United States Air and Radiation EPA420-R-01-018
Environmental Protection April 2001
Agency M6.EVP001
vvEPA Evaluating Resting Loss and
Diurnal Evaporative
Emissions Using RTD Tests
> Printed on Recycled Paper
-------
EPA420-R-01-018
April 2001
RTD
M6.EVP.001
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
This document reports both on the methodology used to
analyze the data from real-time diurnal (RTD) tests on 270
vehicles and on the results obtained from those analyses. The
purpose of the analysis was to develop a model of the diurnal and
resting loss emissions of the in-use fleet to be used in MOBILE6.
This report was originally released (as a draft) in October
1997, and then revised (and re-released) in July 1999. This
current version is the final revision of the July 1999 draft (of
M6.EVP.001). 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 Vehicle Sample 2
3.0 Vehicle Testing 4
4.0 Weighting the EPA Data 5
5.0 Test Parameters 8
6.0 Consolidating Vehicle Parameters for 24-Hour RTD . . 10
6.1 Comparing TBI and PFI Vehicles 10
6.2 Comparing Carbureted and FI Vehicles 12
6.3 Comparing Cars and Trucks 15
6.4 Summarizing Stratification Parameters 18
6.5 Evaluating Untested Strata 19
7.0 Evaporative Emissions Represented by the RTD Test . 19
7.1 Resting Loss Emissions 20
7.2 Diurnal Emissions 20
7.3 Separating Out Gross Liquid Leakers 21
8.0 Characterizing Resting Loss Emissions 23
9.0 Characterizing 24-Hour Diurnal Emissions 27
10.0 Gross Liquid Leakers 32
10.1 Frequency of Gross Liquid Leakers 32
10.2 Magnitude of Emissions 35
10.3 Effects of Vapor Pressure Changes 37
11.0 Other Topics 37
11.1 Temperature Ranges 38
11.2 Heavy-Duty Vehicles 38
11.3 High Altitude Emissions 39
11.4 Motorcycles 40
11.5 Pre-Control Vehicles 41
11.6 Duration of Diurnal Soak Period 43
11.7 1996 and Newer Model Year Vehicles 44
11.8 Tier-2 Vehicles 44
11
-------
TABLE OF CONTENTS (Continued)
Page Number
APPENDICES
A. Temperature Cycles 46
B. Vapor Pressure 47
C. Mean Emissions by Strata 50
D. Modeling Hourly Resting Loss Emissions 53
E. Regression Tables of Diurnal Emissions 54
F. Modeling 24-Hour Diurnal Emissions 59
G. Plots Comparing Diurnal Models to
Means of Measured Data 60
H. Peer Review Comments from H. T. McAdams 68
I. Peer Review Comments from Sandeep Kishan 78
J. Comments from Stakeholders 87
111
-------
Evaluating Resting Loss and Diurnal
Evaporative Emissions Using RTD Tests
Report Number M6.EVP.001
Larry C. Landman
U.S. EPA Assessment and Standards Division
1.0 INTRODUCTION
In previous versions of the highway vehicle emission factor
model (MOBILE), the estimates of the emissions resulting from the
daily rise of the ambient air temperature were based on a one-
hour test (adjusted to simulate an 8-hour test) in which the
heating process was accelerated. As part of the MOBILE model
revision, an effort has been undertaken to use the recently
developed 72-hour real-time diurnal (RTD) test (or a shortened
version) to more accurately estimate those temperature driven
(i.e., diurnal) emissions, as well as the resting loss emissions.
In the RTD test, the ambient temperatures gradually cycle
over a 24 degree Fahrenheit range during the course of each 24
hour period as illustrated below in Figure 1-1:
Figure 1-1
Nominal RTD Temperature Cycle
(Temperatures Cycling Between 72° and 96° F)
100
70°
12
Time (hours)
18
24
-------
-2-
The three hourly temperature cycles used in this study are given
in Appendix A. These three temperature cycles are parallel
(i.e., identical hourly increases/decreases). Each temperature
cycle peaks at hour nine (i.e., at 3PM). The most rapid increase
in temperatures occurs during the fourth hour. For RTD tests
that exceed 24 hours (i.e., 33, 38, or 72 hours), the cycle is
simply repeated.
This document reports both on the methodology used to
analyze the data from these RTD tests and on the results obtained
from those analyses.
2.0 VEHICLE SAMPLE
In this analysis, EPA used real-time diurnal (RTD) test data
from two sources:
1) from five (5) individual testing programs (i.e., work
assignments) performed for EPA by its contractor, and
2) from a testing program performed for the Coordinating
Research Council (CRC).
The RTD testing performed for EPA was done by its testing
contractor (Automotive Testing Laboratories) over the course of
five work assignments from 1994 through 1996 (performed under
three different EPA contracts). A total of 119 light-duty
vehicles (LDVs) and light-duty trucks (LDTs) were tested in these
programs. In the following table (Table 2-1), the distribution
of those 119 test vehicles is given:
1) by work assignment number,
2) by vehicle type (LDV versus LOT),
3) by model year range, and
4) by fuel metering system
carbureted (Carb)
port fuel injected (PFI)
throttle body injection (TBI).
-------
-3-
Table 2-1
Distribution of EPA Test Fleet
Work
Assiqnment No.
2-09
1-05
0-05
0-07
0-11
Vehicle
Type
LDV
LDV
LOT
LDV
LDV
LOT
Model Year
Ranqe
80-85
86-95
80-85
86-95
86-95
71-77
78-79
80-85
86-95
86-95
71-77
78-79
80-85
86-95
Fuel Metering
Carb
5
7
3
1
0
3
1
5
0
0
2
0
5
0
PFI
2
15
4
24
0
0
0
0
0
5
0
0
0
5
TBI
0
10
3
12
2
0
0
0
0
1
0
0
0
4
The recruitment method used for most of the vehicles in the
EPA sample was designed to recruit a larger number of vehicles
that had potential problems with their evaporative control
systems. Specifically, two tests of the integrity of each
vehicle's evaporative control system (a purge test and a pressure
test) were used to screen the candidate vehicles. This resulted,
among the newer vehicles, in a larger proportion of the test
vehicles failing either a purge test or pressure test (but not
both) than did the corresponding vehicles in the in-use fleet.
EPA excluded from its sample all those vehicles that failed both
the purge and pressure tests. Any analyses performed on the EPA
data must, therefore, account for this intentional bias toward
problem vehicles. (See Section 4.0.)
It is important to note that neither the purge test nor the
pressure test is a perfect identifier of vehicles that have
problems with their evaporative control systems. While vehicles
that passed both the purge test and the pressure test had, on
average, lower RTD emissions than similar vehicles that failed
either or both tests, there was a wide overlap of the RTD
emissions of the vehicles that passed both tests with the RTD
emissions of similar vehicles that failed one or both of those
tests. The size of the overlap varied with the strata (see
Section 6.4). But, on average, the cleanest (i.e., vehicles with
the lowest RTD results) one-fourth of the vehicles failing the
purge and/or pressure test(s) had lower RTD test results than the
dirtiest (i.e., highest RTD results) similar vehicles that passed
both the purge and pressure tests. In fact, the vehicle that had
-------
-4-
the highest RTD emissions (other than the seven gross liquid
leakers discussed in section 7.3) was one that passed both tests
The CRC program* involved performing RTD tests on a random
sample of 151 vehicles (mostly LDTs) during 1996. The
distribution of those 151 vehicles (by vehicle type, model year
range, and fuel metering system) is given below in the Table 2-2
Table 2-2
Distribution of CRC Test Fleet
Vehicle
Type
Car
Truck
Truck
Truck
Model Year
Range
71-77
71-77
80-85
86-91
Carb
38
13
47
7
PFI
0
0
2
24
TBI
0
0
1
19
3.0 VEHICLE TESTING
The testing in the EPA study consisted of performing one or
more RTD tests on each vehicle in its "as-received" condition
with the exception that the tank fuel was replaced with specified
fuels. (To restore the vehicle to its "as-received" condition
for subsequent tests, the canister was conditioned to return it
to approximately the condition it was in prior to the first
test.) Up to three temperature cycles were used. (In addition to
the standard 72°-96° F cycle, 60°-84° and 82°-106° cycles were
also used.) Similarly, up to four different fuel volatilities
were specified; specifically, fuels having nominal Reid vapor
pressure (RVP) of 6.3, 6.7, 6.9, and 9.0 pounds per square inch
(psi). Since the actual RVP used in a given test may vary
slightly from the specified target RVP, EPA felt that tests
performed using the 6.7 or 6.9 psi RVP fuel could all be treated
as equivalent to tests performed using a fuel with a nominal RVP
of 6.8 psi.
The testing in the CRC study consisted of performing a
single RTD test on each vehicle in its "as-received" condition.
Each test used the standard temperature profile (i.e.,
temperatures cycling between 72° and 96° F) and was performed
using the fuel already in each vehicle's fuel tank (typically
having an RVP which ranged from 6.7 to 7.0 psi). EPA felt these
D. McClement, J. Dueck, B. Hall, "Measurements of Diurnal Emissions from
In-Use Vehicles, CRC Project E-9", Prepared for the Coordinating Research
Council, Inc. by Automotive Testing Laboratories, Inc., June 19, 1998.
-------
-5-
tests could also be treated as equivalent to tests performed
using a fuel with a nominal RVP of 6.8 psi.
For the purpose of the following analyses, we treated all
testing performed using fuels with RVPs from 6.7 through 7.0 as
if they were all performed using a fuel with a nominal RVP of 6.8
psi. Thus, all the EPA testing performed using fuels with
nominal RVPs of either 6.7 or 6.9 will be combined and then used
with all of the CRC tests.
4.0 Weighting the EPA Data
To correct for the intentional sampling bias toward
"problem" vehicles in the EPA testing programs (described in
Section 2.0), we first determined the number of vehicles in each
stratum in both the recruited sample and the in-use fleet.
Examining the purge/pressure data gathered in the I/M lanes in
Arizona and Indiana, we found 16,637 as-received vehicles for
which successful purge and pressure tests were performed. (These
tested were conducted at the Phoenix, Arizona I/M lane from June
1992 through August 1994 and at the Hammond, Indiana I/M lane
from January 1990 through February 1995.)
Modeling those preceding distributions with smooth (i.e.,
logistic growth) curves as functions of vehicle age* produced the
distributions in Table 4-1. A full discussion of this process is
given in Document Number M6.EVP.006, entitled "Estimating
Weighting Factors for Evaporative Emissions in MOBILE6." The
predicted purge failure rates (i.e., the sum of columns two and
three in the above table) closely approximates those used in the
MOBILES model for vehicles up to 12 years of age. The predicted
pressure failure rates (i.e., the sum of columns three and five)
also closely approximates those used in the MOBILES model for
vehicles up to 12 years of age. Any differences between the
estimates used in MOBILES and those in Table 4-1 should not
affect the analyses in this report. A detailed analysis of the
failure rates on the purge and pressure tests (and, hence on the
appropriate weighting factors) is presented in document number
M6.EVP.006.
This approach assumes that the purge/pressure results are
functions only of age (i.e., independent of vehicle type, fuel
metering system, model year, etc.). To use these distribution
estimates within a given stratum (e.g., 1980-85 carbureted LDVs),
we determined the numbers of vehicles in each of the purge/
pressure categories that we would expect to find in a randomly
selected sample of the in-use fleet. We then calculated the
Vehicle age was estimated by first subtracting the model year from the
test year, and then adjusting so that the final value represents the age
at January first (which is the standard date for the MOBILE model).
-------
-6-
ratio of those expected category sizes to the number of vehicles
actually recruited and tested within each of those four
categories. Those ratios then became the weighting factors for
the analysis of that stratum (only 1995 and older model years).
The values in Table 4-1 will be adjusted to account for the
presence of an I/M program (see document M6.IM.003, entitled
"Estimating Benefits of Inspection/ Maintenance Programs for
Evaporative Control Systems").
NOTE: Since no vehicles in the EPA testing programs were
recruited from among those that failed both the purge and the
pressure tests (the third column in the following table), EPA
used the data from the CRC program to characterize the RTD
emissions of that category. Since (as Table 4-1 indicates) this
stratum is relatively small for newer vehicles, its exclusion had
at most only a slight affect on the estimate of fleet emissions
of those newer vehicles. (See Section 6.5.)
-------
-7-
Table 4-1
Predicted Distribution of Purge/Pressure Results
(By Vehicle Age -- Independent of Model Year)
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
Fail Purge
Pass Pressure
1.77%
1.80%
1.88%
2.02%
2.23%
2.53%
2.95%
3.51%
4.25%
5.23%
6.47%
8.00%
9.76%
11.61%
13.29%
14.51%
15.06%
14.95%
14.41%
13.70%
13.03%
12.50%
12.13%
11.89%
11.74%
1 1 .65%
Results on Purge i
Fail Purge
Fail Pressure
0.09%
0.12%
0.16%
0.23%
0.32%
0.44%
0.60%
0.84%
1.15%
1.58%
2.16%
2.93%
3.95%
5.26%
6.91%
8.93%
11.30%
13.97%
16.83%
19.73%
22.53%
25.08%
27.31%
29.18%
30.68%
31.87%
•md Pressure Test;
Pass Purge
Pass Pressure
95.00%
94.93%
94.72%
94.36%
93.81%
93.03%
91.96%
90.51%
88.57%
85.99%
82.62%
78.30%
72.94%
66.60%
59.58%
52.42%
45.76%
40.14%
35.84%
32.80%
30.81%
29.58%
28.85%
28.45%
28.23%
28.11%
» —
Pass Purge
Fail Pressure
3.14%
3.15%
3.23%
3.39%
3.65%
4.00%
4.49%
5.15%
6.03%
7.20%
8.75%
10.77%
13.35%
16.52%
20.21%
24.14%
27.88%
30.93%
32.93%
33.77%
33.63%
32.84%
31.70%
30.49%
29.35%
28.37%
-------
-8-
5.0 TEST PARAMETERS
Since emissions from vehicles classified as gross liquid
leakers (vehicles identified as having substantial leaks of
liquid gasoline, as opposed to simply vapor leaks) are
characterized separately from those of the remaining vehicles,
the analyses in this section were also performed with those
vehicles omitted (see section 7.3).
There are three testing parameters in the EPA programs that
could affect the RTD test results. (The results of the RTD tests
include both diurnal and resting loss emissions.) Those are:
1) the RVP of the test fuel,
2) the temperature cycle, and
3) the site from which each vehicle was recruited.
Since it is well known that both the ambient temperature and
the fuel volatility will affect evaporative emissions, these two
parameters were automatically included in the calculations. All
of the analyses that used tests performed with fuels ranging from
6.7 to 7.0 psi RVP were conducted assuming the nominal RVP to be
6.8 psi, as noted previously.
The question of whether the "site" variable is significant
was raised because EPA's testing contractor (ATL) recruited
vehicles from two different parts of the country. Twenty-two
(22) vehicles were recruited from and tested in Indiana; the
remaining 97 vehicles were recruited from and tested in Arizona.
Since the higher temperatures in Arizona might have resulted in
higher canister loadings for those as-received vehicles, we
compared the cumulative distributions of the 24-hour RTD results
(weighted to correct for recruitment bias) of the 1986 and newer
LDVs tested at both sites. In Figure 5-1, we compare the six
PFIs tested in Indiana with the 35 in Arizona. In Figure 5-2, we
compare the four TBIs tested in Indiana with the 17 in Arizona.
All of these 24-hour RTD emissions were obtained using 6.7-6.9
psi RVP fuel over the 72°-96° Fahrenheit cycle.
Despite the small sample sizes in the Indiana data (only six
PFIs and four TBIs), the closeness of the distribution curves (in
Figures 5-1 and 5-2) is compelling and suggests that there is no
reason to treat the test data separately. Therefore, the "site"
parameter was dropped from the remaining analyses.
-------
-9-
Figure 5-1
Weighted Cumulative Distributions at Two Sites
RTD Emissions of the 1986 and Newer PFIs
100%
-PFIs in INDIANA
PFIs in ARIZONA
10 20
24-Hour RTD Emissions (grams of HC)
30
Figure 5-2
Weighted Cumulative Distributions at Two Sites
RTD Emissions of the 1986 and Newer TBIs
100%
-TBIs in INDIANA
TBIs in ARIZONA
10 20
24-Hour RTD Emissions (grams of HC)
30
-------
-10-
6.0 CONSOLIDATING VEHICLE PARAMETERS FOR 24-HOUR RTD
Since emissions from vehicles classified as gross liquid
leakers (see section 7.3) are characterized separately from those
of the remaining vehicles, the analyses discussed in this section
were also performed with those vehicles omitted.
When analyzing exhaust emissions, we note that some vehicle
technologies (sometimes identified by model year ranges) have
distinct exhaust emission characteristics. Before beginning the
primary analysis of these evaporative emissions, we examined the
data to determine if analogous technology groupings exist for the
RTD test results. Specifically, it was necessary to determine:
1) whether test results from different model year ranges (i.e.,
1981-85 and 1986-93) can be combined,
2) whether test results from port fuel-injected vehicles (PFIs)
can be combined with throttle body injected vehicles (TBIs)
into a single stratum of fuel-injected vehicles,
3) whether test results from carbureted vehicles can be
combined with fuel-injected vehicles, and
4) whether test results from cars and trucks can be combined
(despite the differences in fuel tank size).
We stratified the test vehicles using the following three
model year ranges:
1) 1972 through 1979,
2) 1980 through 1985, and
3) 1986 through 1995.
Based on the assumption that changes to the EPA certification
requirements for evaporative emissions will result in changes to
vehicles' evaporative control systems, we separated the RTD
results on the pre-1980 vehicles from the results on the 1980 and
newer vehicles. (For the same reason, data from the 1996 and
newer model year vehicles will form a new stratum once we begin
to test those vehicles.) While a similar argument can be made
for an additional break at the 1978 model year point, we lacked
the data to separately analyze the 1978-79 model year vehicles.
A second break point was added between the 1985 and 1986 model
years at the recommendation of some of the automotive
manufacturers who based their suggestion on improvements in the
control of evaporative emissions. Therefore, this second break
point was not based on any changes in EPA test requirements or
applicable standards nor on any analysis of the results of the
RTD tests.
6.1 Comparing TBI and PFI Vehicles
To determine the appropriateness of combining the RTD test
results of PFIs with those of TBIs, we found two samples
containing otherwise similar vehicles:
-------
-11-
1) 1986 and newer trucks in the CRC testing program (see Figure
6 -1) and
2) 1986 and newer LDVs in the EPA testing program (see Figure
6-2) .
In each of those two samples, the testing was performed over the
72°-96° temperature cycle using fuel with an RVP ranging from 6.7
to 7.0 psi. The similarity between PFI and TBI among the 1986
and newer model year trucks in the CRC testing program is
illustrated in Figure 6-1.
Figure 6-1
Cumulative Distributions of PFIs and TBIs
RTD Emissions in the CRC Testing Program
0)
05
s
C
0)
o
JO
3
E
3
O
100%
75%
50%
-CRC 86-91 Truck PFI
CRC 86-91 Truck TBI
25%
10 20 30
24-Hour RTD Emissions (grams of HC)
40
Characterizing those two CRC samples yields (in units of grams
per day over the RTD test):
1986-91 CRC
Truck TBIs
1986-91 CRC
Truck PFIs
Sample
Size
19
24
Median
3.13
2.05
Mean
5.41
5.85
Standard
Deviation
5.70
7.87
The similarity between PFI and TBI among the 1986 and newer model
year LDVs in the EPA testing program is illustrated on the
following page in Figure 6-2.
-------
-12-
Figure 6-2
Weighted Cumulative Distributions of PFIs and TBIs
RTD Emissions in the EPA Testing Program
100%
- EPA 86-95 LDVPFI
EPA 86-95 LDV TBI
0%
10 20 30
24-Hour RTD Emissions (grams HC)
40
Both the distributions shown in Figure 6-2 and the
characterizations of those two EPA samples presented in the
following table (in units of grams per day over the RTD test]
have been weighted to correct for recruitment bias.
1986-95 EPA
LDV TBIs
1986-95 EPA
LDV PFIs
Sample
Size
21
41
Median
4.52
2.08
Mean
9.84
9.32
Standard
Deviation
12.22
19.75
Based on the similarity of the cumulative distribution curves and
on the close fit of the means relative to the respective standard
deviations (in the strata illustrated in Figures 6-1 and 6-2),
the PFI and TBI strata were merged into a single fuel-injected
(FI) stratum for the remaining analyses.
6.2 Comparing Carbureted and Fuel Injected Vehicles
To determine whether test results from carbureted vehicles
can be combined with those from fuel injected vehicles, we
identified the only four samples containing otherwise similar
vehicles:
-------
-13-
1) in the CRC testing program, 43 1986-1991 FI trucks and 7
corresponding carbureted trucks (see Figure 6-3),
2) in the EPA testing program, 64 1986-1995 FI LDVs and 6
corresponding carbureted LDVs (see Figure 6-4),
3) in the CRC testing program, 3 1980-85 FI trucks and 46
corresponding carbureted trucks, and
4) in the EPA testing program, 6 1980-85 FI LDVs and 13
corresponding carbureted LDVs.
However, the two comparisons using the 1980-85 model year
vehicles produced mixed results (possibly due to the small number
of FI vehicles in the samples).
The difference in the distributions between carbureted
(Carb) and FI trucks among the 1986-1991 model year trucks in the
CRC testing program is illustrated in the following table (in
units of grams per day over the RTD test) and in Figure 6-3.
Comparing Carbureted Trucks to Fuel -Injected Trucks
1986-91 CRC
LOT Garbs
1986-91 CRC
LOT FIs
Sample
Size
7
43
Median
6.15
2.85
Mean
9.31
5.65
Standard
Deviation
8.28
6.92
-------
-14-
Figure 6-3
Cumulative Distributions of FIs and Garb Trucks
RTD Emissions in the CRC Testing Program
100%
D)
«J
§5
£
Q)
Q.
o
75%
50%
CRC 86-91 TrkFI
-CRC 86-91 TrkCarb
25%
10 20
24-Hour RTD Emissions (grams of HC)
Similarly, the difference in the distributions between
carbureted (Garb) and FI cars among the 1986-1995 model year LDVs
in the EPA testing program is illustrated in the following table
and in Figure 6-4. Both the distributions shown in Figure 6-4
and the characterizations of those two EPA samples represented in
the following table (in units of grams per day over the RTD test)
have been weighted (using Table 4-1) to correct for recruitment
bias.
Comparing Carbureted LDVs to FI LDVs
1986-95 EPA
LDV Garbs
1986-95 EPA
LDV FIs
Sample
Size
6
64
Median
10.56
3.41
Mean
10.34
9.50
Standard
Deviation
6.73
17.23
-------
-15-
Figure 6-4
Weighted Cumulative Distributions of FIs and Garbs LDVs
RTD Emissions in the EPA Testing Program
cu
O)
2
c
cu
O
5
Q.
cu
'*-J
JO
3
E
3
O
100%
75%
50%
25%
0%
EPA86-95LDVFI
-EPA 86-95 LDVCarb
10 20
24-Hour RTD Emissions (grams of HC)
30
Statistical tests support the hypothesis that the means of
the RTD test results are different for the 1986-1991 model year
trucks. Also, the large standard deviation (relative to the
difference of the means) for the sample of 1986-1995 model year
passenger cars will not allow us to confirm that hypothesis using
statistical tests. However, it is noteworthy that every
carbureted vehicle in each sample had RTD test results higher
than the median of the corresponding fuel injected vehicle
sample. (An unlikely situation if the RTD emissions of the
sample of fuel-injected and sample of the carbureted vehicles
were to be indistinguishable from each other.)
Therefore, EPA will treat the carbureted vehicles and the FI
vehicles as distinct strata for the remaining analyses.
6.3 Comparing Cars and Trucks
Determining the appropriateness of combining the RTD test
results of LDVs with those of LDTs presented different problems.
Specifically, the CRC sample was exclusively trucks except for
the 1971-77 stratum, and the EPA sample (using 6.7-6.9 RVP fuel)
was almost exclusively cars. The obvious solution was to compare
the CRC trucks with the EPA cars. However, because of the
difference in recruitment methods, we first had to omit from the
CRC sample those vehicles which would not have been recruited in
the EPA sample (i.e., those failing both purge and pressure), and
we then re-weighted the remaining results (as we did with the EPA
sample). This produced the following two strata with which to
-------
-16-
investigate the differences in RTD results between cars and
trucks:
1) in the combined EPA and CRC testing programs, the weighted
results of 13 1980-85 carbureted LDVs and 44 corresponding
carbureted trucks (Figure 6-5), and
2) in the combined EPA and CRC testing programs, the weighted
results of 62 1986 and newer FI LDVs and 42 corresponding
carbureted trucks (Figure 6-6).
Figure 6-5
Weighted Cumulative Distribution of Cars and Trucks
RTD Emissions in the EPA and CRC Testing Programs
(1980-1985 Model Year Carbureted Vehicles)
100%
~ 75%
o
D)
<0
4-^
C
0)
% 50%
Q.
0)
'*J
O
3
o
25%
0%
EPA80-85CarbLDVs
CRC 80-85 CarbTrucks
10 20
24-Hour RT D E missions (grams of HC)
30
-------
-17-
Fiqure 6-6
Weighted Cumulative Distribution of Cars and Trucks
RTD Emissions in the EPA and CRC Testing Programs
(1986 and Newer Model Year FI Vehicles)
cu
O)
I
cu
2
cu
Q.
cu
'*—«
JO
3
E
3
O
100%
75%
50%
25%
0%
EPA86-95FILDVS
• - - CRC 86-91 FI Trucks
0 10 20
24-Hour RTD Emissions (grams of HC)
30
The distributions in Figures 6-5 and 6-6 and the
characterizations of those strata (in the following table, in
units of grams per day over the RTD test) have been weighted to
correct for the actual recruitment bias in the EPA sample and the
simulated bias in the CRC sample.
1980-85 LDVs
Carbureted
1980-85 LDTs
Carbureted
86-95 FI LDVs
86-91 FI LDTs
Sample
Size
13
44
62
42
Median
10.22
10.55
3.40
3.11
Mean
11.29
10.58
9.48
5.99
Standard
Deviation
5.04
6.44
17.54
7.67
In Figure 6-5, the distributions of the carbureted 1980-85
cars and trucks are virtually identical. Statistical tests,
using the results from the first two rows in the above table,
also support the hypothesis that the means of the RTD test
results are the same for the carbureted 1980-85 cars and trucks
(relative to the standard deviations). Therefore, EPA will treat
the carbureted cars and trucks as a single stratum for the
remaining analyses.
-------
-18-
In Figure 6-6, the distributions of the FI 1986-95 cars and
trucks appear virtually identical up to about the 75 percentile
point, after which they diverge slightly. However, statistical
tests (using the means, sample sizes, and standard deviations
from the preceding table) do not support the hypothesis that the
means of the RTD test results are the same. Regardless of the
statistical tests, EPA decided to treat the fuel-injected 1986-95
model year cars and trucks the same for the following two
reasons:
4 the similarity of the cumulative distributions up through
the 75 percentile point, and
4 the shortage of the RTD testing of the 1986-95 model year
FI trucks over a range of temperature cycles and fuel
RVPs (which would be necessary to characterize the RTD
emissions if trucks were to be treated differently than
cars).
Therefore, EPA will combine the cars and trucks into a
single stratum for the remaining analyses. This conclusion seems
reasonable based on the fact that the larger fuel tanks (and
hence potentially larger vapor volumes) of trucks are offset by
the reportedly larger canister volumes.
6.4 Summarizing Stratification Parameters
For each combination of the pass/fail results on the
(screening) purge test and pressure test (i.e., recruitment
groups), we stratified the combined 119 vehicle EPA and 151
vehicle CRC data into the following five strata:
Model Year Range
1971-1979
1980-1985
1986 and Newer
Number of
Carbureted
Vehicles
57
65
15
Number of Fuel
Injected
Vehicles
*
12
121
* No data were available for this stratum. We simply
applied the results of the 1971-79 carbureted
vehicles to characterize this stratum.
These five (tested) strata, in the above table, were then
subdivided to include the recruitment criteria and yielded the
substrata listed in Appendix C. Three of these 20 strata were
not tested, and two of the remaining had only limited coverage.
These five missing or poorly covered strata are comprised of
vehicles that failed both the purge and pressure tests.
20
-------
-19-
6.5 Evaluating Untested Strata
As noted in the previous section, the strata that are either
missing or poorly represented in our sample fall into two
categories:
1) No pre-1980 model year vehicles equipped with fuel
injection were recruited because of the small numbers
of pre-1980 model year vehicles in the in-use fleet.
2) The vehicles that failed both the purge and the
pressure tests were:
4 systematically excluded from the EPA sample and
4 missing or poorly represented in CRC's sample of the
newer model year vehicles due to their relative
rarity among the newer vehicles (see Table 4-1).
For the MOBILE model, EPA will estimate the RTD emissions of
the (untested) pre-1980 fuel injected vehicles as being identical
to the corresponding emissions of the pre-1980 carbureted
vehicles. This should be a safe assumption since any actual
differences between these strata should be balanced by the
relatively small number of these fuel injected vehicles in the
in-use fleet. (In fact, MOBILE6 assumes that the pre-1980
vehicles are all carbureted.)
To characterize the vehicles that failed both the purge and
pressure tests, we identified 14 such vehicles that were not
gross liquid leakers (all tested as part of the CRC study).
Thirteen (of those 14) were pre-1980 carbureted vehicles. For
those 13 vehicles, the mean of the (24-hour) RTD emissions was
25.11 grams (with a standard deviation of 12.00). The
corresponding stratum of pre-1980 vehicles that passed the purge
test but failed the pressure test contains 20 vehicles (18 CRC
and 2 EPA) has a mean (24-hour) RTD emissions of 24.39 grams
(with a standard deviation of 7.77). Since the difference
between those means is not statistically significant, we will
combine those two strata into a single stratum of vehicles that
failed the pressure test (regardless of their results on the
purge test). (This approach permits us to avoid having to
estimate emissions from the untested strata of newer vehicles
that fail both the purge and pressure tests.)
7.0 EVAPORATIVE EMISSIONS REPRESENTED BY THE RTD TEST
The results from the real-time diurnal (RTD) tests can be
used to model the following two types of evaporative emissions:
1) "Diurnal" emissions are the pressure-driven emissions
resulting from the daily increase in temperature.
2) "Resting loss" emissions are the relatively stable
emissions that are always present.
-------
-20-
7.1 Resting Loss Emissions
Examinations of the RTD data suggest that, for virtually all
of the tests (regardless of the temperature cycle, fuel RVP, or
vehicle type), the hourly HC evaporative emissions had stabilized
and were relatively constant for hours 19 through 24. (See
Figure 7-1.) This suggests that the average hourly emissions
during the final six hours of the 24-hour RTD cycle correspond to
what this paper refers to (in the previous section) as hourly
"resting loss" emissions.
Figure 7-1
Identifying Resting Losses
(Stable Portion of RTD Hourly Emissions)
100
Hourly Emissions
0
12 24
Time (hours)
U)
c
o
'55
LU
Q
0.0
36
The "resting loss" emissions component of each RTD test was
calculated as the average (i.e., mean) hourly RTD emissions for
hours 19 through 24, at the nominal temperature for the twenty-
fourth hour. In this example, the average emissions for that 6-
hour period (0.10 grams per hour) would represent this vehicle's
hourly resting losses at a stable 72°F with a fuel having RVP of
6.8 psi. The mean hourly resting loss emissions (temperatures of
60°, 72° and 82°) for each of the strata in Section 6.4 are given
in Appendix C.
7.2 Diurnal Emissions
Subtracting the hourly resting loss emissions (calculated in
Section 7.1) from the hourly RTD emissions should yield an
estimate of the hourly emissions that result from the daily rise
in temperature (i.e., "diurnal" emissions). Although the hourly
resting loss emissions will vary as the ambient temperature
-------
-21-
cycles over the full range of the RTD test (see Section 8.0), the
variation is small relative to the RTD hourly emissions. Using a
"temperature adjusted" resting loss value will result in a
slightly higher level of resting loss emissions over the day, and
a corresponding lower level of diurnal emissions over that day.
The total emissions will be unchanged.
In the following figure, the hourly resting loss emissions
correspond to the unshaded area. The remaining (i.e., shaded)
area then corresponds to the hourly diurnal emissions which are
primarily pressure-driven vapor leaks. This approach produces
calculated hourly diurnal emissions that approach zero as the
SHED (i.e., "ambient") temperature drops to near the starting
temperature.
Figure 7-2
Estimating Diurnal Emissions
(Pressure Driven Vapor Leaks)
O
1.5
1.0
>> 0.5
3
O
X
0.0
Pressure Driven
Vapor L eaks
0
12 18
Time (hours)
24
30
The average (mean) 24-hour RTD emissions for each of the
strata in Section 6.4 are given in Appendix C.
7.3 Separating Out "Gross Liquid Leakers" (GLLs)
The largest quantity of RTD data (combining data from the
EPA and CRC programs) was generated using fuel with an RVP
ranging between 6.7 and 7.0 psi over the 72°-96° F temperature
cycle. These test conditions were used by a total of 96 vehicles
in the EPA program and all 151 vehicles in the CRC program.
Using the preceding method to estimate hourly resting loss
emissions (at 72°F) for each of those 247 vehicles, we then
-------
-22-
plotted the full 24-hour RTD emissions versus those hourly
resting loss emissions (Figure 7-3).
Figure 7-3
Comparison of RTD versus Resting Loss Emissions
(72°-96°F Cycle Using 6.7-7.0 RVP Fuel)
ouu
O
0 600
E
2
1 40°
LLJ
Q
5 200
X
CM
n 1
I
•
•
•
•
•
0 5 10 15
Resting Loss Emissions (grams /hour)
20
This graph clearly illustrates that the test results of all
except five of the vehicles are tightly clustered with RTD
results under 100 grams (per 24-hours) and with hourly resting
losses under 1.5 grams per hour. The test results from each of
the remaining five vehicles are quite distinct from those of the
corresponding 242 tightly clustered vehicles. Each of these five
extremely high emitting vehicles was also identified, by the
mechanics who examined them, as having significant leaks of
liquid gasoline (as opposed to simply vapor leaks).
The RTD data in Figure 7-3 suggest that the evaporative
emissions from these five vehicles can exceed the emissions of
corresponding vehicles by one to two orders of magnitude. For
this reason, this report treats these "gross liquid leakers" as a
-------
-23-
separate category of evaporative emitters. It is important to
note that this category (i.e., "gross liquid leakers") is not a
new or previously unaccounted for source of emissions, since the
emissions from these vehicles had previously been included with
the resting loss and diurnal emissions in MOBILES. Thus,
modeling these vehicles separately (in MOBILE6) should have no
impact on the total evaporative emissions.
To define this category of "gross liquid leakers," we first
assumed that the effects of a significant liquid fuel leak should
be evident during the resting loss portion of the RTD test. This
report, therefore, defines a "gross liquid leaker" to be any
vehicle whose resting loss emissions are at least two grams per
hour. These five gross liquid leakers were all part of the CRC
study. Using this definition, we classified two vehicles in the
EPA study as likely gross liquid leakers. (These two are only
"likely" gross liquid leakers because no mechanic's inspections
were performed. We inferred their status based solely on their
resting loss emissions.) These two additional gross liquid
leakers do not appear in Figure 7-3 because they were tested only
on 6.3 and 9.0 psi RVP fuels.
It is important to note that another type of liquid leaker
is possible. Some leaks can occur only if the vehicle is
operating (e.g., leaks associated with the fuel pump).
Preliminary results from a running loss testing program being run
by CRC suggests that vehicles with such leaks may have higher
hourly evaporative emissions (in grams per hour) while they are
operating than the hourly (RTD) emissions from the gross liquid
leakers in this analysis. However, the gross liquid leakers
identified in this analysis have high evaporative emissions every
hour of the day; while, the other type of liquid leaker would
probably have high evaporative emissions only during the hours
the vehicle is actually operating. The effects of that other
type of liquid leaker will be included in the running loss
component of the evaporative emissions in the MOBILE model.
8.0 CHARACTERIZING RESTING LOSS EMISSIONS
Resting loss evaporative emissions are functions primarily
of ambient temperature. There are several distinct mechanisms
contributing to resting loss emissions:
4 permeation of the liquid fuel through the walls of both
hoses and (if applicable) plastic fuel tanks,
4 seepage of vaporized fuel at connectors and through cracks
in hoses, fuel tanks, etc.,
4 permeation and seepage at the canister, and
4 undetected (minor) liquid leaks of fuel.
-------
-24-
Some of these components of resting loss emissions are strongly
related to temperature changes while others such as the minor
liquid leaks are relatively unaffected by temperature changes.
As the first step in characterizing the effects of changes
in temperature and volatility on the hourly evaporative
emissions, we identified 57 vehicles in the EPA program that were
each tested:
4 using both the 6.8 and the 9.0 RVP fuels and
4 over all three temperature cycles.
Using this sample permitted us to have exactly the same vehicles
being tested at each combination of fuel RVP and temperature;
thus, avoiding many of the problems associated with vehicle-to-
vehicle test variability. This sample of 57 vehicles consisted
of:
4 12 1974-85 model year carbureted vehicles and
4 45 1985-94 model year fuel injected vehicles.
In the following graph (Figure 8-1), we plotted the mean hourly
resting loss emissions for the carbureted vehicles and the fuel
injected vehicles.
Based on the graphs in Figure 8-1 (on the following page),
we can make the following observations:
4 Hourly resting loss emissions increase with increasing
ambient temperature.*
4 For the fuel injected (i.e., the larger sub-sample), the
graph appears to contain a slight non-linear component.
However, with measurements at only three temperatures,
there are insufficient data to confirm that observation.
4 For the fuel injected (i.e., the larger sub-sample), the
graph appears to contain a slight non-linear component.
However, with measurements at only three temperatures,
there are insufficient data to confirm that observation.
An increase in hourly resting loss emissions corresponding to an increase
in fuel RVP was also noted (especially for the FI vehicles) . This
apparent relationship is believed to simply be an artifact of the vehicles
always being tested in the same (not randomized) order rather than being a
true relationship between resting loss emissions and Reid vapor pressure.
In the previous version of MOBILE, it was noted that resting loss
emissions appear to be insensitive to the fuel volatility level, and EPA
will continue to use that same assumption in this version of MOBILE.
-------
-25-
Figure 8-1
Mean Hourly Resting Loss Versus Temperature
(averaged at each temperature)
(Sub-Sample of 57 Vehicles)
0.3
0.2 --
0.1 --
0.0
4-
A
o
Carb (12)
Fl (45)
1 Regressions
50°
70° 90°
Ambient Temperature (°F)
110°
Therefore, for these 57 vehicles, the functions that most
reasonably model the hourly resting loss emissions (within the
tested range) are linear functions of temperature. That is:
Where:
Hourly Resting Loss = A + [ B * Temperature (°F) ]
•0.032040
•0.123027
•B1
0.002973
0.002769
For Carb Vehicles
For FI Vehicles
The two slopes (i.e., the "B" values in the above table) are
obviously close to each other in value. Since the difference
between the slopes was not statistically significant, the
regression was rerun, producing a single slope of 0.002812.
Having a single value for the slope (regardless of the stratum)
indicates that an increase in ambient temperature of ten degrees
Fahrenheit will, on average, correspond to an increase of 0.028
grams in each hour's resting loss emissions.
-------
-26-
While only the test results from the 57 vehicles that were
tested over the full range of fuel RVPs and temperature cycles
were used to calculate the coefficient ("B") which determines
the slope of the lines, the full data set was used only to solve
for the individual constant terms ("A").
For each of the strata identified in Section 6.4, we
calculated the value of "A" that would minimize the difference
between the predicted and the actual resting losses (i.e., the
residuals). If more tests had been conducted at a given
combination of temperature and fuel RVP (e.g., 72 °F using 6.8
psi RVP fuel), then the average resting loss emissions at that
combination was then more heavily weighted in the process to
calculate the value "A" .
This process produced a regression equation for each of the
18 strata; however, the predicted results based on the vehicle's
pass/fail status on the purge test were inconsistent. This
inconsistency is not surprising since the types of mechanical
problems that would cause a purge failure are not likely to
contribute to resting loss emissions.* To address this
situation, the population was stratified based simply on whether
the vehicles pass or fail just the pressure test. The regression
equations for each of the 12 resulting strata are given in
Appendix D. The regression equations are unique for each stratum
in which testing was performed. The untested strata of pre-1980
fuel-injected vehicles used the regression equations of the pre-
1980 carbureted vehicles.
Using any one of these 12 equations (in Appendix D), we can
estimate the hourly resting loss emissions for each hour of the
day for the three temperature cycles (in Appendix A) for that
stratum of vehicles. Adding those hourly estimates for the 24
hours of the day produces the daily resting loss emissions (for
that stratum). Repeating that process for all the strata in
Appendix D produces estimated resting loss emissions for all the
12 strata in Appendix D. Since all 12 of those equations are
linear, with the same coefficient for temperature, they produce
similar results: that the full day's resting loss emissions (in
grams) would be 24 times the hourly resting loss (calculated at
the lowest temperature of the day) plus 0.766.
These equations predict resting loss emissions of the
carbureted vehicles to be higher than for the fuel injected
vehicles. While these regressions can be used to calculate
reasonable estimates of resting loss emissions within the range
of temperature and fuel RVPs that were actually tested, we must
determine (see Section 11) how to extrapolate beyond the limits
of the test data.
This is consistent with the previous version of MOBILE, where it was
noted that resting loss emissions are independent of the canister
state (i.e., whether the canister is saturated or fully purged).
-------
-27-
9.0 CHARACTERIZING 24-HOUR DIURNAL EMISSIONS
Diurnal evaporative emissions, like most other evaporative
emissions, are functions of both fuel volatility and temperature
which are themselves interdependent. The RVP is a measure of
vapor pressure (VP) at a single temperature, 100°F. The
Clausius-Clapeyron relationship was used to estimate the vapor
pressure at each temperature and for each of the fuels (RVPs of
6.8 and 9.0 psi) used in this testing program. (in Appendix B, we
illustrate how the Clausius-Clapeyron relationship can be used to
estimate a fuel's vapor pressure at each temperature if the fuel's RVP
i s known.)
To characterize the diurnal emissions, we again (see Section
8.0) identified the 57 vehicles in the EPA program that were
tested over a wide range of vapor pressures. These test vehicles
were distributed among 12 strata (of the 18 potential strata
identified in Section 6.5). Within each stratum, we then
attempted to regress the diurnal emissions against combinations
of fuel volatility and temperature.
A similar approach was attempted to characterize resting
loss emissions (see previous section) but had not been
successful. However, this approach produced more satisfactory
results in characterizing the diurnal emissions even in strata
that were sparsely tested. Most likely this difference was due
to the effect that the test-to-test variability was substantially
larger relative to the smaller resting loss emissions than to the
larger diurnal emissions. Therefore, any test-to-test
variability was less likely to hide patterns evidenced in the
diurnal emissions measurements.
For each RTD test, the Clausius-Clapeyron relationship was
used to estimate the vapor pressure at both the low and the high
temperatures. Using these estimates, we calculated both the
average of the low and the high vapor pressures, as well as the
difference between the low and the high vapor pressures (AVP)
(both measured in kPa). Multiplying these two quantities
together produced a single product term (VP*AVP) that
incorporates the parameters of the RTD test (i.e., both the
temperature cycle and the RVP of the fuel).
The use of this vapor pressure product term (to estimate
diurnal emissions) is a change from MOBILES that used as the
independent variable an "uncontrolled diurnal index" (UDI).
Comparing these two variables (as in the following table), we
find that they are closely related (linearly). Regressing the
values in that table gives us the equation:
Vapor Pressure Product Term = 260.774 + ( 409.919 * UDI )
with an R-squared value of 98.9 percent. Thus, the use of this
VP product term (as the independent variable in MOBILE6) not only
incorporates the parameters of the RTD test, but it is also
consistent with MOBILES.
-------
-28-
Sample Comparing Uncontrolled Diurnal Index (UDI)
To Vapor Pressure (VP) Product Term
RVP
(psi)
9.0
10.5
11.5
11.7
9.0
Low
Temp
60
60
60
60
72
High
Temp
84
84
84
84
96
UDI
1.0000
1 .4567
1.9581
2.0677
1.7448
VP
Product
655.07
888.99
1,063.50
1,100.14
968.66
The mean diurnal emissions (calculated in the previous
section by subtracting a daily resting loss value from the RTD
test results) were repeatedly regressed against a polynomial of
that product term of vapor pressures within each stratum. The
independent variable used in the regressions was either:
1) the product term (i.e., the average vapor pressures
times the difference of the vapor pressures) or
2) the square or cube of that product term (to allow for
possible non-linearity).
We also performed regressions using other combinations of
variables (including RVP). Some of which had improved
statistical "fits" to the test data. Although the equations that
we developed in this analysis are empirical (i.e., data driven)
models, we did impose two sets of restrictions. (The second set
of restrictions is discussed on pages 31 and 32.) The first set
contains the following three restrictions that were based on
engineering experience with diurnal emissions. Many of the
potential models were discarded due to their failure to meet this
set of additional theoretical requirements:
4 The diurnal emissions should decrease with a decreasing
fuel RVP (with all other parameters held constant).
4 The diurnal emissions should decrease with decreasing
temperature cycles (with all other parameters held
constant).
4 For each combination of fuel delivery system (i.e.,
fuel injected versus carbureted) and purge/pressure
category, the diurnal emissions should increase with
each successively older model year grouping (for each
combination of temperature cycle and fuel RVP).
-------
-29-
In each of those 12 strata, we generated both a nonlinear (i.e.,
quadratic) model and a linear model*. A two step process was
used to choose among those models:
1) We performed a visual inspection of the data. (This
approach, in and of itself, is not very precise, but we
wanted to make certain that the model selected would be
both reasonable and accurately represent the test
data.)
2) We compared the statistical parameters associated with
each of those regressions. (That is, we identified the
model that optimized: the F-ratio, the statistical
significance of the independent variable, and the R-
squared value.)
Seven separate strata required additional effort to meet
these three criteria (that were based on engineering experience):
4 the three strata of 1971-1979 model year carbureted
vehicles,
4 the 1980-1985 model year FI vehicles that passed the
pressure test, and
4 the three strata of 1986 and newer model year carbureted
vehicles.
For the 1971-1979 model year carbureted vehicles, we used a
modification of the equations that resulted from the analysis of
the 1980-85 model year carbureted vehicles. Specifically, we
used the same coefficients (i.e., the same corrections for
changes in temperature and RVP), but we modified the constant
terms so that the resulting equations would pass through the
means of the actual (validating) data of the Pre-1980 vehicles.
The stratum of 1980-85 FI vehicle that passed both the purge
and pressure tests was represented by only a single vehicle that
was tested over the full range of temperature cycles and fuel
RVPs. Thus, the results of those tests were combined with the
tests on the three 1980-85 FI vehicles that failed the purge test
but passed the pressure test into a single stratum of vehicles
that passed the pressure test (represented by four vehicles).
The regression of these data was used to determine the
Theoretically, in each of those models, a zero change in daily temperature
(hence, in AVP) should result in zero diurnal emissions. This physical
necessity would result in the constant term in each regression being zero.
However, this requirement was dropped due to:
(1) the resulting low r-squared values,
(2) the lack of test data having diurnal temperature ranges less than 24
degrees, and
(3) our requirement, that for any diurnal emissions to occur, a
difference between the daily high and low temperatures was needed.
-------
-30-
coefficients for both the stratum of 1980-85 FI vehicles that
passed both the purge and pressure tests and the stratum of 1980-
85 FI vehicles that failed only the purge test. The constant
term for each stratum was the value that would make the resulting
equations would pass through the respective means of the actual
(validating) data of the 1980-85 FI vehicles (i.e., cause the
sums of the residuals to equal zero).
The last three problem strata were the 1986 and newer
carbureted vehicles. As is illustrated in Appendix C, only four
combinations of temperature cycle and fuel RVP were tested (in
each of the three purge/pressure substrata). The two untested
combinations were the combinations that would have yielded
results at the highest and the lowest VP values. Having test
data over such a narrow range (i.e., only the four middle values)
of vapor pressures made selecting the proper regression curve
difficult.
We first, therefore, attempted to enlarge the scope of the
data by estimating the diurnal emissions at the two missing
extreme values. We did this by observing that the diurnal
emissions of the 1986-95 carbureted vehicles (at the four tested
combinations of fuel RVP and temperature cycle) were between the
corresponding diurnal emissions of the 1986-95 FI vehicles and
the 1980-85 carbureted vehicles for each tested combination of
fuel RVP, temperature cycle, and purge/pressure result. If this
pattern were to hold true for the two untested combinations, then
the diurnal emissions of the 1986-95 carbureted vehicles would
be:
4 for tests using 6.8 RVP fuel over the 60-86 °F cycle:
44 between 4.815 and 9.519 for vehicles failing the
pressure test,
44 between 4.372 and 5.100 for vehicles failing only
the purge test, and
44 between 0.187 and 2.976 for vehicles passing both
the pressure and the purge tests.
4 for tests using 9.0 RVP fuel over the 82-106 °F cycle:
44 between 28.26 and 45.456 for vehicles failing the
pressure test,
44 between 21.046 and 50.67 for vehicles failing only
the purge test, and
44 between 9.932 and 36.565 for vehicles passing both
the pressure and the purge tests.
We then experimented, using the tested values for the 1986-95
carbureted vehicles with the coefficients determined for the
1980-85 carbureted vehicles and for the 1986-95 FI vehicles to
determine which set would more closely predict the preceding
estimates of the untested configurations. While neither set was
-------
-31-
perfect, the coefficients developed for the 1980-85 carbureted
vehicles came closer to the theoretical values and were selected.
The statistics associated with those regressions are given
in Appendix E. Once the coefficient values of the equation were
determined for each of the 15 strata, we again (as with the both
the Pre-80 vehicles and the 1980-85 FI vehicles) modified the
constant term (for each stratum) to minimize the sum of the
differences between the predicted and calculated diurnal
emissions. The resulting equations are given in Appendix F.
(The coefficients, but not the constant terms, from Appendix E
match those in Appendix F.) Graphical comparisons between the
predictions of those models (i.e., resulting equations) and the
means of the measured RTD test data are given in Appendix G.
In the five strata in which the vehicles passed both the
purge test and the pressure test, the data strongly suggest a
non-linear relationship (i.e., quadratic) between the diurnal
emissions and that "vapor pressure product" term. In the various
strata containing vehicles that failed either the purge or
pressure (or both) tests, the relationship between diurnal
emissions and the vapor pressure product term was sometimes
linear and sometimes non-linear.
On page 28, we noted that two sets of restrictions were
applied to the equations (in Appendix F) that predict diurnal
emissions produced by the regressions in Appendix E. The second
set of restrictions is intended to avoid having unrealistic
predictions when the model extrapolates beyond the limits of the
actual test data. (For example, although no RTD testing was
conducted with a test fuel having an RVP over 9.0 psi, MOBILE6
will produce estimates for an RVP as high as 15.2.)
MOBILE6 attempts to avoid unrealistic estimates by limiting
(i.e., setting "caps" for) the diurnal emissions. These limits
are based on a theoretical approach that is validated by the
means of the observational data in Appendix C. Specifically, we
reasoned that (with all other conditions being the same):
4 Among vehicles with defective evaporative control
systems, those vehicles with severe leaks of liquid
gasoline (GLLs) were likely to have the highest diurnal
emissions. (This was the observed result for all of
the actual tests. The restriction extended this to all
combinations of temperature cycles and fuel RVPs.)
4 The mean diurnal emissions from vehicles with properly
functioning evaporative control systems ("Pass Both")
were likely to be no higher than those from vehicles
with defective control systems. (Again, this
restriction extended this observation to all
combinations of temperature cycles and fuel RVPs.)
4 Among the non-GLL vehicles with defective evaporative
control systems, those vehicles with pressure leaks
(fail pressure) were likely to have higher diurnal
-------
-32-
emissions than those from vehicles with only defective
purge system (fail purge). (Again, this restriction
extended this observation to all combinations of
temperature cycles and fuel RVPs.)
MOBILE6 implements these three additional restrictions (for each
combination of temperature cycle, fuel RVP, model year, and fuel
delivery system) by:
1) limiting the diurnal emissions from the vehicles that
failed the pressure test to those from the GLLs
"Fail Pressure" = Min ["Fail Pressure" , "GLL"]
2) limiting the diurnal emissions from the vehicles that
failed the purge test to those from vehicles that
failed the pressure test
"Fail Purge" = Min ["Fail Pressure" , "Fail Purge"]
3) limiting the diurnal emissions from vehicles with
properly functioning evaporative control systems to
those from vehicles that failed the purge test
"Pass Both" = Min ["Pass Both" , "Fail Purge"]
10.0 GROSS LIQUID LEAKERS
Three issues related to vehicles with gross liquid leaks
need to be addressed:
1) the frequency of the occurrence of gross liquid leakers
(possibly as a function of vehicle age),
2) the magnitude of the emissions from gross liquid
leakers, and
3) the effects of changes in vapor pressure on the diurnal
and resting loss emissions of these gross liquid
leakers.
Analyses of these issues were hampered by a lack of a substantial
number of identified gross liquid leakers. We anticipate
revising the following initial estimates for future models based
on additional data.
10.1 Frequency of Gross Liquid Leakers
In a parallel report (M6.EVP.009, entitled "Evaporative
Emissions of Gross Liquid Leakers in MOBILE6"), EPA used the
results from a test fleet of 270 vehicles (i.e., combined EPA and
CRC samples) to estimate the occurrence of gross liquid leakers
within each of the three model year ranges used in the
recruitment process (the pre-1980, 1980-85, and 1986-95). The
estimated rate of occurrence of the "gross liquid leakers" (at
each of three given ages) is reproduced in the following table
-------
-33-
(Table 10-1, below). The large confidence intervals are the
result of the relatively small sample sizes.
Table 10-1
Frequency of Gross Liquid Leakers
(Based on RTD Testing)
Vehicle
Age (years)
6.12
13.00
21.79
Sample
Size
85
50
51
Frequency at
Vehicle Age
0.20%
2.00%
7.84%
Standard
Deviation
1.41%
1.98%
3.76%
90% Confide
Lower
0.00%
0.00%
1.65%
mce 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 early July (the median date for the
testing).
EPA then found (see Section 3.2 of M6.EVP.009) a logistic
growth curve that closely approximates these three values while
taking into account similar occurrences of "gross liquid leakers"
identified using the hot soak test and the running loss test.
The equation that EPA will use (in MOBILE6) to estimate the
frequency of gross liquid leakers (on the RTD test) is:
Rate of Gross Liguid Leakers
Based on RTD/Resting Loss Testing
0.0865
1 + 55 * exp[-0.259 * AGE]
Plotting this curve and the preceding set of three failure rates
(from Table 10-1) produces Figure 10-1 (on the following page).
A logistic curve that produces an improved "fit" of the values in
Table 10-1 can be obtained (see Section 3.1 of M6.EVP.009) by
reducing (or eliminating) the interdependence with the "gross
liquid leakers" identified using the hot soak test or the running
loss test. However, EPA will use the preceding equation in
MOBILE6.
-------
-34-
Figure 10-1
Frequency of Gross Liquid Leakers
(based on RTD testing)
15%
10%
5%
0%
10 20
Vehicle Age (years)
30
The solid line in Figure 10-1 is the logistic growth
function. Also graphed in that figure are the 90 percent
confidence intervals (as dotted lines) from Table 10-1. Since
the overall effect of the gross liquid leakers is the product (by
model year) of the percentage of gross liquid leakers and the
number of vehicles in the in-use fleet for that model year, the
rapidly increasing proportion of gross liquid leakers in the in-
use fleet tends to be offset by the decreasing number of older
vehicles in the in-use fleet. This graph (as well as the
preceding equation) predicts:
4 Fewer than one-half a percent of vehicles (at each age) up
to eight years of age will be "gross liquid leakers."
4 "Gross liquid leakers" do not reach one percent of any age
group of the in-use fleet until the vehicles exceed 10
years of age.
4 "Gross liquid leakers" reach (or exceed) two percent of
each age group of the in-use fleet for vehicles exceeding
13 years of age.
4 The portion of the fleet that is "gross liquid leakers"
then rises by vehicle age (almost linearly) to about eight
percent for vehicles that are 22 years old.
4 The increase in the frequency of "gross liquid leakers"
then levels off and the frequency approaches just over
eight and one-half percent (about age 30).
-------
-35-
It is important to note that this model of the frequency of
gross liquid leakers is based on the assumption that modern
technology vehicles will show the same tendency toward gross
liquid leaks as do the older technology vehicles at the same age.
However, if the modern technology vehicles were to 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, e.g., fuel tank protection and
elimination of fuel line leaks), the effect would be to replace
that single logistic growth function with a family of two or more
curves. (This assumed lower rate is used to create a different
curve for the 1999 and newer vehicles. See Section 11.7.)
Since EPA has no data to indicate that the multiple curve
scenario is the correct approach, EPA will use the single curve
approach to estimate the occurrence in the in-use fleet of these
vehicles that have substantial leaks of liquid gasoline (i.e.,
"gross liquid leakers").
10.2 Magnitude of Emissions from Gross Liquid Leakers
In that concurrent report on "gross liquid leakers"
(Document Number M6.EVP.009), EPA used the RTD test results from
ten (10) vehicles to estimate the mean diurnal emissions from the
stratum of "gross liquid leakers." Each of these 10 vehicles:
4 had diurnal emissions (RTD minus resting loss) of at least
15 grams per day
4 had an observed liquid leak
4 but, were not necessarily "gross liquid leakers."
EPA then assumed that the distribution of the diurnal emissions
from these leaking (but not necessarily "gross liquid leakers")
vehicles was lognormal (i.e., the logarithms of the emissions,
rather than the emissions themselves, are assumed to be normally
distributed). (Distributions other than the lognormal were
examined, but none came as close to approximating the observed
distribution.) That lognormal distribution was then used to
estimate the frequency associated with each possible diurnal
emission level. 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. (See Section 2.1
of M6.EVP.009 for details.)
EPA will use 104.36 grams per day as the average full-day's
diurnal emissions from "gross liquid leakers" over a day for
which the maximum daily temperature is exactly 24°F above the
daily low temperature. In report number M6.EVP.002, EPA derives
an equation to estimate full-day diurnal emissions over different
temperature cycles (having a difference between the daily high
and low temperatures of at least 10 degrees Fahrenheit) as:
-------
-36-
Total 24-Hour Diurnal Emissions (grams)
= 40.5533 + ( 2.658611 * Diurnal_Temperature_Range )
Where the Diurnal_Temperature_Range is the difference of the daily high
temperature minus the daily low temperature.
Note, that equation predicts a 24-hour total diurnal
emission of 40.48 grams for a day during which the temperatures
do not change. This is not reasonable since diurnal emissions
result from the daily rise in ambient temperatures. Therefore,
EPA will set the 24-hour diurnal equal to zero for a diurnal
temperature range of zero degrees Fahrenheit. For a diurnal
temperature range of exactly ten degrees Fahrenheit, the equation
predicts the 24-hour diurnal for gross liquid leakers to be
67.011 grams. If daily temperature range is between zero and 10
degrees, then EPA will interpolate, producing:
Total 24-Hour Diurnal Emissions (grams) = 6.701075 * Diurnal_Temperature_Range
Earlier versions of MOBILE limited the pressure driven leaks
(i.e., 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.
The preceding approach was repeated for resting loss
emissions. (Again, see Section 2.1 of M6.EVP.009 for details.)
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.
EPA will use 9.16 grams per hour as the average hourly
resting loss emissions from "gross liquid leakers."
On page 26, we noted that the daily resting loss emissions
(assuming a daily temperature profile similar to those in
Appendix A) would be 24 times the hourly resting loss (at the
lowest temperature of the day) plus 0.766. Since including the
0.766 term will increase the day's total resting loss (from
"gross liquid leakers") less than 0.4 percent, and since the
mechanism responsible for the vast majority of the resting loss
emissions from these vehicles is the fuel leaking out of the
vehicle which is not dependent upon the ambient temperature or
fuel volatility, we will assume the resting loss emissions from
"gross liquid leakers" are completely independent of temperature
(see Section 11.1). Therefore, based on the means in the
preceding table, EPA will use, in MOBILE6, for the category of
gross liquid leakers:
• DAILY RESTING Loss = ( 24 * HOURLY RESTING Loss )
( 24 * 9.16 ) = 219.84 (GRAMS / DAY)
and
-------
-37-
• Full-Day's DIURNAL = MEAN RTD - DAILY RESTING Loss
104.36 (GRAMS / DAY )
Thus, while the occurrence of these gross liquid leakers is
relatively rare among newer vehicles (Section 10.1), their
presence has a substantial effect on the total resting loss and
diurnal emissions of the in-use fleet.
10.3 Effects of Vapor Pressure Changes on Gross Liquid Leakers
As previously discussed, the true vapor pressure is a
function of both the ambient temperature and the Reid vapor
pressure of the fuel. Since only two of the seven vehicles that
have been identified as gross liquid leakers were tested over a
range of fuel RVPs, there are not enough data to relate changes
in diurnal and resting loss emissions to changes in fuel RVP.
However, as noted in the preceding section, changes in fuel RVP
are expected to have only minimal (proportional) effects on the
total diurnal and resting loss emissions of vehicles whose
primary mechanism of evaporative emissions is leaking liquid
gasoline. Thus, until additional data are available, EPA will
treat the diurnal and resting loss emissions of the gross liquid
leakers as independent of fuel RVP.
In the previous section, EPA treated the hourly resting
emissions of these gross liquid leakers as if they are
independent of ambient temperature as well. In a concurrent
report (document number M6.EVP.002, entitled "Modeling Hourly
Diurnal Emissions and Interrupted Diurnal Emissions Based on
Real-Time Diurnal Data"), EPA was able to use the hourly diurnal
emissions to estimate the effects of temperature changes on the
diurnal emissions of these gross liquid leakers. That report
concludes that the full-day's diurnal emissions of gross liquid
leakers is dependent only upon the daily temperature range (i.e.,
the difference between the daily high and low temperatures).
Thus, for any of the three temperature cycles in Appendix A, the
mean of the full-day's diurnal emissions of gross liquid leakers
is the constant 104.36 grams (calculated in the previous
section).
Therefore, EPA is proposing that both the hourly resting
loss emissions and full-day's diurnal emissions of gross liquid
leakers are independent of vapor pressure for each of the three
temperature cycles in Appendix A.
11.0 Other Topics
Several topics were not discussed in the preceding analysis
because either:
1) They will be discussed in forthcoming reports.
or
-------
-38-
2) No changes are planned in how they were handled in
MOBILES.
11.1 Temperature Ranges
All of the tests used in this analysis were performed using
one of the three temperature cycles in Appendix A. This results
in all of the resting loss emissions being measured at only three
temperatures (i.e., 60, 72, and 82 °F). In Section 8, we
developed regression equations to estimate hourly resting loss
emissions at theoretically any temperature. We will limit that
potentially infinite temperature range as we did in the previous
version of MOBILE, specifically:
1) We will assume, for light-duty vehicles other than gross
liquid leakers, there are no resting loss emissions when the
temperatures are below or equal to 40°F. (This assumption
was used consistently for all evaporative emissions in
MOBILES.)
2) We will assume, for light-duty vehicles other than gross
liquid leakers, that when the ambient temperatures are above
105°F that the resting loss emissions are the same as those
calculated at 105°F.
Since vehicles classified as gross liquid leakers were not
handled separately in MOBILES, we will now make a new assumption
concerning those vehicles' emission performance as relates to
temperatures. Specifically:
3) For the vehicles classified as gross liquid leakers, we will
assume the resting loss emissions are completely independent
of temperature, averaging 9.16 grams per hour.
The equations developed in this report to estimate full-day
diurnal emissions theoretically could also be applied to any
temperature cycle. We will limit those functions by assuming
that the 24-hour diurnal emissions will be zero for any
temperature cycle in which the difference between the daily high
and low temperatures (i.e., the "diurnal temperature range") is
zero degrees Fahrenheit (i.e., constant daily temperature). As
with the "gross liquid leakers," if the daily temperature range
is between zero and 10 degrees, we will interpolate.
11.2 Heavv-Dutv Gasoline-Fueled Vehicles (HDGVs)
The analyses in this report were based on RTD tests of only
light-duty gasoline-fueled vehicles (LDGVs) and light-duty
gasoline-fueled trucks (LDGTs). Since the data did not indicate
a significant difference between the RTD emissions from LDGVs and
LDGTs, they were combined in a single group for analyses.
Since no RTD testing was performed on any HDGVs, we will use
the same approach that was used in the earlier version of MOBILE.
That is, the ratio of diurnal emissions of the HDGVs to those of
-------
-39-
the LDGTs is proportional to the corresponding ratios of the
evaporative emission standards. For each strata of HDGV (2b, 3,
4, 5, 6, 7, 8a, 8b and heavy-duty gas busses) that are not "Gross
Liquid Leakers," we will assume that their "full day" diurnal
emissions will be a multiple of the "full day" emissions of the
corresponding strata of LDGTs (or LDGVs since the emissions are
the same). Therefore, the heavy-duty vehicles in classes lib and
3 will have evaporative emissions of 1.5 times the evaporative
emissions of the corresponding LDGT strata (determined by fuel
metering and purge/pressure tests). And, the heavy-duty vehicles
in classes 4 through 8 plus busses will have evaporative
emissions of 2.0 times the evaporative emissions of the
corresponding LDGT strata.
Translating these assumptions into equations yields the
following cases:
4 For model years prior to 1985, the diurnal emissions from
HDGVs were uncontrolled. Therefore, for all model years
prior to 1985, we will apply that multiplier to only the
LDTs (of the appropriate fuel delivery system) that failed
the pressure test (for each model year). That is, for all
of the pre-1985 heavy-duty trucks that are not gross liquid
leakers, we will assume that their full day diurnal
emissions are a simple multiple (1.5 or 2.0) of the light-
duty trucks of that model year that also fail the pressure
test.
4 For model years prior to 1979, trucks with gross vehicle
weight ratings (GVWR) between 6,000 and 8,500 pounds were
considered to be heavy-duty trucks. Therefore, MOBILE6 will
set their evaporative emissions equal to those of the trucks
with GVWR between 8,500 and 10,000 pounds (HDGV-2b). These
vehicles are identified in MOBILE6 as LDGT-3 and LDGT-4.
4 For the 1985 and newer model years, we will apply that
multiplier to each purge/pressure fuel-delivery stratum of
the LDTs (for each model year).
We will use the same formulas for resting losses (obviously
changing to "diurnal emissions" to "hourly resting losses").
11.3 High Altitude Evaporative Emissions
We will continue to use the multiplicative adjustment factor
of 1.30 (from previous version of MOBILE) to adjust both the
resting loss and diurnal emissions for high altitude for all
vehicles that are not "gross liquid leakers." For the "gross
liquid leakers," we will assume that there is no difference in
either resting loss or diurnal emissions between low and high
altitude.
-------
-40-
11.4 Motorcycles (MC)
RTD evaporative emission tests were not performed on
motorcycles (MC). In MOBILES, the resting loss and diurnal
emissions from motorcycles were modeled using carbureted vehicles
equipped with open-bottom canisters. That approach will continue
with MOBILE6.
We first identified 109 RTD tests of carbureted vehicles
equipped with open-bottom canisters (all 1988 or earlier model
years), and calculated both the hourly resting loss (associated
with the test's low temperature) and the full-day's diurnal for
each of those 109 tests. The diurnal emissions were then
regressed against both the vapor pressure product term (developed
in Section 9) and the age of each test vehicle. As illustrated
in Table 11-1, each of those variables is statistically
significant. MOBILE6 will use the linear regression equation
generated by that analysis to calculate the full day's diurnal
emissions.
Table 11-1
Regression of Diurnal Emissions
(Simulated Motorcycle Fleet)
Dependent variable is
No Selector
R squared = 59.0%
s= 10.20 with 109-
Source
Regression
Residual
Variable
Constant
age
VP Product
R squared (adjusted)
= 58.3%
Diurnal
3 = 106 degrees of freedom
Sum of Squares
15892.9
11024.5
Coefficient
-36.7971
0.855491
0.058251
df
2
106
s.e. of Coeff
4.5620
0.1894
0.0051
Mean Square
7946.46
104.005
t-ratio
-8.07
4.52
11.5
F-ratio
76.4
prob
< 0.0001
< 0.0001
< 0.0001
Translating that regression analysis into an equation yields:
24-Hour Diurnal Emissions (grams) of Motorcycles
= -36.7971 + ( 0.855491 * Vehicle_Age ) + ( 0.058251 * VP_Product_Term )
EPA will use this equation to estimate the 24-hour diurnal
emissions from motorcycles.
-------
-41-
Similarly, the hourly resting loss emissions were regressed
against both the temperature at which those values were
calculated (i.e., the daily low temperature) and the age of each
test vehicle. As illustrated in Table 11-2, only the vehicle age
is statistically significant. It is possible that temperature
was not found to be statistically significant simply due to the
fact that most of the resting loss emissions were calculated at
the same temperature (72 °F). Since resting loss emissions
should be an increasing function of temperature, EPA will use for
MOBILE6 the linear regression equation generated by the analysis
(in Table 11-2) that uses both variables (despite the low
statistical significance).
Table 11-2
Regression of Hourly Resting Loss Emissions
(Simulated Motorcycle Fleet)
Dependent variable is:
No Selector
R squared = 5.6% R
s= 0.1346 with 109-
Source
Regression
Residual
Variable
Constant
age
Temperature
squared (adjusted)
= 3.8%
Hourly
Resting Loss
3 = 106 degrees of freedom
Sum of Squares
0.114078
1.92123
Coefficient
0.044345
0.006134
0.000859
df
2
106
s.e. of Coeff
0.1572
0.0025
0.0022
Mean Square
0.057039
0.018125
t-ratio
0.282
2.45
0.399
F-ratio
3.15
prob
0.7784
0.0159
0.6909
Translating this regression analysis into an equation yields:
Hourly Resting Loss Emissions (grams) for Motorcycles
= 0.044345 + (0.006134 * Vehicle_Age ) + ( 0.000859 * Hourly_Temperature )
EPA will use this equation to estimate the hourly resting loss
emissions from non-leaking motorcycles at temperatures between 40
and 105 degrees Fahrenheit (see Section 11.1) .
11.5 Pre-Control Vehicles
Non-California vehicles prior to the 1971 model year were
not required to meet an evaporative emission standard. These
uncontrolled vehicles would simply vent vapors to the atmosphere
as pressure built up. Since that situation is similar to that of
a controlled vehicle with a vapor leak, we hypothesized that the
resting loss and diurnal evaporative emissions of the pre-1971
-------
-42-
vehicles would be comparable to the emissions of the pre-1980
vehicles that had failed the pressure test.
To characterize the hourly resting loss emissions from these
pre-control vehicles, we proceeded in a similar fashion to the
approach in Section 8. We first identified the two pre-1980
vehicles in our study that both had failed the pressure test and
were tested over the full range of fuels and temperature cycles.
Possibly due to that small sample size, a regression of those
data produced a slope of resting loss versus temperature that was
not statistically different from zero. We, therefore, decided to
use the same slope (0.002812) that was developed in Section 8.
Since most of the RTD tests (i.e., 37 of 47) that were performed
on the 34 candidate vehicles were run over the same temperature
cycle (i.e., 72 to 96 degrees), the variable "temperature" would
not make a useful independent variable to analyze those 47
resting loss results. However, the variable "age" was found to
be statistically significant. Combining the results of
regressing the data against age with the previously calculated
temperature slope yields the following equation:
Hourly Resting Loss (grams) = -0.768438
+ (0.002812 * Temperature in °F )
+ ( 0.040528 * Vehicle Age in Years )
EPA will use this equation to estimate the hourly resting loss
emissions from pre-control vehicles with the restriction that the
calculated value must be at least the estimated hourly resting
loss of the (newer) 1971-79 model year vehicles (as calculated in
Appendix D).
To characterize the full day's diurnal emissions from these
pre-control vehicles, we proceeded in a similar fashion to the
approach in Section 9. In the preceding paragraph we noted that
only two of the candidate vehicles (i.e., pre-1980 vehicles that
failed the pressure test) were tested over the full range of
fuels and temperature cycles. Attempting to analyze the resting
loss emissions of those two vehicles as a function of temperature
produced only mediocre results. However, the corresponding
analysis for diurnal emissions as a function of the vapor
pressure product term produced satisfactory results, as shown in
Table 11-3 (on the following page).
-------
-43-
Table 11-3
Regression of Diurnal Emissions
(Simulated Pre-Control Fleet)
(Based on Two Vehicles)
Dependent variable
No Selector
R squared = 92.3%
s = 5.503 with 6 -
Source
Regression
Residual
Variable
Constant
VP_Product
is:
R squared (adjusted) =
2 = 4 degrees of freedom
Sum of Squares
1456.41
121.136
Coefficient s
-6.52265
0.05115
90.4%
df
1
4
.e. of Coeff
6.175
0.0074
Mean Square
1456.41
30.284
t-ratio
-1.06
6.93
Diurnal
F-ratio
48.1
prob
0.3504
0.0023
Similar to the statements in the preceding material on the
resting loss emissions from these test vehicles, the diurnal
emissions from these tests are almost exclusively from tests
performed over the 72 to 96 degree temperature cycle using a
single fuel RVP. Thus, using a variable for vapor pressure for
the full set of 47 tests would not be productive. However, as
with the resting loss emissions, we used the preceding
coefficient (0.05115) to estimate diurnal emissions (based on the
vapor pressures) and then regress the calculated residuals
against vehicle age. That regression analysis yields the
following equation:
24-Hour Diurnal (grams) = -40.67512
+ (0.05115 * VP Product Term )
+ (1.41114 * Vehicle Age in Years )
EPA will use this equation to estimate the 24-hour diurnal
emissions from pre-control vehicles with the restriction that the
calculated value must be at least the estimated full-day's
diurnal of the (newer) 1971-79 model year vehicles (as calculated
in Appendix E).
11.6 Duration of Diurnal Soak Period
The analyses in this report were based on diurnals of
exactly 24 hours in length. In the real-world, the soak period
could run for longer or shorter periods of time.
Estimating diurnal emissions when the soak period is less
than 24 hours are analyzed in report number M6.EVP.002 (entitled
-------
-44-
"Modeling Hourly Diurnal Emissions and Interrupted Diurnal
Emissions Based on Real-Time Diurnal Data").
Estimating diurnal emissions when the soak period is more
than 24 hours are analyzed in report number M6.EVP.003 (entitled
"Evaluating Multiday Diurnal Evaporative Emissions Using RTD
Tests") .
11.7 1996 and Newer Model Year Vehicles
Starting with the 1996 model year, EPA began certifying some
of the new LDGVs, LDGTs, and HDGVs using an enhanced test
procedure (ETP) which includes the RTD test. (The phase-in
continued through the 1998 model year. By the 1999 model year,
all the vehicles in the affected classes were ETP vehicles.) The
methods used by EPA to estimate the resting loss and diurnal
emissions from these vehicles are detailed in report number
M6.EVP.005 (entitled "Modeling Diurnal and Resting Loss Emission
from Vehicles Certified to the Enhanced Evaporative Standards").
Summarizing the results in that report, EPA found that (for ETP
vehicles that are not "gross liquid leakers") the diurnal
emissions predicted by those analyses:
• approximated the corresponding pre-enhanced (i.e., 1990-95),
for the ETP vehicles that failed either the purge test or
the pressure test and
• were approximately one-half the corresponding pre-enhanced
(i.e., 1990-95), for the ETP vehicles that passed both the
purge test and the pressure test.
These results support the assumptions made by EPA at the time the
ETP rules were proposed; therefore, EPA will continue to use
those assumptions in MOBILE6. That is, EPA will use a single set
of equations to predict the diurnal emissions from the 1986 and
newer vehicles that fail either the purge test or the pressure
test, similarly for the gross liquid leakers. Also, for the ETP
vehicles that pass both the purge and the pressure tests, EPA
will estimate their diurnal emissions to be exactly one-half the
diurnal emissions of the 1986-95 model year vehicles that also
pass both the purge and the pressure tests and are not gross
liquid leakers.
EPA will (in MOBILE6) also continue using the assumption
that the resting loss emissions of the (ETP) vehicles that are
not gross liquid leakers will be reduced by 75 percent.
11.8 Tier-2 (2004 and Newer Model Year) Vehicles
Beginning with the 2004 model year, vehicles that meet the
more stringent Tier-2 standards will begin to be phased-in.
Estimating the effects of those requirements for MOBILE6 is
discussed in two parallel reports: "Modeling Diurnal and Resting
Loss Emissions from Vehicles Certified to the Enhanced
Evaporative Standards" (report M6.EVP.005) and "Accounting for
-------
-45-
the Tier 2 and Heavy-Duty 2005/2007 Requirements in MOBILE6"
(report M6.EXH.007). MOBILE6 will estimate the diurnal emissions
of those Tier-2 vehicles that are not "gross liquid leakers" and
that pass both the purge and pressure tests:
• For passenger cars (i.e., LDGVs), we will assume that the
diurnal emissions will be reduced by 0.75 (compared to the
pre-Tier-2 ETP vehicles in Section 11.7).
• For LDGT-1 and LDGT-2 (GVWR <, 6000) , we will assume that the
diurnal emissions will be reduced by 0.675 (compared to the
pre-Tier-2 ETP vehicles in Section 11.7).
• For LDGT-3 and 4s (6000 < GVWR <. 8500) , we will assume that
the diurnal emissions will be reduced by 0.525 (compared to
the similar pre-Tier-2 ETP trucks in Section 11.7).
• For HDGTs with GVWR up to 14,000, emissions will be 1.474
times the corresponding emissions of the Tier-2 LDGTs with
GVWR from 6,001 to 8,500 (i.e., proportional to the
certification standards).
• For HDGTs (GVWR > 8500), we will assume that the diurnal
emissions will be 2.000 times the corresponding emissions of
the Tier-2 LDGTs with GVWR from 6,001 to 8,500 (i.e.,
proportional to the certification standards).
EPA will assume that for Tier-2 vehicles that fail either the
pressure test or the purge test, their diurnal (and resting loss)
emissions will be the same as the corresponding pre-Tier-2
vehicles.
EPA will (in MOBILE6) assume that the resting loss emissions
of the affected (Tier-2) vehicles will exhibit the same percent
reduction as the diurnal emissions.
-------
-46-
Appendix A
Temperature Cycles (°F)
Hour
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
—Temper
60°-84°F
60.0
60.5
63.5
68.3
73.2
77.4
81.1
83.1
83.8
84.0
83.5
82.1
79.7
76.6
73.5
70.8
68.9
67.0
65.2
63.8
62.7
61.9
61.3
60.6
60.0
atures Cycling B<
72°-96°F*
72.0
72.5
75.5
80.3
85.2
89.4
93.1
95.1
95.8
96.0
95.5
94.1
91.7
88.6
85.5
82.8
80.9
79.0
77.2
75.8
74.7
73.9
73.3
72.6
72.0
jtween —
82°-106°F
82.0
82.5
85.5
90.3
95.2
99.4
103.1
105.1
105.8
106.0
105.5
104.1
101.7
98.6
95.5
92.8
90.9
89.0
87.2
85.8
84.7
83.9
83.3
82.6
82.0
Change in
Temperature
0.5
3.0
4.8
4.9
4.2
3.7
2.0
0.7
0.2
-0.5
-1.4
-2.4
-3.1
-3.1
-2.7
-1.9
-1.9
-1.8
-1.4
-1.1
-0.8
-0.6
-0.7
-0.6
* The temperature versus time values for the 72-to-96 cycle are
reproduced from Table 1 of Appendix II of 40CFR86.
These three temperature cycles are parallel (i.e., identical
hourly increases/decreases). The temperatures peak at hour nine.
The most rapid increase in temperatures occurs during the third
and fourth hours.
For cycles in excess of 24 hours, the pattern is repeated.
-------
-47-
Appendix B
Vapor Pressure
Using the Glausius-Clapeyron Relationship
The Clausius-Clapeyron relationship assumes that the
logarithm of the vapor pressure is a linear function of the
reciprocal (absolute) temperature. This relationship is a
reasonable estimate of vapor pressure (VP) over the moderate
temperature ranges* (i.e., 60° to 106°F) that are being
considered for adjusting the diurnal emissions.
In an earlier EPA work assignment, test fuels having RVPs
similar to those used in EPA's RTD work assignments were tested,
and their vapor pressures (in kilo Pascals) at three different
temperatures were measured. The results of those measurements
are given below in the following table:
Nominal
RVP
7.0
9.0
Measured
RVP
7.1
8.7
Vapor Pressure (kPa)
75° F
30.7
38.2
100°F**
49.3
60.1
130°F
80.3
96.5
** The VPs at 100° F are the fuel RVPs (in kilo Pascals).
Plotting these six vapor pressures (using a logarithm scale for
the vapor pressure) yields the graph (Figure B-l) on the
following page.
For each of those two RVP fuels, the Clausius-Clapeyron
relationship estimates that, for temperature in degrees Kelvin,
the vapor pressure (VP) in kPa will be:
Ln(VP) = A + (B / Absolute Temperature), where:
RVP A B
8.7
7.1
13.5791
13.7338
-2950.47
-3060.95
C. Lindhjem and D. Korotney, "Running Loss Emissions from Gasoline-Fueled
Motor Vehicles", SAE Paper 931991, 1993.
-------
-48-
Figure B-l
Comparison of Vapor Pressure to Temperature
100
10
0.0030
0.0031 0.0032
Reciprocal of Temp (1/°K)
0.0033
0.0034
Since MOBILE6 will estimate diurnal emissions by using the
vapor pressure of the typical (local) fuel at two temperatures
(the daily low and high temperatures), we need to create a
similar VP curve for any local fuel. Since that curve is a
straight line (in log-space), all we need is the vapor pressure
of the local fuel measured at two different temperatures. (That
is, two points determine a straight line.) Unfortunately, all we
usually have available is the Reid vapor pressure (RVP) which is
the VP at 100 degrees Fahrenheit. To obtain a second point (to
determine the VP curve), EPA will use the preceding graph (Figure
B-l). In that graph the two lines are not parallel, they
intersect at a point. (That point of intersection has meaning
only in a mathematical context. In an engineering context, both
the temperature (825.8 °F) and VP (12,679 kPA) are so high as to
be meaningless. This point would correspond to the "point at
infinity" in perspective drawings.)
Combining the reported VP of the fuel at 100 degrees
Fahrenheit (i.e., RVP) with this artificial VP value at 825.8
degrees Fahrenheit, we obtain the linear equation:
Ln(VP) = A + ( B / Absolute Temperature), where:
B = -3565.2707 + ( 70.5114 * RVP )
and
A = Ln( 6.89286 * RVP ) - ( B / 310.9 )
Despite the artificial nature of that second point, this
equation accurately predicts the vapor pressure (in kPa) of the
-------
-49-
two test fuels (in Figure B-l) as well as producing reasonable
estimates for the range of fuels and temperatures modeled in
MOBILE6. Therefore, EPA will use this equation to estimate the
values of VP (that are used as an intermediate step in MOBILE6)
to predict the diurnal emissions.
-------
-50-
Appendix C
Mean Evaporative Emissions by Strata
By Vapor Pressure Products
Strata
Pre-1 980 Carbureted
Fail Purge/
Fail Pressure
Pre-1 980 Carbureted
Fail Purge/
Pass Pressure
Pre-1 980 Carbureted
Pass Purge/
Fail Pressure
Pre-1 980 Carbureted
Pass Purge/
Pass Pressure
1980-85 Carbureted
Fail Purge/
Fail Pressure
1980-85 Carbureted
Fail Purge/
Pass Pressure
Fuel
RVP
6.8
6.8
6.8
9.0
6.8
9.0
9.0
6.8
6.3
6.8
9.0
6.3
6.8
9.0
9.0
6.8
6.8
9.0
6.8
9.0
9.0
6.8
6.8
6.3
6.8
9.0
6.3
6.8
9.0
9.0
Temp.
Cvcle
72.TO.96
60.T0.84
72.TO.96
60.TO.84
82.T0.106
72.T0.96
82.TO.106
60.T0.84
72.Tg.96
72.Tg.96
60.T0.84
82.T0.106
82.Tg.106
72.Tg.96
82.TO.106
60.Tg.84
72.TO.96
60.T0.84
82.Tg.106
72.Tg.96
82.T0.106
72.T0.96
60.T0.84
72.Tg.96
72.Tg.96
60.T0.84
82.T0.106
82.Tg.106
72.Tg.96
82.T0.106
VP
times
AVP
567.02
374.77
567,02
655,07
789.30
968.66
1323.87
374.77
489,32
567,02
655.07
683.98
789.30
968.66
1323.87
374,77
567.02
655.07
789.30
968.66
1323.87
567.02
374.77
489,32
567,02
655.07
683.98
789.30
968.66
1323.87
Count
13
1
7
1
1
1
1
2
1
20
3
1
2
3
2
1
11
1
1
1
1
1
3
1
11
4
1
3
4
3
Mean
RTD
Test
Results
25.111
16.229
21.055
17.511
36.321
44.222
76.801
21.284
17.426
24.385
21.572
24.328
42.799
35.331
72.263
7.861
13.240
17.423
32.292
38.297
100.094
27.401
8.834
16.541
17.756
16.823
14.962
19.669
25.415
55.324
Mean
Hourly
Resting
Loss
0.452
0.250
0.307
0.218
0.204
0.250
0.259
0.238
0.140
0.227
0.103
0.175
0.174
0.107
0.274
0.167
0.263
0.239
0.293
0.204
0.062
0.265
0.124
0.185
0.163
0.172
0.146
0.169
0.163
0.162
-------
-51-
Mean Evaporative Emissions by Strata
By Vapor Pressure Products (continued)
Strata
1980-85 Carbureted
Pass Purge/
Fail Pressure
1980-85 Carbureted
Pass Purge/
Pass Pressure
1986+ Carbureted
Fail Purge/
Fail Pressure
1986+ Carbureted
Fail Purge/
Pass Pressure
1986+ Carbureted
Pass Purge/
Fail Pressure
1986+ Carbureted
Pass Purge/
Pass Pressure
1980-85 Fuel Injected
Fail Purge/
Fail Pressure
1980-85 Fuel Injected
Fail Purge/
Pass Pressure
Fuel
RVP
6.8
6.3
6.8
9.0
6.3
6.8
9.0
9.0
6.8
6.3
6.8
9.0
6.3
6.8
9.0
9.0
N/A
6.8
9.0
6.8
9.0
6.8
9.0
6.8
9.0
6.8
9.0
6.8
9.0
N/A
6.8
6.8
9.0
6.8
9.0
9.0
Temp.
Cvcle
60.TO.84
72.T0.96
72.TO.96
60.Tp.84
82.T0.106
82.T0.106
72.Tp.96
82.TO.106
60.Tp.84
72.Tp.96
72.T0.96
60.T0.84
82.Tp.106
82.Tp.106
72.T0.96
82.T0.106
N/A
72.Tp.96
60.Tp.84
82.T0.106
72.T0.96
72.T0.96
60.T0.84
82.Tp.106
72.TO.96
72.Tp.96
60.Tp.84
82.T0.106
72.T0.96
N/A
60.Tp.84
72.Tp.96
60.T0.84
82.T0.106
72.Tp.96
82.TO.106
VP
times
AVP
374.77
489.32
567,02
655,07
683.98
789.30
968,66
1323.87
374,77
489.32
567.02
655.07
683,98
789,30
968.66
1323.87
N/A
567,02
655,07
789.30
968.66
567.02
655.07
789.30
968.66
567,02
655,07
789.30
968.66
N/A
374,77
567,02
655.07
789.30
968,66
1323.87
Count
2
1
8
3
1
2
3
2
3
3
38
7
3
4
7
3
0
1
1
1
1
2
2
2
2
10
1
1
1
0
3
3
4
3
4
4
Mean
RTD
Test
Results
13.383
20.741
16.508
27.768
43.384
31.965
45.319
53.615
5.302
16.308
9.081
1 1 .352
22.047
14.999
21.089
43.900
N/A
10.230
12.840
25.720
17.670
15.865
21.765
21.480
26.265
9.481
6.440
8.630
8.140
N/A
4.329
7.910
6.556
10.744
11.506
26.730
Mean
Hourly
Resting
Loss
0.121
0.253
0.139
0.127
0.444
0.216
0.276
0.308
0.065
0.195
0.107
0.147
0.170
0.169
0.194
0.274
N/A
0.100
0.097
0.155
0.148
0.233
0.342
0.124
0.308
0.138
0.092
0.102
0.075
N/A
0.010
0.011
0.045
0.041
0.086
0.123
-------
-52-
Mean Evaporative Emissions by Strata
By Vapor Pressure Products (continued)
Strata
1980-85 Fuel Injected
Pass Purge/
Fail Pressure
1980-85 Fuel Injected
Pass Purge/
Pass Pressure
1986+ Fuel Injected
Fail Purge/
Fail Pressure
1986+ Fuel Injected
Fail Purge/
Pass Pressure
1986+ Fuel Injected
Pass Purge/
Fail Pressure
1986+ Fuel Injected
Pass Purge/
Pass Pressure
Fuel
RVP
6.8
6.8
9.0
6.8
9.0
9.0
6.8
6.8
9.0
6.8
9.0
9.0
N/A
6.3
6.8
6.3
6.8
9.0
6.3
6.8
9.0
9.0
6.3
6.8
6.3
6.8
9.0
6.3
6.8
9.0
9.0
6.3
6.8
6.3
6.8
9.0
6.3
6.8
9.0
9.0
Temp.
Cycle
60.TO.84
72.T0.96
60.T0.84
82.Tp.106
72.Tp.96
82.T0.106
60.Tp.84
72.T0.96
60.T0.84
82.Tp.106
72.Tp.96
82.T0.106
N/A
60.Tp.84
60.T0.84
72.T0.96
72.Tp.96
60.Tp.84
82.T0.106
82.T0.106
72.Tp.96
82.TO.106
60.Tp.84
60.Tp.84
72.T0.96
72.T0.96
60.Tp.84
82.Tp.106
82.T0.106
72.T0.96
82.TO.106
60.T0.84
60.Tp.84
72.Tp.96
72.T0.96
60.T0.84
82.Tp.106
82.Tp.106
72.T0.96
82.T0.106
VP
times
AVP
374,77
567.02
655.07
789,30
968,66
1323.87
374,77
567.02
655.07
789.30
968,66
1323.87
N/A
321,73
374.77
489.32
567,02
655,07
683.98
789.30
968,66
1323.87
321,73
374,77
489.32
567.02
655,07
683.98
789.30
968.66
1323.87
321.73
374,77
489,32
567.02
655.07
683,98
789,30
968.66
1323.87
Count
2
3
2
2
2
2
1
4
2
2
2
1
0
3
12
5
18
17
5
15
17
12
1
12
4
19
19
4
16
19
12
2
16
6
69
31
6
24
31
21
Mean
RTD
Results
19.624
19.482
25.861
39.424
39.065
50.255
12.943
8.541
7.845
11.861
13.330
25.503
N/A
3.002
5.413
6.027
9.083
7.802
1 1 .068
14.498
11.734
23.895
5.206
6.600
10.259
9.202
8.611
14.842
15.824
16.193
32.116
0.602
1.611
2.345
7.166
2.398
3.576
5.487
4.426
13.640
Mean
Hourly
Rst Lss
0.198
0.206
0.184
0.300
0.231
0.252
0.296
0.080
0.157
0.218
0.227
0.348
N/A
-0.009
0.011
0.024
0.060
0.034
0.064
0.073
0.056
0.087
0.037
0.042
0.038
0.094
0.053
0.088
0.110
0.114
0.129
-0.001
0.027
0.032
0.062
0.034
0.049
0.073
0.064
0.123
-------
-53-
Appendix D
Modeling Hourly Resting Loss Emissions
From Light-Duty Gas Vehicles and Trucks
As Functions of Temperature (°F)
In each of the following 12 strata, resting loss emissions (in grams per
hour) are modeled using a pair of numbers (A and B), where:
Hourly Resting Loss (grams) = A + ( B * Temperature in °F)
Where
B = 0.002812 (for ALL strata) and
"A" is given in the following table:
Fuel Delivery
Carbureted
Fuel Injected
Model Year
Range
Pre-1980
1980-1985
1986-1995
Pre-1980*
1980-1985
1986-1995
Pass Pressure
Test
0.05530
-0.05957
-0.07551
0.05530
-0.09867
-0.14067
Fail Pressure
Test
0.07454
-0.02163
0.05044
0.07454
0.02565
-0.10924
* The untested stratum (Pre-1980 FI vehicles) was
represented using the Pre-1980 model year carbureted
vehicles.
Calculating the 24 hourly resting loss emissions using any
temperature profile in which the hourly change in temperature is
proportional to the corresponding hourly changes in the cycles in
Appendix A, and then sum all of the 24 hourly results, we find:
24-Hour Resting Loss (grams) =
(24 * Hourly_Resting_l_oss_at_l_owest_Temperature)
+ ( 0.03193 * Diurnal_Temperature_Range )
Where the Diurnal_Temperature_Range is the difference of the daily high
temperature minus the daily low temperature. This equation is
used to predict the full-day's resting loss emissions that can
then be subtracted from the RTD test results yielding the full-
day's diurnal emissions.
-------
-54-
Appendix E
Regression Analyses of Mean 24-Hour Diurnal
Versus Vapor Pressure Product Term
1971-79 Carbureted Vehicles Passing the Pressure Test
(based on 2 vehicles)
Dependent variable
No Selector
R squared = 97.4%
s= 5.175 with 6-
Source
Regression
Residual
Variable
Constant
Square of
VP_Prod 71,000
is:
R squared (adjusted) =
2 = 4 degrees of freedom
Sum of Squares
3958.84
107.107
Coefficient s
-3.971210
0.048250
96.7%
df
1
4
.e. of Coeff
3.4920
0.0040
Mean Square
3958.84
26.7769
t-ratio
-1.14
12.2
Means of Diurnal
F-ratio
148
prob
0.3190
0.0003
1971-79 Carbureted Vehicles Failing the Pressure Test
(based on 2 vehicles)
Dependent variable
No Selector
R squared = 94.1%
s = 4.824 with 6 -
Source
Regression
Residual
Variable
Constant
Square of
VP_Prod 71,000
is:
R squared (adjusted) =
2 = 4 degrees of freedom
Sum of Squares
1484.47
93.0907
Coefficient s
12.65690
0.029546
92.6%
df
1
4
.e. of Coeff
3.2560
0.0037
Mean Square
1484.47
23.2727
t-ratio
3.89
7.99
Means of Diurnal
F-ratio
63.8
prob
0.0177
0.0013
-------
-55-
Appendix E (continued)
Regression of Mean Diurnal Emissions
1980-85 Carbureted Vehicles Passing Both Purge and Pressure Tests
(based on 3 vehicles)
Dependent variable is
No Selector
R squared = 97.2%
s = 2.349 with 6 - 2 =
Source
Regression
Residual
Variable
Constant
Cube of VP Prod/
1,000,000
R squared (adjusted) =
= 4 degrees of freedom
Sum of Squares
766.002
22.0621
Coefficient s
1 .74328
0.014639
96.5%
df
1
4
.e. of Coeff
1.299
0.0012
Mean Square
766.002
5.51554
t-ratio
1.34
11.8
Means of Diurnal
F-ratio
139
prob
0.2509
0.0003
1980-85 Carbureted Vehicles Failing the Pressure Test
(based on 2 vehicles)
Dependent variable
No Selector
R squared = 95.5%
s = 3.054 with 6 -
Source
Regression
Residual
Variable
Constant
Square of
VP_Prod 71,000
is:
R squared (adjusted) =
2 = 4 degrees of freedom
Sum of Squares
790.710
37.3085
Coefficient s
10.0859
0.021564
94.4%
df
1
4
.e. of Coeff
2.061
0.0023
Mean Square
790.710
9.32713
t-ratio
4.89
9.21
Means of Diurnal
F-ratio
84.8
prob
0.0081
0.0008
-------
-56-
Appendix E (continued)
Regression of Mean Diurnal Emissions
1980-85 Carbureted Vehicles Failing ONLY the Purge Test
(based on 3 vehicles)
Dependent variable is
No Selector
R squared = 96.8%
s = 3.262 with 6 - 2 =
Source
Regression
Residual
Variable
Constant
Cube of VP Prod/
1,000,000
R squared (adjusted) =
= 4 degrees of freedom
Sum of Squares
1285.00
42.5584
Coefficient s
5.18176
0.018960
96.0%
df
1
4
.e. of Coeff
1.805
0.0017
Mean Square
1285.00
10.6396
t-ratio
2.87
11.0
Means of Diurnal
F-ratio
121
prob
0.0454
0.0004
1980-85 Fuel-Injected Vehicles Passing the Pressure Test
(based on 4 vehicles)
Dependent variable is:
No Selector
R squared = 98.2%
s = 0.8350 with 6 - 2
Source
Regression
Residual
Variable
Constant
Square of
VP_Prod 71,000
R squared (adjusted) = 97.
= 4 degrees of freedom
Sum of Squares
154.859
2.78884
Coefficient s.e.
2.40746 0
0.00954291 0
8%
df
1
4
of Coeff
.5635
.0006
Mean Square
154.859
0.697209
t-ratio
4.27
14.9
Means of Diurnal
F-ratio
222
prob
0.0129
0.0001
-------
-57-
Appendix E (continued)
Regression of Mean Diurnal Emissions
1980-85 Fuel-Injected Vehicles Failing the Pressure Test
(based on 2 vehicles)
Dependent variable
No Selector
R squared = 93.5%
s = 3.290 with 6 -
Source
Regression
Residual
Variable
Constant
VP_Product
is:
R squared (adjusted) =
2 = 4 degrees of freedom
Sum of Squares
619.704
43.3046
Coefficient s
0.286889
0.033366
91 .8%
df
1
4
.e. of Coeff
3.692
0.0044
Mean Square
619.704
10.8262
t-ratio
0.078
7.57
Means of Diurnal
F-ratio
57.2
prob
0.9418
0.0016
1986-95 Fl Vehicles Passing Both Purge and Pressure Tests
(based on 16 vehicles)
Dependent variable is:
No Selector
R squared = 91 .6%
s = 0.9934 with 6 - 2
Source
Regression
Residual
Variable
Constant
Square of
VP_Prod 71,000
Means of Diurnal
R squared (adjusted) = 89.5%
= 4 degrees of freedom
Sum of Squares
43.1082
3.94721
Coefficient s.
-0.834330
0.005035
df
1
4
e. of Coeff
0.6704
0.0008
Mean Square
43.1082
0.986802
t-ratio
-1.24
6.61
F-ratio
43.7
prob
0.2813
0.0027
-------
-58-
Appendix E (continued)
Regression of Mean Diurnal Emissions
1986-95 Fuel-Injected Vehicles Failing the Pressure Test
(based on 11 vehicles)
Dependent variable
No Selector
R squared = 95.1%
s= 2.221 with 6-
Source
Regression
Residual
Variable
Constant
Square of
VP_Prod 71,000
is:
R squared (adjusted) =
2 = 4 degrees of freedom
Sum of Squares
380.824
19.7226
Coefficient s
1.75768
0.014965
93.8%
df
1
4
.e. of Coeff
1.499
0.0017
Mean Square
380.824
4.93065
t-ratio
1.17
8.79
Means of Diurnal
F-ratio
77.2
prob
0.3059
0.0009
1986-95 Fuel-Injected Vehicles Failing ONLY the Purge Test
(based on 12 vehicles)
Dependent variable is:
No Selector
R squared = 85.8%
s = 2.492 with 6 - 2
Source
Regression
Residual
Variable
Constant
Square of
VP_Prod 71,000
R squared (adjusted) =
= 4 degrees of freedom
Sum of Squares
150.611
24.8387
Coefficient s
4.14550
0.009411
82.3%
df
1
4
.e. of Coeff
1 .6820
0.0019
Means of Diurnal
Mean Square
150.611
6.20969
t-ratio
2.46
4.92
F-ratio
24.3
prob
0.0693
0.0079
-------
-59-
Appendix F
Modeling 24-Hour Diurnal Emissions
As Functions of Vapor Pressure (kPa)
In each of the following strata, 24-hour diurnal emissions are modeled
using four constants: A, B, C, and D. Where,
24-Hour Diurnal (grams) =
= A + B * [(Mean VP) * (Change in VP)]
+ C * [(Mean VP) * (Change in VP)]2 /1,000
+ D * [(Mean VP) * (Change in VP)]3 /1,000,000
For each of the 19 strata, the four constants used to model diurnal emissions are
given below in the following table:
Fuel
System
Both
FI&
Garb
Garb
Fl
Garb
Fl
Both
FI&
Garb
Model Yr
Ranqe
72-79
80-85
80-85
86-95
86-95
1999+
1999-2003
2007+
Purge /
Pressure
Fail Press
Fail Purge
PASS Both
Fail Press
Fail Purge
PASS Both
Fail Press
Fail Purge
PASS Both
Fail Press
Fail Purge
PASS Both
Fail Press
Fail Purge
PASS Both
Fail Press
Fail Purge
PASS Both
PASS Both
A
6.90895
-4.58719
-8.09426
9.71190
6.88852
2.97845
-3.14389
2.39612
-1.29432
1.51716
4.11975
1 .42298
0.47846
3.25800
0.38830
0.47846
3.25800
0.19415
0.09222
B
0
0
0
0
0
0
0.033366
0
0
0
0
0
0
0
0
0
0
0
0
C
0.029546
0.048250
0.048250
0.021564
0
0
0
0.009543
0.009543
0.02156
0
0
0.014965
0.009411
0.005035
0.01497
0.00941
0.00252
0.00119
D
0
0
0
0
0.018960
0.014639
0
0
0
0
0.01896
0.01464
0
0
0
0
0
0
0
The ETP phase-in years (96-98) will be weighted averages of the pre-ETP and
the ETP models. Similarly, for vehicles passing both purge and pressure, the
Tier-2 phase-in years (2004-2006) will be weighted averages of the pre-Tier-2
and the Tier-2.
-------
-60-
Appendix G
Plots Comparing Diurnal Models to Means of Measured Data
(Graphing of Appendix C versus Appendix F)
In the following 15 graphs, we plot on the same graph both
the equations (from Appendix F) that are used in MOBILE6 to model
diurnal emissions (as a function of the vapor pressure product
term) as well as the corresponding mean diurnal emissions from
Appendix C (after subtracting daily resting loss emissions from
the RTD test results). To make the comparisons between graphs
easier, the same scale is used on all 15 graphs.
Note: The three strata of 1986-95 model year carbureted
vehicles were each tested at only four of the six possible
temperature cycle / RVP combinations (see page 30).
Note: The points that are most distant from the curves (e.g.,
the single test at the highest VP on the pre-1980 vehicles
passing both the pressure and purge tests) are those that are
averages of only a small number of RTD test results.
Figure G-l
Carbureted 1971-79 - Failing Pressure Test
iuu -
>s 1C
rams/da
n -
3 t
3 ou
"(0
'^T or
Q £O
n
MOBILE6
B RVP 6.8
• RVP 9.0
-J— ^
•
**
-*'*'*
•^
^'*
^^
^
300
600 900 1,200
VP .Product Term (kPa**2)
1,500
-------
-61-
Figure G-2
Carbureted 1971-79 - Failing Purge Test
iuu -
^ 1C -
rams/da
n -
3 t
3 ou
"fO
i_
3
— 9C
n -
MOBILES
B RVP 6.8
• RVP 9.0
• ^^*"
^ .*•
• ^-
^* ^^
*^
•*•
•
<•••
»-»''**
» ^^^
* ^
x*
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
Figure G-3
Carbureted 1971-79 - Passing Both Pressure and Purge
iuu -
*ii -7C .
rams/da
n ••
3 c
3
"fO
c
3
9C
n
MOBILE6
B RVP 6.8
• RVP 9.0
• ^'tf
_ ^
M
^*
<*S'
**> "**
^ **"
•-*»
»
S
^•
^
X
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
-------
-62-
Figure G-4
Carbureted 1980-85 - Failing Pressure Test
iuu -
^ 1C -
rams/da
n -
3 t
3 ou
"fO
i_
3
— 9C
n
MOBILES
B RVP 6.8
• RVP 9.0
-••-"•
. • _f •"
* ^-*'*'
.x*
*-***
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
Figure G-5
Carbureted 1980-85 - Failing Purge Test
iuu -
>K -1C .
CO 75
T3
~5>
E
CO
i- en -
"fO
c
n
MOBILE6
m RVP 6.8
• RVP 9.0
"I
i
> *%
^
^^
^
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
-------
-63-
Figure G-6
Carbureted 1980-85 - Passing Both Pressure and Purge
iuu -
^ 1C -
rams/da
n -
3 t
s ou
"fO
1_
2
— 9C
n
MOBILES
B RVP 6.8
• RVP 9.0
__> m
-»- - * " "
i
. -^»"-"
^*
S
,'*'
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
Figure G-7
Carbureted 1986-95 - Failing Pressure Test
iuu -
>K -1C .
CO 75
T3
~5>
E
CO
i- en -
"fO
c
n
MOBILE6
m RVP 6.8
• RVP 9.0
~ — *~ *
^ -«•-*-
^-^ '
> "^
^^ -^
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
-------
-64-
Figure G-8
Carbureted 1986-95 - Failing Purge Test
iuu -
^ 1C -
rams/da
n -
3 t
3 ou
"fO
i_
3
— 9C
n
MOBILES
B RVP 6.8
• RVP 9.0
__ ,-
•^
~*>~""~*
^*- '
•
^^ ^
--***
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
Figure G-9
Carbureted 1986-95 - Passing Both Pressure and Purge
iuu -
*ii -7C .
rams/da
n ••
3 c
3
"fO
c
3
9C
n
MOBILE6
B RVP 6.8
• RVP 9.0
^
-v--l"~^
i
•
**
S
^
«•
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
-------
-65-
Figure G-10
Fuel-Injected 1980-85 - Failing Pressure Test
iuu -
^ 1C -
rams/da
n -
3 t
3 ou
"fO
i_
3
— 9C
n
MOBILES
B RVP 6.8
• RVP 9.0
^M-~~^m
B^-
-»^-
_^ ^ — — '
--*"***
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
Figure G-ll
Fuel-Injected 1980-85 - Failing Purge Test
iuu -
*ii -7C .
rams/da
n ••
3 c
s
"fO
C
3
9C
n
MOBILE6
B RVP 6.8
• RVP 9.0
•.
-» •• — "
. % ~'
-i-* "" ""
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
-------
-66-
Figure G-12
Fuel-Injected 1980-85 - Passing Both Pressure and Purge
iuu -
^ 1C -
rams/da
n -
3 t
s ou
"fO
1_
2
— 9C
n
MOBILES
B RVP 6.8
• RVP 9.0
. -~*
-^ 1
' *•**"*"
- "" "*
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
Figure G-13
Fuel-Injected 1986-95 - Failing Pressure Test
iuu -
*ii -7C .
rams/da
n ••
3 c
3
"fO
c
3
9C
n -
MOBILE6
m RVP 6.8
• RVP 9.0
J^^-»
- — ••""""
' "^V*"
-j -* "**
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
-------
-67-
Figure G-14
Fuel-Injected 1986-95 - Failing Purge Test
iuu -
^ 1C -
rams/da
n -
3 t
3 ou
"fO
i_
3
— 9C
n -
MOBILES
B RVP 6.8
• RVP 9.0
.•_--••
-»- - * ~ "
""V*" "*
.---'
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
Figure G-15
Fuel-Injected 1986-95 - Passing Both Pressure and Purge
iuu -
*ii -7C .
rams/da
n ••
3 c
3
"fO
c
3
9C
n
MOBILE6
m RVP 6.8
• RVP 9.0
^
1 ""*
300
600 900 1,200
VP.Product Term (kPa**2)
1,500
-------
-68-
Appendix H
Response to Peer Review Comments from H. T. Me Adams
This report was formally peer reviewed by two peer reviewers
(H. T. McAdams and Sandeep Kishan). In this appendix, comments
from H. T. McAdams are reproduced in plain text, and EPA's
responses to those comments are interspersed in indented italics.
Comments from the other peer reviewer appear in the following
appendix (Appendix I).
************************************
Evaluating Resting Loss and Diurnal Evaporative Emissions Using
RTD Tests
By
Larry C. Landman
Report Number M6.EVP.001
Review and Comments
By
H. T. McAdams
1. REPORT CLARITY
Report clarity is determined by a number of factors, among them
semantics, readability and logical rigor. The role played by
semantics is multifaceted but can be roughly summarized as
matching the report to the readership. Readability connotes
directness and simplicity to the extent that the subject matter
allows. Readability, therefore, is closely associated with
style. Though it cannot alone guarantee clarity, most readers
would probably agree that a report that is easily read is more
likely to get its points across than a report that is weighted
down with long, involved sentences and clumsy organization. Also
clear writing does not insure sound logic, but it is difficult to
envision one without the other.
In the review of Report M6.EVP.005, certain stylistic revisions
were recommended in the interest of readability. Sometimes,
however, grammar and syntax may be pristine but the writing
conveys a mixed message. An example is found in Landman's
discussion of the frequency of gross liquid leakers (Section
10.1, page 30 et seq.
On page 30 is the following table.
Vehicle Frequency of
Gross Liquid Leakers
5.62 0.28%
12.50 2.00%
21.29 7.84%
-------
-69-
Th is table together with Figure 10-1 and comments on page 31 make
it unclear whether the percentages refer to the fraction of
vehicles of a particular age or to the fraction of the entire
fleet. The table leads one to believe that 2.00%, for example,
refers to 2% of vehicles that are 12.5 years old. The y-axis of
the figure says simply "Frequency (%)" without identifying the
sample space that the percentages are based on. Inasmuch as the
x-axis denotes age, however, it would seem logical to assume that
percent connotes percent of vehicles of a given age.
EPA revised this document to remove this source of possible
ambiguity.
Comments following the graph, however, are less than clear. For
example, the statement is made that "the rapidly increasing
proportion of gross liquid leakers in the in-use fleet tends to
be offset by the decreasing number of older vehicles in the in-
use fleet." The statement seems to be either contradictory or
circular: the proportion can not both "increase rapidly" and yet
"be offset" by the decreasing number of older vehicles.
The report should make it clear, however, that two functions have
to be considered. One plots vehicle age along the x-axis and the
proportion of vehicles of that age that are leakers along the y-
axis. The other plots vehicle age along the x-axis and the
fraction of the fleet represented by vehicles of that age along
the y-axis. For any given vehicle age, it is the product of the
two functions that must be considered. For example, if 1% of 8-
year-old vehicles are leakers, but 8-year-old vehicles make up
only 40% of the fleet, then only 0.4% of the fleet will consist
of 8-year-old gross liquid leakers.
To estimate the portion of the fleet consisting of gross liquid
leakers of all ages, one would have to make the same computation
for all vehicle ages and sum the results. If vehicle ages are
discrete, the computation would be simply the product of two
vectors. If age is represented as continuous, then one would
have to integrate the product function over time.
Revision of this part of the report is strongly recommended. It
would also be prudent to look for similar situations where some
simplification and clarification might be in order.
EPA agrees with those suggestions, and this report has been
revised to incorporate those changes.
2. OVERALL METHODOLOGY
The general methodology of this report essentially parallels that
of Report No. M6.EVP.005. Both are driven by the objective to
be realized: to model real-time diurnal (RTD) emission tests over
a 72-hour period as a more realistic way to measure diurnal
emissions. Separating resting loss emissions from diurnal
emissions is a corollary aim. Resting loss emissions, the report
-------
-70-
says, are always present, whether there are any pressure-driven
emissions or not.
We will examine, first, the mechanics of resting losses and
secondly, the philosophy underlying the statistical approach.
2.1 Resting Losses: Real or Definitional?
Resting loss emissions are isolated by finding that region in the
time series when emissions are essentially constant. According
to the report, this time period encompasses hours 19 through 24.
The hourly resting loss emissions are taken as the average of
those losses over the 6-hour time period. This average is then
subtracted from what was originally defined as diurnal losses to
obtain a new measure of diurnal losses.
The methodology seems straightforward enough, but somehow not
logically consistent. One senses that there are some aspects of
the approach that need to be examined in more depth. In Report
M6.EVP.005 one of the dictums laid down is that evaporative
emissions over a zero time cycle (i.e., at constant temperature)
are defined to be zero. Evidently, then, the resting loss
emissions, as determined by the emissions during the 19th to 24th
hours, must be pressure driven if only by the small pressure
range associated with the small (5 degrees or so) temperature
range.
This interpretation is not correct. Report M6.EVP.005
actually states: "The 24-hour diurnal emissions will be
zero grams for any temperature cycle in which the diurnal
temperature range is zero degrees Fahrenheit (i.e., a
constant temperature throughout the entire day)." Thus, the
reviewer's argument concerning resting loss emissions is not
applicable.
If that is so, why not run a 6-hour test over that temperature
range and measure the resting loss directly? Should we expect to
get the same results that we would get by isolating the 19th
through 24th hours from the 24-hour cycle? Do the losses during
the preceding 18 hours somehow have an effect on the next six
hours? Is there, in fact, a lag effect?
To admit that possibility would mean that a pressure active at
time t0 might not take effect until some later time t0 + delta.
Actually, it would seem, pressure acting over zero time would
have no effect on emissions. Moreover, if one were to look at
pressure cumulatively over the pressure range, the pressure
active at time t0 + 1 hour would be effective over a considerably
longer time than the pressure active at time t0 + 5 hours.
Viewed in this light, resting losses are produced by the same
mechanism as losses associated with any other small pressure
interval and would seem to exist only by arbitrary definition.
-------
-71-
More analysis needs to be addressed to this phenomenon. Computer
modeling seems to be suggested and possibly serial correlation to
determine the "decay time" for any given pressure.
While the suggestion may have merit; EPA is not able, at
this time, to conduct a new vehicle testing program to
validate this hypothesis (i.e., determining whether that
approach would produce results different from those
calculated by EPA in this report). Therefore, EPA will
estimate resting loss emissions (in MOBILE6) by averaging
the hourly evaporative emissions produced during the final
six hours of the 24-hour RTD test.
2.2 Philosophy Underlying the Statistical Approach
Special attention is directed to the statistical approach used to
analyze RTD data, because it might seem to break with tradition.
Vehicles are selectively recruited in such a manner that the
sample is enriched with malfunctioning evaporative emission
control systems. At first glance, it might appear that
deliberately biasing the sample is against all sampling precepts.
Actually, though, the procedure is just an ingenious adaptation
of stratified sampling. The purge and pressure tests define the
strata, each of which can be sampled as desired and the results
weighted in accordance with the relative frequency of the strata
in the fleet.
The choice of test parameters - e.g., fuel RVP and temperature
cycle - is consistent with engineering knowledge. However, the
way the temperature-related variables are redefined may merit
further attention, as will become evident later in this
discussion.
In addition to the variables that are known to affect evaporative
emission, however, there are error sources that contaminate the
emission measurements and complicate the construction of a
predictive model. In the report, it is recognized that the same
results may not be obtained at different test sites. It is also
acknowledged that light-duty vehicles (LDV) may perform
differently from light-duty trucks (LDT), and that carbureted
vehicles may perform differently from fuel-injected (FI)
vehicles. These differences, however, are the source of what
statisticians call "fixed effects." Often they are treated as
"dummy variables" having only two values, 0 and 1.
In the problem before us, it is likely that these fixed effects
extend all the way to the level of individual vehicles. That is
certainly true of exhaust emissions. In the development of the
Complex Model for RFG, vehicle-to-vehicle variation was found to
account for more than 90% of the total variation. A case was
even cited in the reports under review in which the vehicle
variation exceeded the variation accounted for by a predictor
variable. As elsewhere noted, vehicle effects can often be
isolated by way of dummy variables.
-------
-72-
The hierarchy of errors are well described by this bit of
doggerel verse.
Dogs have fleas,and fleas have fleas
Upon their backs to bite "em,
And they in turn have smaller fleas,
And so on ad infinitum.
So it is with emissions. Beyond the level of vehicle-to-vehicle
variation arising from assignable causes, there are errors
arising from unassignable causes. These are the error sources
that give rise to the fact that repeated tests of a particular
vehicle do not give the same results. Repeated tests such as
these, known in statistical parlance as replications, can be used
to estimate the magnitude of these random and unassignable
errors. By proper experiment design, these errors can be
determined by removing all identifiable effects and assigning the
remainder to error, as is done in regression analysis. Any fixed
effects not removed will inflate the error term and thereby
weaken the ability of the test to detect effects that would
otherwise be called statistically significant.
Running a testing program to characterize the variability of
individual in-use vehicles as well as vehicle-to-vehicle
variability will be a major undertaking. This is not
something that can be done in time for MOBILE6.
The methodology used in this investigation, for the most part,
acknowledges this hierarchy of effects, but there are instances
where error-management is less than optimum. In particular,
error bounds are not specified except indirectly by statistical
tests of significance used as the basis for including or
excluding certain fixed effects. Though the approach used in the
report is generally accepted under similar circumstances, the
arbitrariness of significance tests leaves something to be
desired.
Although Dr. McAdams believes this approach "leaves
something to be desired," he does acknowledge that "the
approach used in the report is generally accepted under
similar circumstances," and he makes no specific
recommendations. Therefore, EPA will continue to use the
same statistical approach as in the draft version.
3. APPROPRIATENESS OF DATASETS SELECTED
The available data comes from two very different sources, EPA
"work assignments" and CRC tests. The EPA data is collected so
as to be biased toward defective emission-control systems; that
data is then "weighted" according to the frequency with which
these defective vehicles are found in the real-world fleet. The
CRC data is collected randomly but is biased toward trucks.
Certainly this combination of data sets can not be said to be
ideal.
-------
-73-
For purposes of seeing just how far from ideal the combined
sample is, it might be informative to list the characteristics of
both side by side, as is done in the comparison of two products
in the advertising media. Undoubtedly there would be many blank
entries in the combined-sample column.
Landman has done a commendable job in adapting this less-than-
optimum dataset to the purposes of his report. In some instances
it was necessary to find a suitable surrogate for the missing
data, and in other instances it was necessary to take a leap of
faith. Though such tactics would not be admissible in a well-
designed experiment, attention must be turned to "next best"
options. Several are to be found in the report. The only
suggestion that might be made here is to attempt to put bounds on
the possible differences between what we have and what we would
like to have.
4. DATA ANALYSIS AND STATISTICAL METHODOLOGY
Much of the concern in the report is with how to classify
vehicles into appropriate subsets or strata. The methodology
employed relies heavily on comparison of cumulative distribution
plots for those data sets that are considered to be candidates
for merging into a single set.
Landman starts with a list of classes or strata that are
physically discernable, such as cars vs trucks, vehicles that
pass certain tests vs vehicles that fail those tests, and so on.
He then essentially arranges the RTD emissions for these vehicles
into ascending order and constructs the sample cumulative
probability distribution for each vehicle set. Finally, he plots
the cumulative distributions for sets that are candidates for
merging and relies on judgment to decide whether or not to
combine these sets.
This procedure is subject to criticism on two counts. First,
standard practice in the plotting of cumulative distributions is
to plot measurements along the x-axis and cumulative probability
along the y-axis.
The formats of the graphs have been revised.
Secondly, judgemental decision making is highly subjective and,
in this case, need not be, because a number of statistical tests
are at hand for comparing two sets of measurements. These
include contingency tables, discriminant functions and cluster
analysis, as well as such simple procedures as a t-test for the
difference between two means.
The discussion has been revised to include more objective
methods.
First, a word about the mechanics of plotting cumulative
distributions. Though there is nothing wrong about Landman's way
of plotting, it is unconventional and may cause the reader to re-
-------
-74-
orient his thinking. Probability graph paper is designed so that
the measurement axis is horizontal and the probability axis is
vertical. Indeed, it might be interesting to plot Landman's data
on such paper. Inasmuch as most data tends to follow a normal
distribution, data tends to plot as a straight line, with the
intercept and slope being measures of central tendency and
dispersion, respectively. Accordingly, the linearized plots may
be simpler to compare and interpret. (An alternative to
probability graph paper is to transform cumulative percents to
multiples of standard deviation in a normal [0,1] distribution.)
The software that we are using does not easily lend itself
to graphing these cumulative distributions using a
probability graph scale.
Next, let us look at ways to compare two distributions. If the
distribution of the data can be identified as being a sample from
a known theoretical distribution, it suffices just to compare the
estimated parameters of the theoretical distribution. For
example, if data are known or suspected to be a sample from a
normal distribution, it suffices to compare the means and
standard deviations for the two distributions being compared.
Since the form of the data distribution is not actually known,
Landman evidently seeks a comparison that circumvents any
assumptions about the form of the underlying statistical
distribution. That fact opens the door to an array of so-called
nonparametric tests.
Plotting the cumulative distributions for the two samples being
compared is a step in the right direction. From here, however,
Landman's approach devolves to purely personal "eye-balling." In
comparing Indiana and Arizona test sites, for example, he says
"Despite the small sample size in the Indiana data ... the
closeness of the distribution curves is compelling and suggests
that there is no reason to treat the test data separately."
Elsewhere, in comparing the cumulative distributions for PFIs and
TBIs, he says "Based on the similarity of the cumulative
distribution curves and on the close fit of the means ... the
PFI and TBI strata were merged into a single fuel-injected (FI)
stratum for the remaining analyses." Though most analysts might
agree with Landman on the above two cases, agreement becomes much
less likely when we come to comparison of such pairs as cars vs
trucks and EPA vs CRC testing programs.
Let us return to Landman's contentions with regard to the "close
fit" of means and medians. The median is only one point on the
cumulative distribution - namely the 50th percentile - and would
seem to be redundant when comparing the entire cumulative
distributions. If the entire cumulative distributions are "close
fits," there are no "extra points" for the fact that the medians,
also, essentially match. And, with regard to the means, how
close is "close enough" to make the assertion that they are a
"close fit?"
-------
-75-
Comparison of the means of two sets of data is one of the first
problems presented in elementary statistics courses. It is first
addressed in comparing the means of two sets of data when the
number of cases is the same for both sets. Later, a method is
presented for the instance of unequal sample sizes. To implement
the test, one needs the sample size, mean and standard deviation
for both samples. From that, one computes the applicable value
of t in the t-distribution. Even though the underlying
distribution for the two samples might not be normal, the
distribution of sample means tends to approach the normal by
virtue of the Central Limit Theorem. This observation is true
even for distributions as far from normal as a uniform
distribution and for sample sizes as small as three or four.
Many of the comparisons used only the "eye-balling" approach
since the curve "fits" appeared to be obvious; however, this
report has been amended to include more objective methods.
As illustrations, we present results for (1) comparison of PFI vs
TBI (p. 10) and (2) carbureted vs FI trucks (p.12).
Sample Application of t-test to Coalesence of Sets
Sets
Compared
TBI Trucks
PFI Trucks
Carb.Trucks
FI Trucks
Sample
Size
19
24
7
43
Mean
5.41
5.85
9.31
5.65
Std.
Dev.
5.70
7.87
8.28
5.92
t-ratio Prob.
-0.2047 0.8388
1.2640 0.2113
The above test is based on the assumption that the two sets of
data have the same standard deviation. When this assumption can
not be made, we face a situation known in the statistical
literature as the Behrens-Fisher problem, for which
approximations are available.
Unfortunately, standard deviations are not given for most of the
comparisons made in the report; among them are the more
problematic cases, those that could benefit most from a
statistical test of some kind. Although standard deviations are
given for the CRC data, they are not given for the EPA data. If
those statistics are available, it would be worthwhile to perform
other t-tests to bolster the decision to merge or not to merge
the strata.
The various tables have been expanded to include the
standard deviations, and the discussions in the text have
also been revised.
It is to be pointed out that the t-test gives insight only with
regard to the mean, into which all individual observations map.
If one is interested in a more comprehensive measure, such as the
cumulative distribution function, one might employ a non-
parametric test, such as the Kolmogoroff-Smirnoff test or chi-
square.
-------
-76-
Since the goal of the MOBILE model is simply to estimate the
mean of the emissions, EPA did not consider it necessary to
employ other statistical methods to estimate the
distributions.
5. APPROPRIATENESS OF THE CONCLUSIONS
As in Report M6.EVP.005, conclusions are not stated explicitly.
The thrust of the report, however, can be summarized as follows.
1) Real-time diurnal (RTD) tests of evaporative emissions make
it possible to evaluate "resting losses" that are always present,
even when there are no pressure-driven losses.
2) Resting losses per hour can be computed as the average RTD
over hours 19 through 24 of the cycle.
3) Both diurnal and resting losses can be simply modeled by
means of a set of regression equations in which the predictor
variables are RVP and a VP-product term referred to as prod in
this review.
4) The product term may enter equations as prod1, prod2 or
prod3 or some combination of these powers.
5) Whether two sets of data can be pooled can be judged by
comparison of their cumulative distribution functions.
6. RECOMMENDATIONS FOR ALTERNATE DATASETS AND/OR ANALYSES
Several explicit recommendations are made with regard to the
report being reviewed, Number M6.EVP.001. These recommendations
stem mainly from the report's treatment of stratification and the
manner in which it adjusts, or fails to adjust, for vehicle-to-
vehicle differences. Of particular concern are those instances
in which the standard deviation of the constant term approaches
or exceeds its estimated value. It is likely that this inflation
of the error of estimating the constant is caused by vehicle-to-
vehicle differences in their overall level of evaporative
emissions. Wherever feasible, these vehicle effects should be
removed and the regression coefficients recomputed.
The effect of the statistical uncertainty in the constant
term is most significant when the vapor pressure product
variable is near zero. However, that situation will not
occur in MOBILE6 because interpolation will be used for
scenarios in which the daily temperature difference (daily
temperature rise) is less than ten degrees Fahrenheit.
Therefore, EPA is less concerned with the statistical
uncertainty in the constant term than with the ability of
the individual formulas to predict the emissions in the
likely range of the vapor pressure product term.
-------
-77-
The following recommendations are made with regard to
improvements that might be implemented within the time frame of
finalization of this report.
1) Wherever possible, perform quantitative statistical tests
to support the subjective decision made to pool or not to pool
certain datasets.
The report has been revised to include the use of more
objective statistical tests.
2) Attempt to resolve the objection voiced earlier in this
discussion with regard to the product term prod and its powers.
Investigating the uncertainty in the resulting regression
equations produced by the vehicle-to-vehicle variability is
beyond what can be done prior to completing MOBILE6.
3) Wherever possible, use dummy variables to prevent the
aliasing of vehicles or other extraneous variable with predictor
variables.
The data were reanalyzed using categorical (dummy) variables
instead of separate analyses for each stratum. However, the
results were either unacceptable from an engineering
perspective (i.e., predicting increasing emissions for
decreasing fuel RVPs) or produced poorer "fits" within some
strata to the actual test data.
4) For representative cases, give error bounds for regression
coefficients and for diurnal and resting loss emissions as
computed by the applicable regression equation.
The regression tables (in Appendix E) include the standard
error of the coefficient for the regressions of diurnal
emissions.
As pointed out earlier in this discussion, resting losses need to
be defined more rigorously. To do so may require a better
understanding of the loss process.
With regard to the modeling of diurnal and resting emissions,
further attention needs to be directed to the reciprocity aspect
of the product term prod and the "indifference" of powers of that
term when selected as a predictor variable.
These are issues that might better be postponed for further
systematic study. In that category, also, is a re-orientation of
thinking with regard to more thoughtful application of
statistical tests, regression analysis, and non-parametric
procedures.
1-20-99
htm
-------
-78-
Appendix I
Response to Peer Review Comments from Sandeep Kishan
This report was formally peer reviewed by two peer reviewers
(H. T. McAdams and 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 November 20, 1998) that do not necessarily match
the page numbers in this final version. Comments from the other
peer reviewer appear in the preceding appendix (Appendix H).
************************************
This memorandum provides peer review comments on two EPA
documents: "Evaluating Resting Loss and Diurnal Evaporative
Emissions Using RTD Tests", Document No. M6.EVP.001, November 20,
1998 and "Modeling Diurnal and Resting Loss Emissions" Report
Number M6.EVP.005, October 1, 1998. Both of these are draft
reports.
The original peer review covered two of the MOBILE6
documents. Only the portion of that review pertaining to
this report is reproduced in this appendix. The remainder
of the peer review is reproduced in report number M6.EVP.005
(Appendix G of that report) .
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.001 (November 20, 1998)
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.
Overall, this report was clearly written and the overall
methodology seems alright. We were comfortable with the
stratification based on purge and pressure tests results and the
appropriateness of the datasets. We do not have any
recommendations of any alternate datasets. We think that our
most important comments involve the statistical analysis in which
the average emissions of vehicles were regressed against input
variables. We think that better results could be obtained by
using a class variable to identify each vehicle in a regression.
This would allow more experimental data to be used in each
regression and it would provide better estimates of the
-------
-79-
regression coefficients. Additionally, we do not understand the
background behind the vapor pressure product term especially
since it seems to require patches for RVP functionality and for
predictions of diurnal temperature increases shorter than those
in the dataset.
The report has been revised to include additional background
on this vapor pressure product term, as well as, comparisons
between this term and the corresponding independent variable
used in the previous version of MOBILE.
I. It would be helpful if the individual car data were provided
in an appendix to the report. The data should provide the
identity of the car and the evaporative emissions values
obtained under each different test condition. This
information provides the reader with a way to evaluate the
raw data if he desires.
The raw data is too extensive to easily fit within an
appendix of this report. EPA will make it available
electronically, upon request, in spreadsheet files.
2. Page 7, 4th paragraph - The phrase in parentheses weighted to
correct for recruitment bias was not immediately clear. We
presume that the weighting corrects for the recruitment bias
of vehicles that failed purge and pressure tests. The
abundance of vehicles that failed these tests in the
population would be lower than they were in the vehicle test
fleet.
The reviewer's presumption is correct. (The phrase in
question is in the fourth paragraph of Section 5.0.)
3. In Section 5.0, beginning on Page 7, comparisons for
different strata of vehicles were made using RTD emissions.
We think that the report should state somewhere near the
beginning of this section, that these RTD comparisons really
include both diurnal and resting loss emissions. The
question then also arises as to whether the strata
evaluations or strata comparisons would be the same if the
RTDs had first been split into diurnal and resting losses
first and then comparisons of diurnal emissions in strata
made separately from comparisons of resting loss emissions
in strata.
A notation has been added to remind the readers that the
results of the RTD tests include both resting loss and
diurnal emissions. The alternative approach of splitting
the resting loss and diurnal emissions prior to the
comparisons may be considered in a future analysis (i.e.,
after the completion of MOBILE6).
4. Figures beginning with 5-1 on Page 8 - It would be helpful
if the legends in these figures include the number of
vehicles on which each curve in the plot is based. This
-------
-80-
would give the reader a way to judge the confidence of the
location of the curve on the plot.
The sample sizes are included in the text. Attempting to
include the counts in the legends of each graph
unfortunately reduces the readability (with the software
packages we are using).
, nd
5. Page 9, Section 6.0, 2n paragraph, Item 2 - One of the
questions asked was whether PFI and TBI vehicles could be
combined into a single stratum of fuel injected vehicles.
Another possibility is to consider combining TBI with
carbureted vehicles. Combining of vehicles in this fashion
has been used in other studies and might be considered for
this study.
Since a major source of diurnal emissions is from the
carburetor bowl, carbureted and fuel injected vehicles have
historically be placed into separate strata. The reviewer
did not indicate that this approach is flawed, only that an
alternate approach may exist. Therefore, EPA will continue
with this approach. The reviewers suggestion will be tested
in future analyses.
6. Page 11, Section 6.1 - Comparison of strata statistics at
the bottom of the page. This table no longer shows a value
for standard deviation as was presented in the previous
table of this sort in the report. In discussions with Larry
Landman, Larry noted that the measures on page 11 were a
combination of several strata of purge/pressure failures and
that it was difficult to calculate a combined standard
deviation. Also, it seems by examining these statistics,
the purpose is to determine if the distributions are the
same. Another possibility would be to perform a test to see
if the distributions are significantly different given the
sample sizes. It might be possible to perform a Komolgorov-
Smirnoff test on each of these distributions to back up the
visual appearance of the plots with a quantitative
assessment.
That table (and others) has been expanded to include the
standard deviations, and the discussions in the text have
been revised to also to include the standard deviations.
7. In Section 6.0 the comparisons of different strata are made
as the report states after weighting the results to correct
for recruitment bias. It would be helpful to have an
appendix in the report describing how these weightings were
accomplished. Larry pointed out that there was another EPA
report that described this.
The weighting method is the standard approach used when the
sample is obtained using stratified random (targeted)
recruitment.
-------
-81-
8. Page 18, last paragraph of Section 6.5 - while it is
probably true that this paragraph documents what has been
done with the different strata, it is quite difficult to
follow. Please consider modifying this paragraph to make it
more clear to the reader.
That paragraph has been revised.
9. Page 20, first paragraph of Section 7.2 - this section seems
to say that we do not need to bother modeling the resting
loss as a function of RVP and temperature but we can just
make it constant and subtract off the constant value for
every hour of the day from the RTD hourly emissions. This
gives the impression that we are introducing errors in the
deduced diurnal emissions values. Why don't you just model
resting losses as a function of RVP and temperature and
subtract them from the RTD emissions to get the diurnal
emissions? In addition to comment #0, it may be helpful to
include plots of several vehicles' RTD emissions versus time
in an appendix. Considering the plots may bring a better
understanding of the schematic of Figure 7-2.
Actually, the analyses were performed using estimates of
hourly resting loss emissions that were "corrected" to
account for the changing temperatures. The paragraph has
been revised.
As to supplying more sample plots (of RTD emissions versus
time), all of the individual (hourly) results will be
provided electronically. Thus, if any reader who is
interested can easily generate those plots.
10. Page 22, third full paragraph - while hot soak emissions are
not being studied in this report, this paragraph might say,
if you believe it is true, that leaks associated with the
fuel pump might appear as high hot soak emissions.
Yes, that is what is being suggested. While we have not
tested such a vehicle, it seems possible that some of the
leaked gasoline (e.g., from a leaky fuel pump) might still
be present in liquid form after the engine has been shut
off. That liquid gasoline would then be counted as part of
the hot soak emissions.
11. Pages 22 and 23, first two paragraphs of Section 8.0 - We
also expect that resting loss evaporative emissions are
functions only of ambient temperature but it seems that you
are immediately discarding the possibility of an RVP
influence even though you have data that would seem to allow
you to evaluate whether that influence is really present or
not. So, why not test for the RVP effect? This could be
done using the 57 vehicles in the EPA program that were
tested with both 6.8 and 9.0 RVP fuels and three temperature
cycles. Again, in discussions with Larry, he said that the
sequence of the tests and RVP effects may be confounded due
to the test procedure used. EPA should consider randomizing
-------
-82-
the tests in future programs so that such effects can be
properly investigated.
True. In future testing programs, we will consider
randomizing the order of the testing so that the effects of
RVP on resting loss can be investigated.
12. Page 23, middle of the page - at this point, we wanted to
see the evaporative emissions data on these individual 57
vehicles. This desire leads to the next item.
As note in the response to this reviewer's first comment,
EPA will make the raw data available electronically, upon
request, in spreadsheet files. One of those files will
consist of the data on just these 57 vehicles.
13. Page 24, Figure 8-1 - the average values plotted in the
table look quite good. However, consideration of just the
average values can hide a lot of additional interesting
information. We would like to be able to see the plot with
individual vehicles or by the 12 strata used in the analysis
in different symbols. Are the slopes different for the
different strata? Are the slopes different for different
RVP levels, purge and pressure failures, or model year
groups? These issues potentially have a greater
significance than the tiny amount of curvature seen in the
figure. Larry pointed out that these relationships will be
used in the MOBILE model which primarily considers averages
so that inventories can be developed. However, we still
think that at this stage of model development, it is
important to consider vehicle to vehicle differences to get
a better understanding of emission trends. In many datasets
we have seen that the influence of a parameter on emissions
is more subtle than vehicle-to-vehicle differences.
As note in the responses to the first and twelfth comments,
EPA will make the raw data will be available electronically
for any additional graphs and analyses that the readers may
wish to investigate. This approach (i.e., producing
distinct regression equations of resting loss emissions for
each tested stratum) was attempted and then discarded (as
noted in the third paragraph of Section 9.0) because it
produced predictions contrary to theoretical models.
14. Page 25, first paragraph - in the analysis values for B were
calculated separately for carbureted and fuel injected
vehicles. Did you look for different B's for the different
strata in the dataset? This question is really just a
restatement of question 12. However, it involves a
regression type of solution rather than a graphical one.
While there may insufficient data to detect significant
differences, they might be significant. If it is found that
the values are not significantly different, then this should
be stated along with the error of the estimate in B.
See previous response.
-------
-83-
15. Page 25, paragraph 3 - at this point the report gives the
reason for not considering the influence of purge failure
with respect to resting loss emissions. While the
engineering reason seems to make sense, it also seems that a
regression of the data could reveal whether the presence of
purge failures actually had a significant effect on resting
loss emissions. This would confirm the elimination of purge
failure from consideration and would make the case stronger
for not using it to describe resting loss emissions.
The sample sizes were too small (given the standard
deviations) to permit us to distinguish the resting loss
emissions of those substrata; thus, we relied upon
engineering judgment.
16. Page 25, paragraph 4 - the last sentence ending in 0.766
leaves the reader wondering where that value came from. It
would seem that either the value should not be mentioned or
it should be explained. I think the source is in Appendix
D, but I am not sure.
More detail has been added to the explanation that was
present.
17. Page 26, Section 9.0, paragraph 4 - the logic of using
VP.A.VP eludes me. Is there some theory to back this up or
some reference that can be referred to which contains
theoretical development?
The beginning of Section 9.0 (pages 26 and 27) has been
revised to provide more background on this VP product term,
including earlier uses of this variable as well as
comparisons with the variable (UDI) used in MOBILES.
18. Page 26, Section 9.0, paragraph 5 - the mean of the diurnal
emissions were modeled. As soon as the mean of individual
measurements are used, the error associated with those
individual measurements is lost. In addition, because the
number of tests performed at different conditions are
unbalanced, the regression has no way of distinguishing the
different uncertainties in the mean values. When individual
values are used in a regression, the model will attempt to
fit the measured data better for those conditions where
there are more observations. Without using individual
values, and instead using averages for each condition, a
regression model will put equal emphasis on each mean
whether that mean was calculated from one observation or 30.
Use of means can also result in a functional form which may
be actually unsupported by the individual vehicle data.
In the sample being analyzed, the same test vehicles and the
same number of test results were present at each point
(i.e., at each of the six values of the VP product term).
Thus, we avoided the potential problems that the reviewer
-------
-84-
noted. (This was one of the reasons that EPA restricted
this portion of the analyses to the 57 vehicles that were
tested at each of the six combinations of temperature cycle
and fuel RVP.)
19. Page 26, Section 9.0 paragraph 9 - the diurnal emissions
were repeatedly regressed against. What is meant by
repeatedly is not clear. We assume that this means playing
with the model statement until a satisfactory model is
achieved.
This assumption of the reviewer is correct.
20. Top of page 27 - doesn't the fact that RVP helped the
regression when the vapor pressure product was already in
the regression indicate that the vapor pressure product
doesn't do a satisfactory job of modeling the emissions?
This then causes us to wonder again what the theory of the
vapor pressure product is.
The vapor pressure product term does a satisfactory job of
modeling the diurnal emissions; it does not do a perfect
job. However, our attempts at improving those models
produced equations that closely predicted the diurnal
emissions at the six actual test conditions but erred at the
intermediate points (e.g., at 8.0 RVP).
21. Top of page 27 - a third step which should be used to choose
among the models is to validate or evaluate the alternative
regression equations against the diurnal emissions of the
vehicle data which were not used to build the models. The
models for which the measured and modeled diurnal emissions
data agree best would be top candidates for selection.
Yes. That was how the final models were chosen.
22. Page 27, last paragraph - the text says that you imposed the
following three restrictions. How did you impose these? Or
did you really just check the predicted values of models for
making sense based on engineering experience? The phrase
"impose restrictions" suggests that you did something during
model building rather than after model building.
The candidate models that did not meet those restrictions
were rejected. Therefore, we "imposed restrictions" after
the models were built, not during the building process
itself.
23. Footnote at the bottom of Page 27 - doesn't the fact that
the diurnal emissions should be zero when the temperature
increase is zero and the fact that the models do not
properly predict this boundary condition imply that the
model statement could be improved upon? It seems that it
would be possible to come up with a functional form for the
model statement that included this zero/zero condition and
-------
-85-
also describe reasonably well the measured emissions values
when the temperature increase was not zero.
Since the model should predict zero diurnal emissions (for
vehicles not "gross liquid leakers") when the daily
temperature is constant, we first tested regressions that
predict zero diurnal emissions when the temperature change
is zero. (That is, regressions that simply connect the
origin to the mean of the data.) None of those regressions
(neither linear nor non-linear) were able to accurately
predict the diurnal emissions at the six actual test
conditions. As a compromise, we used the "best fit"
equations for days in which the difference between the daily
high and low temperatures is at least 10 degrees and, for
days in which the temperature difference is less than 10
degrees, we simply interpolate between the prediction at 10
degrees and zero.
24. Page 28, second paragraph - we would like to see the
regression results before the constant terms are adjusted.
By the way, the use of the expression "altered the constant
terms" sounds like you are fudging the models. We would
suggest the use of a different expression and a little more
explanation on the reason for the adjustments.
Those regression results (prior to modifying the constant
terms) are contained in the tables in Appendix E. Following
the suggestion of the reviewer, the wording (concerning
"altering" the constant terms) has been revised.
25. We recommend that the report contain a plot of measured data
and model curves with appropriate symboling to describe the
different strata and/or test variables. If this type of
plot proves to be to busy or complex, at a minimum, we would
recommend showing parity plots, that is predicted versus
measured emissions, for each of the models. This will give
the reader an indication of how accurately the models
predict emissions and whether there are any major deviations
across the range of emissions.
Appendix G has been added to the report to illustrate the
predicted versus the (means of the) measured data.
26. Page 29, third paragraph "we then transformed the constant
term" - We think you don't really mean transformed but
really mean adjusted as you have described earlier in the
section. The word transformed, to us, means made a
transformation. For example, a natural log transformation
or a power law transformation. Also, two lines later, the
report uses the term "calculated diurnal emissions". We
assume you are referring to the fact that the diurnal
emissions were deduced from the RTDs by subtracting off
estimated resting losses. We would prefer to see the word
"estimated" or "deduced" to replace the word "calculated."
The wording has been revised similar to comment number 24.
-------
-86-
27. Last paragraph on Page 29 - you state that sometimes the
data suggests that a non-linear relationship for the diurnal
emissions with respect to the vapor pressure product term,
and sometimes it appeared to be linear. One approach to
determining if a relationship is non-linear is to scale the
inputs to have a mean of zero and then including in the
model statement a linear term and a non-linear term such as
a squared term. This scaling separates the linear and
curvature effects so that the significance of the
coefficient on the squared term can be used to answer the
question is there significant curvature in the relationship
between the inputs and the outputs. Was this technique used
to answer the question of whether the relationship was
linear or non-linear? Or was a different technique used?
The significance of the non-linear term was calculated using
the unsealed data.
28. Page 30, Section 10.1 - we suggest that you mention the
number of vehicles used to determine the logistic growth
curve for the frequency of gross liquid leakers.
The details are given in report M6.EVP.009.
29. Appendix B, page 43 - The discussion of the Clausius-
Clapeyron method. We can see how you calculate the values
of A and B from the two RVP values from the previous EPA
work assignment. However, we think that once the RVP, A,
and B values are determined, it will then not be possible to
ensure that the curves pass through the appropriate pressure
at 100°F. The equation seems to be over specified.
The discussion in Appendix B has been revised to improve
clarity. The results are unchanged.
30. Appendix D, page 47 - We don't follow the equation at the
bottom of the page or how it was obtained.
An explanation of that equation was added to Appendix D.
-------
-87-
Appendix J
Response to Written Comments from Stakeholders
The following comments were submitted in response to EPA's
posting a draft of this report on the MOBILE6 website. The full
text of each of these comments is posted on the MOBILE6 website.
Comment Number: is
Name / Affiliation: John Walsh / EPA Region 2
Date: April 3, 1997
Comment:
"We would assign this [diurnals, resting losses, & liquid
leaks] a low priority unless significant differences as
anticipated over estimates derived from current testing
methods."
EPA's Response:
We could not determine in advance whether these new
approaches would result in significant differences. These
new approaches are being used because we believe they more
closely represent the real world.
Comment Number: 28
Name / Affiliation: Chris Saricks / Center for Transportation
Research (Argonne National Lab)
Date: April 29, 1997
Comment:
"The CTR is generally pleased to learn that corrections are
planned in MOBILE6 ... for recalibrating the share of trip
emissions attributable to diurnal and resting emission
losses in recognition."
EPA's Response:
No response is necessary.
-------
-88-
Comment Number: 30
Name / Affiliation: Dale Aspy / EPA Region 4
Date: April 30, 1997
Comment:
Acceptable methodology needs to be developed for testing of
resting loss emissions.
EPA's Response:
The commentor appears to be concerned that the recruitment
method(s) used be able to account for the rare, high-
emitting outlier. We agree. We believe that the outliers
will be represented.
Comment Number: 32
Name / Affiliation: John Walsh / EPA Region 2
EPA's Response:
This comment is simply an exact duplicate of Comment No. 18.
Comment Number: 49
Name / Affiliation: Marc Houyoux / North Carolina
Supercomputing Center - Environmental
Programs
Date: October 30, 1997
Comment:
Relative to analyzing resting loss emissions, the commentor
does not like regression of means, he prefers to see scatter
plots.
EPA's Response:
If we were trying to predict the distribution of resting
loss emissions as a function of temperature, then we would
agree. However, since we are attempting to estimate fleet
(i.e., mean) emissions, we believe this approach is
acceptable.
-------
-89-
Comment Number: 54
Name / Affiliation: David Lax / API
Date: December 17, 1997
Comment:
API is "very concerned that the approach used in M6.RTD.001
to weight resting loss and diurnal evaporative emissions
test data based on pressure / purge test result status may
lead to biased representations of the in-use fleet."
EPA's Response:
EPA agrees that neither the purge test nor the pressure test
is the best choice to use in order to stratify the fleet.
Unfortunately, the vehicle recruitment process used them as
recruitment criteria. But, since each purge /pressure bin
is adequately represented, weighting the results should
produce an unbiased representations of the in-use fleet.
Comment:
API wants the statistics that EPA used to characterize the
resting losses so that they can confirm EPA's analyses.
EPA's Response:
Appendix C was added to the report to provide those data.
Comment:
API wants the statistics for characterization of 24-Hour
diurnal emissions to be able to confirm EPA's analyses.
EPA's Response:
Appendix F was added to the report to provide those data.
Comment:
In the section of their comments entitled "Accounting for
Liquid Leaks," API suggested that EPA "consider and evaluate
more data before finalizing an algorithm for this component
in the MOBILE model. In particular, ... the data on the
liquid leaks observed in the running loss test program
recently conducted by the CRC."
-------
-90-
EPA's Response:
We did consider those data (and others). As a result, we
revised our estimates.
Comment:
In the section of their comments entitled "Accounting for
Liquid Leaks," API suggested that EPA use the results of
their recent survey program to detect leakers.
EPA's Response:
We did use the results of their recent survey program to
detect leakers. We were not able to follow up on their
suggestion to study the effects of the behavior of the
driver on real-world leaks.
------- |