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.

-------