25%
0.85%
3.03%
90% Confide
Lower
0.00%
0.00%
1 .27%
snce Interval
Upper
3.10%
2.60%
1 1 .23%
The observed frequencies from Table 3-3 are plotted in Figure 3-3
(below). Also graphed in that figure are the 90 percent
Since none of the mean ages in Table 3-3 exceeded 10 years, EPA chose
approaches different from those used with the diurnal or running loss
emissions. Rather than predicting the occurrence on the hot soak test of
GLLs among older vehicles based only on data from newer vehicles, EPA
estimated those rates based on the rates of GLLs on both the RTD an
running loss tests.
-------
-22-
confidence intervals (as dotted lines) from Table 3-3 and the
predicted frequencies (as the solid line) from Appendix D. Those
predicted occurrences from Appendix D are based not on hot soak
test results, but on results of running loss tests and RTD tests.
Comparing, in Figure 3-3, the predicted rates of GLLs
occurring with the observed rates of GLLs on the hot soak test,
we observe:
• the predicted rates are all lower than the observed rates
which were based on relatively small samples, but
• the predicted rates are all within the 90 percent
confidence intervals of the observed rates (at each of the
three points).
These differences between the predicted and observed rates may
simply be the result of the small sample sizes.
Figure 3-3
Predicted Frequency of Gross Liquid Leakers
With Observed Frequencies and 90 Percent Confidence Intervals
On the Hot Soak Test
(Based on RTD and Running Loss Testing)
15%
B Observed Frequencies
• • "90% Confidence Interval
Predicted Frequencies
10 20
Vehicle Age (years)
30
Again, the newly acquired data (noted at the end of Sections
3.1.1 and 3.1.2) in which no GLLs were found during hot soak
testing in a sample of 50 late-model year vehicles (mean age of
4.5 years) are consistent with the preceding hot soak
predictions.
-------
-23-
3.2 Second Approach to Estimate Frequency
The second approach employed by EPA was to use all of the
observations (in Tables 3-1 through 3-3) to find logistic
functions that optimize (simultaneously) all of the predictions
This approach produced the following two equations:
Rate of Gross Liquid Leakers
Based on RTD/Resting Loss Testing
0.0865
1 + 55 * exp[-0.259 * AGE]
Rate of Gross Liquid Leakers
Based on Running Loss Testing
0.058
1 + 70 * exp[-0.48 * AGE]
These two equations (and their union which estimates GLLs on
hot soak tests) predict rates of occurrence that are all within
one-half of the corresponding standard deviations at each of the
nine observations (in Tables 3-1 through 3-3). We can again
graph those data (i.e., observed rates and confidence intervals)
from Tables 3-1 through 3-3, but now in figures with curves from
these new predictions (Figures 3-4 through 3-6). The only
differences between the three figures in Section 3.1 and these
new corresponding figures are the solid lines designating the
predicted frequencies.
Figure 3-4
Predicted Frequency of Gross Liquid Leakers Using Second Approach
With Observed Frequencies and 90 Percent Confidence Intervals
Based on RTD Testing
15%
10%--
Observed Frequencies
•90% Confidence Interval
•Predicted Frequencies
O
c
0)
t 5%
10
20
30
Vehicle Age (years)
-------
-24-
Figure 3-5
Predicted Frequency of Gross Liquid Leakers Using Second Approach
With Observed Frequencies and 90 Percent Confidence Intervals
Based on Running Loss Testing
15%
10%
Obs erved
Frequencies
•90% Confidence
Interval
10
20
30
Vehicle Age (years)
Figure 3-6
Predicted Frequency of Gross Liquid Leakers Using Second Approach
On the Hot Soak Test with Observed Frequencies and 90 Percent Confidence Intervals
(Based on RTD and Running Loss Testing)
15%
O
c
0)
10%
£ 5%
0%
Observed Frequencies
•90% Confidence Interval
•Predicted Frequencies
10 20
Vehicle Age (years)
30
-------
-25-
A visual inspection of these three figures (3-4 through 3-6)
indicates that this approach produces predicted rates (of the
occurrence of GLLs) that are all well within the 90 percent
confidence intervals of the observed rates (at each of the nine
points). In fact (as noted earlier in this section), all nine
predicted rates are within one-half of the corresponding standard
deviations at each of the observations.
3.3 Selection of Approach to Estimate Frequency
In choosing between these two methods (which in EPA's
opinion are the two best candidates) of predicting the frequency
of GLLs, we first observed that the greatest difference between
these two methods was in estimating the rate of GLLs on the hot
soak test. In Figure 3-7 (on the following page), we reproduced
the estimated frequency curves from Figures 3-3 and 3-6. In this
figure, the "dashed" line is the estimate produced using the
first method (i.e., from Figure 3-3 in Section 3.1.3), and the
solid line is the estimate produced using the second method
(i.e., from Figure 3-6 in Section 3.2).
Figure 3-7
Comparing Predicted Frequency of Gross Liquid Leakers
On the Hot Soak Test
15%
10% --
5%
_
Method 1
0% -P
0
10 20
Vehicle Age (years)
30
-------
-26-
A visual inspection of this figure indicates that:
• The two predicted rates are similar for vehicles at least
17 years of age or older.
• For vehicles newer than 17 years of age, the second method
predicts a substantially higher occurrence of GLLs. (For
vehicles up through the age of 10, the second method
predicts more than twice as many GLLs as does the first
method.)
To decide between these two models, EPA made use of a recent
testing program run jointly by the CRC and the American Petroleum
Institute (API) . [5] This program was specifically designed to
determine the frequency of vehicles with liquid leaks. Since
actual measurements of evaporative emissions were not performed
in this program, we cannot determine which of those vehicles
identified as having liquid leaks would have met our criteria for
GLLs.
In that API/CRC program, 1,000 vehicles were inspected for
any signs of leaks with the engine operating (during at least a
portion of the visual inspection). (This protocol was expected
to permit identification of vehicles exhibiting fuel leaks on the
RTD, hot soak, or running loss tests.) The vehicles were then
classified by the mechanic according to the severity of the
observed leaks. The visible liquid leaks were classified as
either:
• small liquid leaks (e.g., single drops) or
• larger leaks (e.g., steady flow of drops).
This classification was based on a visual inspection rather than
on the results of a test of the actual evaporative emissions.
The results of that study are summarized in the following table:
Table 3-4
Frequency of Leaking Vehicles
In API/CRC Testing Program
Model
Year
Ranqe
Pre-80s
80-85
86-91
92-98
Mean
Age
(years)
22.329
14.394
9.429
3.979
Sample
Sizes
70
155
352
423
Vehicles
with
Small
Leaks
5
10
2
0
Vehicles
with
Larger
Leaks
2
1
2
0
Total
with
Any
Leaks
7
11
4
0
90% Conf Interval
Lower
4.10%
3.70%
0.21%
0.00%
Upper
15.90%
10.49%
2.07%
0.49%
-------
-27-
The 90 percent confidence intervals in Table 3-4 are based on the
(total) number of vehicles with either small or large visible
leaks. Those vehicles which were identified as having large
visible liquid fuel leaks were almost certainly GLLs, and many of
the vehicles which were identified as having small visible liquid
fuel leaks were possibly GLLs as well. Thus, EPA considers the
upper bound of the confidence intervals as a conservative
estimate of the occurrence of the GLLs. If we reproduce Figure
3-7, and include the 90 percent confidence intervals from Table
3-4 (as dotted lines), we produce Figure 3-8.
Figure 3-8
Comparing Predicted Frequency of Gross Liquid Leakers
On the Hot Soak Test
(New Confidence Intervals from Table 3-4)
IO7o '
10% -
CO/
Method
I I '' ^-
x ^^
+ l^^,
[''
I t I
^^^_i^^^^_i_ ^^^_i^^_^^
^^l/i ^^^^
\ I / J^
\ | ^ / F~^~ — Method 1
^^^ ^^^^^^^^^^^^
kX ^ t '
Jr * \
l~2--\' I
10 20
Vehicle Age (years)
30
A visual inspection of Figure 3-8 strongly suggests the
second method for predicting the frequency of GLLs over predicts
the actual occurrence of GLLs for vehicles under the age of 13
years. (The conclusion that the second method "OVERPREDICTS" the
frequency is based on EPA's choice of basing the confidence
intervals on Table 3-4 instead of Table 3-3. That choice
reflects primarily the relatively large sample sizes in Table 3-4
compared with those in Table 3-3.)
Therefore, EPA will use the first method (Section 3.1) to
estimate the frequencies of the occurrence of GLLs on the three
types of tests for evaporative emissions. The results of that
method are given in Appendix D.
-------
-28-
3.4 Overall Occurrence of GLLs in the In-Use Fleet
The equations in Section 3.1 (or the results in Appendix D)
predict the occurrence of GLLs identified on the RTD test to
range between 0.02 to 8.55 percent by vehicle age, and for those
identified on the running loss test to range between 0.05 and
5.97 percent by vehicle age. It is reasonable to ask what is the
overall percentage of these vehicles in the entire in-use fleet.
To answer that question, we referred to another report which
provides an estimate of the national distribution by age of
light-duty vehicles (LDVs) and light-duty trucks (LDTs). (See
reference [11].) Applying the percentages from Appendix D to
those estimated vehicle counts produces Table 3-5 (on the
following page). The predicted total counts in Table 3-5 suggest
that GLLs represent approximately 1.2 to 1.6 percent of the
entire in-use fleet.
-------
-29-
Table 3-5
Predicted Occurrence of Gross Liquid Leakers
In the National In-Use Fleet of LDVs and LDTs
(as of January 1995)
Calendar
Year Minus
Model Year
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24 and older
TOTALS:
Vehicle
Counts
9,581,160
12,690,223
12,595,718
12,479,871
12,328,489
12,124,815
11,850,006
11,484,110
11,007,677
10,404,139
9,663,040
8,783,860
7,508,980
6,076,245
4,896,767
3,929,300
3,140,650
2,503,094
2,030,454
1,710,242
1,451,096
1 ,240,664
1,069,132
928,705
3,724,043
175,202,480
GLLs
Identified on:
RTD
2,052.19
3,924.61
5,621.77
8,033.14
11,433.59
16,178.24
22,702.78
31,499.85
43,050.78
57,685.81
75,350.55
95,286.08
111,677.02
121,573.75
128,727.45
131,947.97
130,511.75
124,468.78
116,862.35
110,464.75
102,385.03
93,514.33
84,580.52
76,080.74
312,764.31
2,018,378
Runninq Loss
4,750.99
9,349.56
13,760.93
20,159.55
29,321.45
42,197.16
59,817.53
83,045.60
112,104.66
145,891.73
181,302.31
213,090.08
226,678.25
219,360.63
203,502.92
181,708.71
157,112.78
132,479.39
111,810.15
96,801.22
83,696.51
72,483.90
63,008.21
55,054.76
221,641.23
2,740,130
-------
-30-
4.0 REFERENCES
1) Larry Landman, "Evaluating Resting Loss and Diurnal
Evaporative Emissions Using RTD Tests," Report numbered
M6.EVP.001, April 2001.
2) Larry Landman, "Modeling Hourly Diurnal Emissions and
Interrupted Diurnal Emissions Based on Real-Time Diurnal
Data," Report numbered M6.EVP.002, April 2001.
3) Louis Browning, "Update of Hot Soak Emissions Analysis"
prepared by Louis Browning of ARCADIS Geraghty & Miller,
Inc. for EPA, Report numbered M6.EVP.004, September 1998
4) Larry Landman, "Estimating Running Loss Evaporative
Emissions in MOBILE6," Report numbered M6.EVP.008, April
2001.
5) D. McClement, "Raw Fuel Survey in I/M Lanes", Prepared for
the American Petroleum Institute and the Coordinating
Research Council, Inc. by Automotive Testing Laboratories,
Inc., June 10, 1998.
6) D. McClement, "Real World Evaporative Testing of Late Model
In-Use Vehicles, CRC Project E-41", Prepared for the
Coordinating Research Council, Inc. by Automotive Testing
Laboratories, Inc., December 17, 1998.
7) Melvin Ingalls, "Mobile Source Exposure Estimation,"
prepared by Southwest Research Institute for EPA, EPA Report
Number EPA460/3-84-008, March 1984, Appendix A.
8) Edward L. Glover, "Hot Soak Emissions as a Function of Soak
Time," Report numbered M6.EVP.007.
9) D. McClement, "Measurement of Running Loss Emissions from
In-Use Vehicles (CRC Project E-35)", CRC Report No. 611,
Prepared for the Coordinating Research Council, Inc. by
Automotive Testing Laboratories, Inc., February 1998.
10) Larry Landman, "Modeling Diurnal and Resting Loss Emissions
from Vehicles Certified to the Enhanced Evaporative
Standards," Report numbered M6.EVP.005, April 2001.
11) Tracie R. Jackson, "Fleet Characterization Data for MOBILE6:
Development and Use of Age Distributions, Average Annual
Mileage Accumulation Rates, and Projected Vehicle Counts for
Use in MOBILE6," Report numbered M6.FLT.007.
-------
-31-
Appendix A
RTD Emissions of 11 Vehicles with Liquid Leaks
With RTD > 30 and Resting Loss > 0.25
(Arranged in Increasing Order of Estimated Resting Losses)
(ALL of the Leaking Vehicles Were Carbureted)
Vehicle
Number
9095
9037
9046
9042
9098
9148
9049
9054
9129
9087
9111
Real -Time
Diurnal
(RTD) Test
(grams / day)
32.26
33.44
33.76
30.88
45.21
47.97
181.35
316.59
181.79
478.16
777.14
Estimated
Rst Loss
(at 72°F)
(grams / hr)
0.28
0.47
0.62
0.89
0.90
1.27
4.87
10.58
10.77
14.12
16.51
Estimated
Diurnal
(grams / day)
24.85
21.47
18.21
8.83
22.91
16.63
64.55
62.64
IGNORE*
139.22
380.79
* An examination of the hourly RTD data from this
vehicle (in reference [i]) suggests that the leak
actually developed around the tenth hour of the
24-hour test. While the resting loss estimate
(based on hours 19 through 24) is most likely
valid, the estimate of diurnal emissions is
unreliable (in fact, it is negative).
Note that while all 11 of these vehicles have liquid
leaks most of them do NOT qualify as Gross Liquid
Leakers (only the five highest emitting vehicles meet
the necessary criteria).
-------
-32-
Appendix B
Hot Soak Emissions of 14 Vehicles with Liquid Leaks
(With Hot Soak Emissions At Least 2.0 grams / test)
Sorted by Fuel Delivery System
In Increasing Order of Emissions
Proqram
Auto Oil
EPA
EPA
Auto Oil
EPA
EPA
Vehicle
Number
134
177
122
79
173
97
Fuel
System
GARB
GARB
GARB
GARB
GARB
GARB
Temp
(°F)
94
95
105
92
92
110
RVP
(psi)
6.0
6.1
6.1
7.0
6.7
6.7
Hot Soak
(grams HC)
2.54
4.63
5.53
9.49
14.53
14.66
Program
EPA
Vehicle
Number
143
Fuel
System
TBI
Temp
m
94
RVP
(psi)
6.4
Hot Soak
(grams HC)
8.28
Program
Auto Oil
Auto Oil
Auto Oil
EPA
Auto Oil
EPA*
EPA*
Vehicle
Number
35
199
47
33
276
372
266
Fuel
System
PFI
PFI
PFI
PFI
PFI
PFI
PFI
Temp
(°F)
104
96
93
113
87
106
105
RVP
(psi)
6.7
6.5
6.1
6.0
6.3
9.0
9.0
Hot Soak
(grams HC)
2.00
2.26
11.56
46.95
49.39
54.18
88.57
* These two vehicles were tested using a substantially
more volatile fuel.
-------
-33-
Appendix C
Running Loss Emissions of 10 Vehicles with Liquid Leaks
(With Running Loss Emissions At Least 5.0 grams / mile)
(Arranged in Increasing Order of Estimated Resting Losses)
(ALL of the Leaking Vehicles Were Carbureted)
Vehicle
Number
35044
35125
35099
35085
35045
35071
35047
35129
35054
35091
Running
Loss HC
(grams / mile)
5.009
5.297
5.649
6.880
7.469
9.175
13.480
13.566
24.841
42.973
Running
Loss HC
(grams / LA-4)
37.47
39.44
42.17
51.18
55.79
68.84
100.19
100.72
184.96
318.90
Running
Loss HC
(grams / hour)
98.32
103.49
110.65
134.29
146.39
180.63
262.89
264.28
485.32
836.76
-------
-34-
Appendix D
Predicted Frequency of Occurrence of GLLs
Vehicle
Age
(years)
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Resting
Loss/
Diurnal
0.02%
0.03%
0.04%
0.06%
0.09%
0.13%
0.19%
0.27%
0.39%
0.55%
0.78%
1.08%
1 .49%
2.00%
2.63%
3.36%
4.15%
4.97%
5.75%
6.46%
7.05%
7.54%
7.91%
8.19%
8.40%
8.55%
Running
Loss
0.05%
0.07%
0.11%
0.16%
0.24%
0.35%
0.50%
0.72%
1.02%
1 .40%
1.88%
2.43%
3.02%
3.61%
4.16%
4.62%
5.00%
5.29%
5.51%
5.66%
5.77%
5.84%
5.89%
5.93%
5.95%
5.97%
Hot
Soak
0.07%
0.10%
0.15%
0.23%
0.33%
0.48%
0.70%
1.00%
1.41%
1.95%
2.64%
3.48%
4.46%
5.54%
6.67%
7.83%
8.95%
10.00%
10.94%
11.75%
12.42%
12.94%
13.34%
13.63%
13.85%
14.00%
-------
-35-
Appendix E
Response to Peer Review Comments from Sandeep Kishan
This report was formally peer reviewed by one peer reviewer
(Sandeep Kishan). In this appendix, comments from Sandeep Kishan
are reproduced in plain text, and EPA's responses to those
comments are interspersed in indented italics. Each of these
comments refer to page numbers in the earlier draft version
(dated June 30, 1999) that do not necessarily match the page
numbers in this final version.
************************************
This memorandum provides peer review comments on two EPA
documents: "Estimating Running Loss Evaporative Emissions in
MOBILE6," Document No. M6.EVP.008, June 28, 1999, and
"Evaporative Emissions of Gross Liquid Leakers in MOBILE6,"
Report Number M6.EVP.009, June 30, 1999. Both of these are draft
reports.
Overall, we think that the reports are good, and they present
some new data analysis techniques that are attractive. Since, in
the past, we have had to do similar data analyses and modeling
for evaporative emissions from vehicle test data, we can
appreciate many of the difficulties and data limitations you are
subject to. We hope the comments below help you with this
effort.
Document No. M6.EVP.009 (June 30, 1999)
We have the following questions, comments, and recommendations on
this draft report. For each item we give the page number and
paragraph that the comment refers to, if it is a specific
comment.
We found that the first half of the report, which estimates the
average emissions rate of gross liquid leakers, was well written
and, in addition, we thought that the technique of fitting the
sparse data to log-normal distributions was excellent. However,
in the second half of the report which estimates the frequency of
gross liquid leakers of different types in the vehicle
population, we had difficulty understanding the distinction
between the first approach and the second approach. We did
understand the development of the logistic growth curves for each
emission type. However, we think that there is no reason to
average this data, which causes loss of important information,
before building the logistic models. More defensible
relationships could easily be built using the individual car data
rather than averages of data.
-------
-36-
1. In general, the report presents the final story of the
analysis and shows how the data fits that story. In many
cases, we are more accustomed to a method of analysis
reporting that demonstrates how the data reveals what the
most likely story is. Consequently, we have looked to see
if the data have been presented in a way so that the story
holds together.
2. Page 2, Second paragraph from bottom - We agree with the EPA
proposed treatment of considering evaporative emissions for
gross liquid leakers as independent of ambient temperature
and RVP.
EPA, of course, agrees with its own methodology.
3. Page 3, Paragraph 3 - The report seems to begin the
discussion of substantial leakers and gross leakers in a
manner that is confusing to the reader. We suggest, and
perhaps this is the intended meaning of the author, that
substantial leakers are those leakers which have a lower
limit of liquid leak rates than do gross liquid leakers.
For each type of emission a set of substantial liquid
leakers are analyzed. Then, at some point in the
development, only the gross liquid leakers are analyzed.
For example, later in the report for the hot soaks tests,
the substantial liquid leakers have rates of greater than 2
grams per hour and the gross liquid leakers have rates of
greater than 10 grams per hour. Consequently, we suggest
that beginning in Section 2.1, a distinction between
substantial and gross liquid leakers be made. The
parenthetical comments in Section 2.1 seem to say that
substantial liquid leakers and gross liquid leakers are
synonymous. We think that these parenthetical comments only
serve to cloud the distinction between substantial and gross
liquid leakers and, therefore, they should be removed.
These comments appear in the third, fourth and fifth
paragraphs on Page 3.
No, there was no attempt to define a "substantial" leaker
category. The word "substantial" only refers to the
magnitude of the leak. Our intent was to define a "Gross
Liquid Leaker" (GLL) as a vehicle having a substantial leak
of liquid gasoline. The exact magnitude of a "substantial
leak" (in terms of drops of gasoline per hour) was left
vague. However, the reader could use the lower bounds
specified for GLLs (see Table 2-1) to calculate such hourly
rates. The text has been revised to avoid this confusion.
4. Page 3, Paragraph 5 - We agree with the approach of using a
log-normal fit of the sparse data to estimate the gross
liquid leaker average emission rates to avoid simply
-------
-37-
calculating the mean of the sparse values which are
available.
5. Page 4, Second Paragraph - It took me a while to recognize
that the estimated diurnal emissions, which are referred to
in the second paragraph, is equal to the RTD emissions minus
24 times the resting loss emissions. We think it would be
helpful to the reader to insert a short paragraph before
this paragraph to remind the reader of this relationship.
We added that explanation to the beginning of Section 2.1.
6. Page 5, Paragraph 1 - We agree with the technique of
trimming the upper tail off the log-normal distribution for
the purposes of calculating the mean gross liquid leaker
emission rate; it reflects an engineering reality. We also
like the technique of determining the sensitivity of the
mean to doubling the value of the upper cutpoint. However,
we were curious about how much the mean would change if no
upper cutpoint were used, and we suspect that other readers
would have the same curiosity. Our gut feel is that, if the
upper cutpoints were at +infinity, the average emission rate
would be only slightly increased.
The RTD test of the vehicle with the highest diurnal (380.79
grams per day) was aborted after 16 hours because the
technicians were concerned that the SHED was approaching an
explosive concentration level. In this report we calculated
an average diurnal using a maximum of 5 times that
potentially explosive rate. Even that maximum seems too
high. By using still higher values, we risk reducing the
credibility of the analysis. The reader is of course free
to perform that calculation.
7. Page 5, Paragraph 2 - Choosing the value of the lower
cutpoint of the lognormal distribution is more problematic
then choosing the upper cutpoint. We felt that the
discontinuity argument of values between 25 and 62 grams per
day in the second paragraph was pretty weak since there are
larger discontinuities at larger emission values. We think
we agree that a lower limit is needed (on the other hand, it
may be possible that calculating the average value using the
lower tail of the log-normal distribution may not change the
average value much) to avoid double counting of emissions
from gross liquid leakers and the diurnal emissions of non-
gross liquid leakers which will be estimated from a
different routine in a MOBILE code. We think that a more
defensible approach to selecting the lower cutpoint would be
to consider the range of normal (not leakers) diurnal
emission values for the fleet using the existing routines in
MOBILE. In other words, could an analysis of the diurnal
emissions emitter model in MOBILE be done to verify that the
-------
-38-
lower cutpoint chosen for the diurnal emission gross liquid
leaker distribution does not produce a gap or a bump in the
distribution between the normal and the gross liquid leaker
models?
We agree that the selection of the lower bound (threshold)
for the gross liquid leakers is a weak point. It is highly
sample dependent. If a higher threshold value were
selected, the effect would be to increase the estimated
average diurnal emissions (from these GLLs). For example,
doubling the threshold from 25 to 50 grams per day would
increase the estimated average by almost 35 percent. While
this seems to be a large change, the actual effect on total
evaporative hydrocarbon is in consequential. As more data
become available, we may revise these threshold values.
8. Page 5, Paragraph 2 - It would be beneficial to the reader
to have an appendix to show how the average emissions for
the log-normal distribution with the cutpoints on the upper
and lower end are calculated. Most readers won't want to or
won't be able to go through this tricky calculation.
These averages were calculated by computing the area under
curves using Riemann sums (from first semester Calculus).
We see no need to include these calculations in this report.
9. Page 6, Second full paragraph - Comments 4, 6, 7, and 8
above apply generally to all of the different types of gross
liquid leaker calculations in Section 2. From this point
forward, the comments will apply only to specific issues on
individual gross liquid leaker types.
10. Page 7, Paragraph I - The last sentence talks about a
uniform distribution. We think that this is a relatively
minor comment but it did take me a while to understand what
the author was referring to. In the last paragraph on the
page, the report mentions that the uniform distribution
would have a better fit but the only reason that the report
gives to not chose the uniform distribution was for
consistency with other models in the report. However, there
is another reason that could be considered, and perhaps
mentioned, is that the uniform distribution would imply that
40% of the vehicles would have zero resting loss emissions.
We think that you will agree this is probably not the case.
The third paragraph on Page 7 also has a typo in the third
line: the word approached should be approaches.
Good point, the material has been revised to include this.
II. Page 11, Second full paragraph - The reference to the other
report suggests that the relative fuel pressures between TBI
and PFI systems are a factor of 4 different. This is not
-------
-39-
12 .
13.
explicitly stated in this report. Perhaps it should be if
the author believes it still to be correct.
We believe that the critical assumptions were stated. Going
off into a detailed discussion of relative pressures might
only cloud the issue.
Page 15, Table 2-1 - The footnote at the bottom of Table 2-1
brings up an issue. For the RTD data analysis, the resting
losses were removed from RTD to get diurnal emissions. But
the same approach was not used to separate resting losses
from hot soak emissions and running loss emissions. Why is
there a difference in analysis methods? Perhaps, it would
have been just as easy to determine the average RTD
emissions per day and then Table 2-1 would have had an entry
for RTD in place of diurnal.
Since we estimated resting loss emissions by averaging the
emissions during last six hours of the RTD test (i.e., the
hours corresponding to the period from midnight to 6 AM),
the resting loss values were available for each RTD test.
This permitted us to easily calculate for each RTD test a
resting loss / diurnal pair. This was not true of the hot
soak or running loss tests. Therefore, different approaches
were used.
The footnote at the bottom of Table 2-1 also is a surprise
to the reader. At a minimum, we suggest that the reader be
warned that this subtraction will occur by placing an
appropriate statement at the beginning of the analysis
sections for hot soak and running loss average emission rate
determinations.
As the reviewer suggested, an explanation has been added to
both the hot soak section and to the running loss section.
Page 15, Table 2-1 - One of the problems that we had in
following the discussion in the previous sections about the
determination of average gross liquid leaker emission rates
was the values used to determine substantial leakers, gross
leakers, lower cutpoints, and upper cutpoints of the log-
normal distributions. A table placed somewhere in the
report such as the following would help guide the reader
through these different values.
Hot Soak Test
(q/hr)*
Carbureted
TBI
PFI
Liquid Leakers
Substanti
al
>2
Gross
>10
Averaqinq
Range
10-300
10-300
10-300
GLL
Average
16.95
45.00
57.14
-------
-40-
Resting Loss
(q/hr)
Diurnal (g/day)
Running Loss
(g/mile) *
>0.25
RTD>30
>5
>2
>25
>7
2-50
25-1000
7-200
9.16
104.36
17.65
We agree that such a revised table would be useful. We
replaced Table 2-1 with a revised table, similar to this
one. (We noted, in response to the third comment, that
there is no category of "substantial leakers." Therefore,
the revised table is different from this one.)
14. Page 15, Second paragraph - The first assumption states that
for each test of evaporative emissions (RTD, hot soak, and
running loss tests)... Immediately we thought, where are the
resting losses? Aren't the frequencies of occurrence for
gross liquid leaker resting losses going to be estimated?
This seems to be a glaring omission.
The second paragraph of Section 2.1 explains that the RTD
test is used to obtain both the diurnal emissions and the
resting loss emissions.
15. Page 15, Paragraph 3 - In the second assumption, we think
that it is important to bring in engineering concepts about
how gross liquid leaks are related to the different types of
evaporative emissions. For example, if gross liquid leaks
are related to fuel pressure, they could occur for running
losses and hot soaks but not occur for resting losses and
diurnals. We think that this type of discussion would lend
engineering support to Assumption 2.
We have added that assumption (and example) to Section 3.0.
16. Page 15, Paragraph 3 - We think that we can understand what
Assumption 2 says. However, we do not follow the reasoning
behind the assumption. It seems to us that there should be
gross liquid leakers for each of the four types of
emissions. We do not understand why the report suggests
using two types to estimate the third type (that is, the
running loss and the RTD results to determine the hot soak
results). Because we could not understand the reasoning
behind this assumption, we did not understand why, on Page
16, the second step in Section 3.1 was necessary and, of
course, when it came to understanding the distinction
between Approach 2 and Approach 1, we were lost.
The approach was necessary because EPA lacked data on GLLs
on the hot soak test at ages over 10 years. This statement
has been added to the beginning of Section 3.0.
-------
-41-
17. Page 16, Section 3.1.1 - We think that using averages of
frequencies of occurrence of gross liquid leakers for the
three model year groups used in recruitment causes a large
amount of information to be lost from the data.
Additionally, since Assumption 1 states that gross liquid
leaker frequencies will be assumed to be the same for older
and newer technologies, there is no need to divide vehicles
into model year groups. A better and more defensible
approach for determining the logistic growth functions would
be simply to use logistic regression on the gross liquid
leak leaker indicators for each vehicle that was tested. A
logistic regression procedure, which is simple to use, is
available in SAS. For each logistic regression, the input
variable would be vehicle age and the response variable
would be an indicator variable that would have a value of
zero for a non-gross liquid leaker and one for a gross
liquid leaker. The procedure would fit the data to a model
with the same shape as shown in Figure 3-1. The procedure
also has options for outputting the confidence limits of the
predicted values.
True. However, we believe that Figure 3-1 (and the similar
figures that follow) illustrating the resulting equation
(curve) closely approximating the three averaged rates
(frequencies) is far more informative than having the same
cumulative distribution curve drawn through a cloud of data.
Additionally, the stratification into model year groups was
based upon the stratified (targeted) recruitment that was
used, not on potential differences in the rates of GLLs.
18. Page 17, Figure 3-1 - The figures such as Figure 3-1 could
still be used to show trends in the data and the model
results when logistic regression is used to build the
models. For example, the plot could be made to have the
average frequency for every five years of age and the model
resulting from logistic regression and the confidence limits
could be drawn as curves on the plot. The confidence limits
provided by the model would span the entire range of the
data.
That approach would produce a graph similar to the existing
Figure 3-1, with the exception that the individual points
would be equally spaced, but with more variance at each
point. We see no advantage to this, but the reader is free
to reanalyze the data.
Use of logistic regression would also appropriately solve
the logistic growth expression for gross liquid leakers
based on running loss testing, which is shown on Page 19 in
Figure 3-2. In this instance, using the average values for
the three model year groups caused a problem which has
probably occurred by chance alone, in that the oldest model
-------
-42-
year average had a lower value than the middle model year
value.
This approach only solves the "problem" by obscuring the
fact that the calculated rate of occurrence of GLLs in this
sample is slightly lower for the older vehicles. We still
believe (as noted in our response to comment 17) that this
graphical approach is more useful to the readers.
19. Page 19, Section 3.1.3 - This section starts with the phrase
"to estimate the rate of occurrence of gross liquid leakers
on the hot soak test..." since we did not understand
Assumption 2 fully, we do not understand why the gross
liquid leaker rate of hot soak needs to be estimated when it
could have been modeled just like it was for RTD and running
loss. We think that perhaps a Venn diagram would help in
clarifying the gross liquid leakers. We think the report is
using the following Venn diagram with two overlapping
circles for diurnal and running loss with the union of the
circles being hot soak gross liquid leakers.
The explanation that was added to the beginning of Section
3.0 (in response comment 16) was repeated (in the beginning
of Section 3.1.3) for emphasis.
We think that the Venn diagram for the gross liquid leakers
should start with the following Venn diagram which has four
overlapping circles for resting loss, diurnal, running loss,
and hot soak emissions. Then the report should consider
engineering relationships to see if it is possible to
simplify the diagram.
We do not believe that Venn diagrams are necessary.
20. Page 21, Section 3.2 - The only clue that we have as to how
the second approach differs from the first approach are the
two words "optimize simultaneously." If the frequency of
gross liquid leakers in the fleet is calculated
simultaneously (we assume this means a vehicle would be a
gross liquid leaker for all types of emissions) then
wouldn't there be just one equation to predict the gross
liquid leaker rate of occurrence? Because we could not
understand Assumption 2 and the distinction between Approach
1 and Approach 2, we could not comment intelligently on
Section 3.3.
As the reviewer pointed out (in comment 15), a vehicle might
qualify as a GLL on only one or two the three evaporative
tests that we used. (An explanation of that was added to
the end of Section 3.0.) Thus, it is not only possible, it
is likely that there would be three distinct equations
(curves) for the frequency of the different types of GLLs.
-------
-43-
Appendix F
Response to Comments from Stakeholders
No comments were submitted in response to EPA's posting a
draft of this report on the MOBILE6 website.
------- |