United States
Environmental Protection
Agency
Office of TransP°rtation EPA420-P-05-003
and Air Quality March 2005
MOVES2004 Energy and
Emission Inputs
Draft Report
-------
EPA420-P-05-003
March 2005
MOVES2004 Energy and Emission Inputs
Draft Report
John Koupal
Larry Landman
Edward Nam
James Warila
Carl Scarbro
Edward Glover
Robert Giannelli
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 that are currently available.
The purpose in the release of such reports is to facilitate an exchange of
technical information and to inform the public of technical developments.
-------
Table of Contents
1. Introduction 3
2. Overview of MOVES2004 Energy & Emission Inputs 5
3. Overview of Data Used in MOVES2004 5
3.1 Data Gathering and Processing Steps 6
3.2 Data Sources 7
4. Total Energy Rates 9
4.1 Source Bin Definitions 9
4.2 Operating Mode Definitions 11
4.3 Running Energy Rate Development 12
4.4 Start Energy Rate Development 36
4.5 Extended Idle 44
5. Petroleum and Fossil Energy Calculations 46
6. Carbon Dioxide (CO2) Calculations 47
7. Methane (CH4) and Nitrous Oxide (N2O) Rates 49
7.1 Source Bin Definitions 49
7.2 Operating Mode Definitions 49
7.3 Data Sources 50
7.4 Running and Start Rate Development 50
7.5 Advanced Technologies & Alternative Fuels 55
8. CO2 Equivalent Calculation 56
9. Adjustments 57
9.1 Temperature 57
9.2 Air Conditioning 64
-------
10. Well-To-Pump Energy & Emission Rates 65
11. Data Source Identification 67
12. Acknowledgments 69
Appendix A: Binning Methodology Proof-Of-Concept 70
Appendix B: Calculation of Running Energy Consumption Rates Using
"The Binner Program" 82
Appendix C: Proof-Of-Concept Assessment of Hole Filling Methods 104
Appendix D: Algorithm for Running Energy Hole Filling Using
Interpolation /Copying 120
Appendix E: Start Energy Rates 135
Appendix F: CH4 & N2O Rates by Model Year 139
Appendix G: Comments on Emission Analysis Plan 147
Appendix H: Pre-Publication Peer Review Comments 159
-------
1. Introduction
The basis of energy and emission estimation in MOVES2004 is the energy and
emission rates associated with each pollutant (including energy)a and emission process
estimated by MOVES. This report serves to document the development of these inputs.
Included in this report are the structure of emissions rates in terms of vehicle and
operating mode characterization, including supporting analyses; data sources used in the
development of emission rates; methods for producing emission rates, including the
"binning" process, and "hole filling" where adequate data were not available for a
particular rate; and advanced technology modeling using the Physical Emission Rate
Estimator (PERE) and other sources.
Energy and emission rates for all pollutants and processes are stored in a table in the
MOVES Default database named EmissionRate, with the exception of well-to-pump
energy and emissions, which are stored the table GREETWellToPump. All of the work
described in this report pertains to the development of entries in EmissionRate or
GREETWellToPump. These tables are therefore considered the "results" of the work
presented in this document. The default EmissionRate table has over 20,000 records, and
the process to generate advanced technology and future rates adds over 90,000 records, in
total accounting for the full range of energy and emission rates by factors including
pollutant, process, operating mode, fuel, engine technology, model year group, regulatory
class, loaded weight and engine displacement.
MOVES2004 implements a number of "firsts" with regard to EPA's mobile source
emission models, including: modeling energy consumption, N2O and CH4 explicitly;
defining rates by speed and power-based operating modes; modeling a broad array of
advanced technology vehicles, including several permutations of hybrid applications;
modeling periods of extended idling (e.g. heavy-duty "hoteling") as a separate emission
process; relying primarily on second-by-second data to develop emission rates; and
including well-to-pump energy emission estimates to enable well-to-wheel (e.g. life
cycle) analysis. Several of these features will be carried over and expanded in the
continued development of MOVES over the next few years. Thus, in addition to
documenting the development of specific inputs used in MOVES2004, this report
presents the foundation of methodologies which will be applied to future versions of
MOVES.
A report entitled "Draft Emission Analysis Plan for MOVES GHG", published in
Fall 2002, contains much of the background for the studies and methods documented
here. This report, referred to herein as the "emission analysis plan", underwent
stakeholder and formal peer review in accordance with EPA Peer Review guidelines.
Appendix G of this report contains a summary of comments from these review processes
along with responses to these comments. Because much of the background for the
a For brevity, within this report energy consumption is sometimes grouped within the term "emission"
or "pollutant".
-------
MOVES emission analysis work is covered in the emission analysis plan, the reader is
encouraged to consult with the emission analysis plan to provide a broader context for the
work presented here. A copy of the emission analysis plan may be requested directly via
email to mobile@epa.gov.
A pre-publication version of this report underwent formal peer review by Professor
Lawrence Caretto of California State University Northridge. The resulting comments and
our responses to these comments are contained in Appendix H.
This report is one of a series which documents the design, technical inputs and use of
MOVES2004. More detail on some of the concepts and methodologies presenting in this
report are found in the following reports:
• "MOVES2004 Software Design Reference Manual" (referenced herein as the
Design Manual), which provides detail on MOVES design, mathematical
formulation and database structure.
• "MOVES2004 Highway Vehicle Population and Activity Data" (referenced
herein as the Fleet and Activity Report), which provides detail on
methodology and data used to generated inputs related to parameters such as
vehicle population, vehicle miles traveled, average speed distributions, etc.
• "Fuel Consumption Modeling of Diesel, Motorcycle, and Advanced
Technology Vehicles in the Physical Emission Rate Estimator (PERE)"
(referenced herein as the PERE report), which provides detail on the PERE
model used to fill data holes and generate advanced technology rates.
• "Update of Methane and Nitrous Oxide Emission Factors for On-Highway
Vehicles", which provides detail on the analysis of CH4 and N2O test data
used to generate MOVES2004 emission rates.
• "GREET User Manual and Technical Input for MOVES Integration"
(referenced herein as the GREET report), which provides detail on the use of
and default assumptions in the version of Argonne National Laboratories'
GREET model developed to to provide well-to-pump energy and emissions in
MOVES2004
A more general overview of the MOVES model and documentation is contained in the
document "A Roadmap to MOVES2004". These reports will be available on the
MOVES website, http://www.epa.gov/otaq/ngm.htm, or may be requested directly via
email to mobile@epa.gov
-------
2. Overview of MOVES2004 Energy & Emission Inputs
Emission rates in MOVES2004 are categorized according to pollutant and emission
process. The pollutants in MOVES2004 are: total energy, fossil-based energy,
petroleum-based energy, methane (CFLt) and nitrous oxide (N2O).b Of these, however,
most are calculated in MOVES2004 from estimates of three "root" pollutants: total
energy (from which fossil energy and petroleum energy are calculated), CH4 and N2O.
Thus, the EmissionRate table only contains rates for total energy, CH4 and N2O.
The emission "processes" in MOVES2004 are: running operation, i.e. emissions from
on-road vehicle operation once the vehicle has warmed up; start operation, i.e. excess
emissions due to vehicle start-up; extended idle operation, i.e. emissions due to long
periods of off-network idling; and well-to-pump, i.e. emission produced during the
process of getting fuel from raw material to the pump. In MOVES2004, all emission
processes employ all pollutants, except extended idling, which applies only to energy
consumption.
Pollutant and process combinations vary significantly in terms of data availability,
which dictate the level of sophistication for which different combinations could be
modeled. Specifically, running energy consumption rates were developed with a great
level of detail, broken down by speed and power-based operating modes based on
analysis of millions of seconds worth of in-use data covering all vehicle types.
Conversely, CH4 and particularly N2O were based on relatively small samples of
laboratory bag (e.g. standard Federal Test Procedure) data, and were not split into
operating modes. Start energy consumption and emission rates where based largely on
FTP bag data, and hence are expressed as aggregate per-start quantities without further
breakdown. Very limited data are available on extended idle energy consumption rates
for heavy-duty trucks, hence these rates were developed as adjustments to in-traffic idle
rates based on a recent EPA study.
Cold start temperature effects and air conditioning usage were modeled as adjustment
factors in MOVES2004, and are documented in this report.
3. Overview of Data Used in MOVES2004
The MOVES model was designed to generate emission rates primarily from second-
by-second emissions data. This represents a large shift from MOBILE, which generated
emission rates from Federal Test Procedure (FTP) bag emission, with additional
corrections to account for "off-cycle" driving (although this shift was only employed for
running energy consumption in MOVES2004, while start energy consumption, CFLi and
N2O rates were developed from traditional bag data). Using second-by-second data
b Also presented in this report are inputs for the calculation of Atmospheric CO2 and CO2 equivalent.
However, these emissions are not included in MOVES2004 at this time. Atmospheric CO2 includes CO2
emitted directly from the tailpipe, plus HC and CO emitted from the tailpipe which is oxidized into CO2 in
the atmosphere.
-------
allows a much broader range of data to be used in developing emission rates, but also
opens up a host of issues regarding data quality assurance.
No new testing was performed specifically to provide data for MOVES2004. Instead,
data were compiled from previous EPA test programs and from several external sources,
quality assured and corrected, and (if not already in the database) entered into EPA's
Mobile Source Observation Database (MSOD).l EPA contracted with Eastern Research
Group Inc., (ERG) to assist in the acquisition, quality checking and compilation of data
collected by outside parties; the reports resulting from this work contain the specifics of
the test programs collected, and should be consulted for additional detail.2'3'4 Data
checking and loading software was developed by the contractor Britton Information
Systems.5'6 This data acquisition process was carried out over the course of a year and
resulted in changes in EPA software, hardware, and processes for obtaining,
documenting, and reformatting emissions data from non-agency data sources. This
section discusses the processes used to obtain, check, and distribute the data to the
MOVES modeling team followed with a description of the data sets used for MOVES.
3.1 Data Gathering and Processing Steps
The overall data acquisition and management process for all MOVES data followed
these steps (some steps were not necessary for data already in the possession of EPA, or
if the data were bagged emissions rather than second-by-second):
1. Identify candidate emission test programs based on literature search and contact
with non-EPA parties conducting emission research.
2. Determine suitability of emissions data and vehicle information data from these
test programs according to quality ranking guidelines for data and test program
documentation
3. Obtain the data from the data source if not currently with EPA.
4. Reformat the data for processing by EPA quality checking software
5. Check the data for errors and reasonableness (e.g. out-of-range, data
discontinuity, and analyzer "freezing").
6. Load the data into EPA's Mobile Source Observation Database (MSOD).
7. Check and (if needed) shift time alignment for second-by-second data based on
pre-determined algorithm correlating CO2 emissions and vehicle power.
8. Deliver aligned emissions data and vehicle information in a suitable format to the
MOVES modeling team for processing.
-------
The process on a dataset-by-dataset basis came to a halt for a variety of reasons.
Many data sets were incomplete in content and description. They could not physically be
loaded into MSOD. EPA and ERG had to contact the organization that originated the
data many times to make amendments or corrections. This rework was significant and
necessary to make the data useable. In the end some of the supplied data was not used.
3.2 Data Sources
The measurement and collection of second-by-second vehicle emissions data is not a
new practice. It has been done for the certification of diesel engine vehicles for total
hydrocarbon and oxides of nitrogen (NOX) for a decade. The practice is done for all
pollutants as a tool in the development of emissions control systems by manufacturers for
years and done in volume daily by emission and inspection programs across the United
States and Canada. However, in-use emission assessment programs are relatively new to
the practice.
EPA collected a great deal of second-by-second emission data in the development of
the I/M240 test procedure in the early 1990's and began to collect it on a routine basis
across all emissions factor test programs in 1997; in particular, a large effort was made to
collect second-by-second data on many vehicles over many inventory drive schedules to
support the development of MOBILE6, although the second-by-second data wasn't used
in MOBILE6 itself. These data are stored in MSOD and was available to MOVES
immediately.
The Agency's in-use emissions test program during that period was small in
comparison to past efforts, however, and the volume of data on-hand was considered
insufficient for MOVES. To supplement the MOVES2004 dataset, a list of potential data
sources was compiled by MOVES team members and presented to ERG, who was given
the task of contacting the principle investigators and determining the suitability of the
data based on quality ranking criteria developed by EPA. ERG then obtained
documentation on the test programs the data represented.
Although second-by-second data was the focus of the data gathering effort, bag
emission data was also sought where available, particularly in some cases where we knew
second-by-second data wouldn't be used - primarily for N2O emissions.
Sources for test data outside of EPA were:
• California Air Resources Board (CARB)
• Development of Unified Correction cycles (UCC) in 1996
• 16th Vehicle Surveillance Program with N2O bag data
• Coordinating Research Council (CRC)
• Study in 1997 to determine the effects of sulfur levels in fuel on vehicles
(E-42 & -47)
• Auto/Oil Air Quality Improvement Research Program (early 1990's)
-------
• Study in 2002 on Heavy -duty Vehicle Chassis Dynamometer Testing for
Emissions Inventory (E-55)
• The New York State Instrumentation/Protocol Assessment Study which
compared the standard IM240 test procedure and instrumentation with the New York
Transient Emissions Short Test (NYTEST)
• North Carolina State University study to determine the emissions savings that
could be achieved through better traffic management
• University of California Riverside College of Engineering Center for
Environmental Research and Technology (CE-CERT)
• NCHRP 25-11 Comprehensive Modal Emission Model and Vehicle Emissions
Database, Version 2.02
• Ammonia from Light-duty vehicles
• Heavy-Duty Diesel Truck study
• Environment Canada's study on the Effects of Aged Catalysts and Cold Ambient
Temperatures on Nitrous Oxide Emissions
• Texas Department of Transportation and the University of Texas study in 2002-
03 on the use of new fuels in heavy -duty diesel vehicles
• West Virginia University testing of heavy duty vehicles using their portable
dynamometer (four distinct datasets were obtained)
The above-referenced ERG reports cover the efforts to obtain, format, and deliver the
data to EPA. These reports cover the number of tests, the number and vehicle types
tested, test procedures followed, test conditions, and test fuels. The reports indicated the
limitations of the test data and quality of the documentation for the test programs that the
data represented.
In addition to the programs listed above, this effort also gathered second-by-second
data on tens of thousands of vehicles from Inspection / Maintenance (I/M) Programs:
• Arizona I/M program (Car Care). Data from January 1, 2002 to June 30, 2002
• British Columbia I/M program (Air Care). Data from January 1, 2001 to June 3,
2002
• Colorado (I/M) program (Air Care). Data from January 1, 1999, to September 1,
2002
These data were not used in MOVES2004, however, since I/M program data introduces
another layer of data concerns which couldn't be adequately addressed in time for
inclusion in the model. We will be considering the use of these data, along with
additional I/M programs and remote sensing device (RSD) data, for the criteria pollutant
version of MOVES.
-------
All data used in MOVES2004, and/or available in the MSOD, can be obtained via
request to EPA. Interested parties may request the data via email to the mobile@epa.gov
inbox.
4. Total Energy Rates
4.1 Source Bin Definitions
Source bins are groupings of parameters which distinguish differences in energy and
emission rates according to physical differences in the source, e.g. weight, engine size,
model year group etc. In MOVES the entire on-road fleet is characterized by source
bins, which are mapped to the activity-based Source Use Types within the model. In the
MOVES design, source bin definitions are able to vary by pollutant, since in the real
world the vehicle characteristics which are most influential to energy consumption and
emissions do vary depending on the pollutant. The reader should consult the Design
Manual for additional background on these concepts.
Source bins fields and the definitions within each field were based on quantitative
analysis of available data to determine the most important variables to consider, and
important breakpoints to account for within each field; and qualitative assessment of
important categories to define for full coverage of the current and future fleet, or
consistency with existing methodologies.
Source bin fields for energy consumption were developed based on a quantitative
assessment of the most important factors influencing CO2 emissions (as a surrogate for
fuel and energy consumption) performed by ERG.7 This work analyzed several thousand
chassis dynamometer emission tests on light-duty gasoline and diesel vehicles and
several hundred chassis dynamometer emission tests on heavy-duty diesel vehicles. A
ranking of the most important factors to consider in modeling CO2 and CH4 emissions
was developed. The analysis was performed using ANOVA, with the pollutant as the
dependent variable and numerous vehicle, fuel, and operating parameters as the
independent variables. The analysis examined both the partial r2 (the contribution of a
particular factor in explaining overall variability) and the relative size of the parameter
estimate of a particular factor. Based on this analysis, ERG ranked the factors as "very
important," "somewhat important," "not important," and "not significant."
For CO2, the conclusion we derived from the ERG analysis is that power (for the
ERG analysis characterized by average total power, since emission results were only
available at the bag level) is by far the most critical factor for both the light-duty gasoline
and heavy-duty diesel vehicles, followed by which driving cycle the vehicle was run on.
The most important vehicle characteristic was the ratio of vehicle displacement to weight,
which was a surrogate for vehicle power/weight ratio. Many variables one would
assume are influential turned out not to be, such as start versus running or vehicle
mileage. One anomaly is the importance of air conditioning operation, which was
classified as "not important" in the ERG analysis. This result was likely driven by the
-------
fact only a small portion of the dataset was collected when the air conditioning was on.
Another study, which focused specifically on this subset of data, found significant
differences in fuel consumption when the air conditioning was engaged.8
The ERG analysis led us to conclude that it was important to define source bins
by vehicle (loaded) weight and engine displacement (the most important parameters of
average total power and driving cycle are accounted for with the use of operating mode
bins described in Section 4.2). We further expanded source bins to make sure that
MOVES had the flexibility to model different mixes of vehicle technologies - adding
fuel type, engine technology and model year group. The source bin fields and the bin
definitions are shown in Table 4-1.
Table 4-1 - MOVES Source Bin Definitions for Total Energy
Fuel Type
Gas
Diesel
CNG
LPG
Ethanol (E85)
Methanol (E85)
GasH2
Liquid H2
Electric
Engine Technology
Conventional 1C (CIC)
Advanced 1C (AIC)
Moderate Hybrid - CIC
Full Hybrid - CIC
Moderate Hybrid - AIC
Full Hybrid - AIC
Fuel Cell
Hybrid - Fuel Cell
(See Table 4- 14 for
combinations of fuel type
and engine technology used
in MOVES2004)
Model Year
Group
1980 and earlier
1981-85
1986-90
1991-2000
2001-2010
2011-2020
2021 and later
Loaded
Weight
Null
<= 2000 Ibs
2001-2500
2501-3000
3001-3500
3501-4000
4001-4500
4501-5000
5001-6000
6001-7000
7001-8000
8001-9000
9001-10,000
10,001-14,000
14,001-16,000
16,001-19,500
19,501-26,000
26,001-33,000
33,001-40,000
40,001-50,000
50,001-60,000
60,001-80,000
80,001-100,000
100,001-130,000
>=130,001
Engine Size
Null
< 2.0 liters
2.1-2.5 liters
2.6-3.0 liters
3. 1-3.5 liters
3.6-4.0 liters
4. 1-5.0 liters
> 5.0 liters
A source bin is a unique combination of values across each category. For example,
all gasoline vehicles, conventional internal combustion engines, model year 1991 - 2000,
loaded weight 2501 - 3000 Ibs, engine displacement <2.0 liters would define a single
source bin with a unique set of emission rates. It is important to note that, for total
energy, traditional definitions of vehicle regulatory class (e.g. LDV, LDT1, LDT2 etc.)
are not part of the source bin definition for total energy. While vehicle class inherently
influences tailpipe emission results, energy consumption is primarily a function of fuel,
technology, vehicle weight, engine size, and model year regardless of whether it is a car
or a truck.
To keep track of the myriad of source bins in the EmissionRate data table, a 19-
digit Source Bin ID schema has been developed. While seemingly Byzantine at first,
we're confident that MOVES users will learn to recognize Source Bin IDs as quickly as
10
-------
their own phone numbers. Table 4-2 shows the source bin schema, which differs for total
energy, CH4 and N2O.
Table 4-2: Source Bin ID Schema
Digit
1
2-3
4-5
6-7
8-9
10-13
14-17
18-19
Field
Leading digit
Fuel type
Engine Technology
Regulatory Class (CH4,
N20 only)
Model Year Group
Engine Size
Loaded Weight
Trailing zeros
The key values (IDs) for each field can be located in the Design Manual, or by
directly querying the FuelType, EngineTech, ModelYearGroup, EngineSize and
WeightClass tables in the MOVES default database.
4.2 Operating Mode Definitions
We subdivide total activity into categories that differentiate emissions, known as
operating mode bins. Operating mode bin definitions are allowed to vary by pollutant
and emission process, to accommodate variability in the important activity-based effects
on energy and emissions. 17 operating mode bins have been defined for running total
energy; braking, idle, and 15 subdivisions of cruise and acceleration defined by
instantaneous vehicle speed and vehicle specific power (VSP). A detailed account of
how these bins were established, plus proof-of-concept validation analysis, is found in
Appendix A. The specific bin definitions for total energy are shown in Table 4-3.
Table 4-3 Operating Mode Definitions for Running Total Energy
Braking (Bin 0)
Idle (Bin 1)
VSP \ Speed
< 0 kw/tonne
Oto3
3 to 6
6 to 9
9 to 12
12 and greater
6 to 12
<6
0-25mph
Bin 11
Bin 12
Bin 13
Bin 14
Bin 15
Bin 16
25-50
Bin 21
Bin 22
Bin 23
Bin 24
Bin 25
Bin 26
>50
Bin 36
Bin 3 5
Bin 3 3
The Bin ID numbers shown in the tables are the key field IDs used in the
EmissionRate table to identify the bins. Emission rates in the EmissionRate table are
keyed by a) source bin, b) pollutant / process, and c) operating mode bin.
11
-------
Figure 4-1 provides a graphical example of energy rates by operating mode bin
for two different source bins differentiated only by weight class: conventional gasoline
1986-1990 model year vehicles with engine size in 2.0 - 2.5 liter range, with a loaded
weight of 2501 - 3000 pounds, and 3501 - 4000 pounds.
Figure 4-1: Energy Rates by Operating Mode for Two Source Bins
Source Bin: Gasoline / 86-90 MY / 2.0-2.5 liter
• 2501-3000 Ibs
3501-4000 Ibs
100.0
80.0
S
CO
OL
60.0
0)
c
UJ
i 40.0
0
20.0
0.0
25 mph|
25 -50 mph
VSP (KW/tonne)
4.3 Running Energy Rate Development
4.3.1 Data Used
4.3.1.1 EPA Test Programs
Data used to generate total energy rates were from the EPA and non-EPA test
programs discussed in Section 3. All data used for generating total energy rates were
second-by-second (1 Hz) resolution data. The EPA data sources used for total energy
rates in MOVES are found summarized in Table 4-4. Many of the driving cycles listed
under "test cycles" are the facility-based cycles developed for use in MOBILE6 (e.g.
FWY or ART, which stand for freeway or arterial, along with a Level of Service
designation); more detail on these cycles can be found in MOBILE6 reports documenting
12
-------
their development.9 Under the "seconds of data" heading, "raw" is the number of
seconds of data from each test program in the MSOD, before any processing or filtering.
"After QA/QC" is what remained after the process discussed in Section 3.1; this was the
sample passed to the binning program discussed in the next section. "After filtering" is
what remained after the filtering performed by the binner program, discussed in
Appendix B; this is the sample used directly in developing the total energy rates.
Table 4-4: EPA-sponsored test programs supplying data to MOVES2004
MSOD
Program ID
98N2OA
CDHOT_PM_A
CYCLES_A
LDV_AC_A
LHDDT A
LHDT_A
Description
Nitrous Oxide (N20) Study on Tier 1 LDVs, LDTs, and
LEVs (LDVs) at various mileages.
Determining the Relationship of Opacity and Exhaust
Emissions (Including Total PM) in In-use Gasoline
Powered Vehicles During an IM240
Determining Basic Exhaust Emission Rates for Light Duty
Cars and Trucks using Multiple Drive Schedules and with
the Air Conditioning On and Off
Determining Basic Exhaust Emission Rates for Light Duty
Cars and Trucks using Multiple Drive Schedules
Investigation on diesel LHDT exhaust emissions on various
cycles, play-loads including measuring for toxics, PM and
unregulated pollutants.
Investigation of gasoline 10 LHDTs on 8 driving cycles at
different loads, start emissions, IMs, FTPs and different
fuel sulfur levels
Vehicles
TT J
Used
23
107
43
62
6
10
MY
Range
1995-99
1969-98
1990-96
1983-96
1998-99
1991-97
Test
/~i nlnn
Cycles
F505
FTP
FWY-AC
US06
3IM240
ART-AB
ART-CD
ART-EF
FWY-AC
FWY-D
FWY-E
FWY-F
FWY-G
FWY-HI
LA4
LA92
LOCAL
NONFRW
NYCC
RAMP
ST01
US06
ART-AB
ART-CD
ART-EF
FWY-AC
FWY-D
FWY-E
FWY-F
FWY-G
FWY-HI
LA92
LOCAL
NONFRW
NYCC
RAMP
ST01
ART-CD
FTP
FWY-E
FWY-G
FWY-HI
LA92
NYCC
ST01
US06
ART-EF
F505
FTP
FWY-D
FWY-F
Seconds of Data
Raw
171,240
74,673
610,671
671,704
74,902
399,258
After
QA/QC
170,579
74,488
606,385
670,367
74,828
397,611
After
filtering
77,255
74,176
349,767
552,492
73,304
387,097
13
-------
LHDT B
LHDT C
LHDT_EVAP
LHDT LOT
OBD A
OEM_2100
ROVER_A
SHOOT_OUTA
TIER_1
Investigation on gasoline LHDT exhaust emission on
various driving cycles, payloads, and fuels (sulfur)
Determining Basic Exhaust Emission Rates for Light
Heavy Duty Trucks using Multiple Drive Schedules and
Payloads
LHDTs with evaporative emission tests at 3 different temp.
ranges and two different fuels (6.3 and 9.0) including FTPs.
Gas cap on/off followed .
Inventory Cycles/LA92 Exhaust Emissions Data Collection
and Amendment #1 Sulfur Fuel Testing
Determining the Effectiveness of Onboard Diagnostic
Systems in Identifying Vehicles that Fail the FTP
An Investigation of OEM 2100's Capabilities to Accurately
Measure Emissions of Late-Model Gasoline Vehicles
Determining the Viability of Gathering with ROVER
Exhaust and Vehicle Information for Diesel and Gasoline
Powered Light-Heavy Duty Trucks
On- Road Emission Test Data from 15 Light- Duty Vehicles
and 15 Heavy Duty Diesel Trucks for On-Board Emission
Data Analysis and Collection for the New Generation
Model
Determining Basic Exhaust Emission Rates for Tier 1 Light
Duty Cars using Multiple Drive Schedules
2
18
2
46
47
2
8
35
38
1993-95
1989-97
1993-97
1988-97
1996-98
1996-98
1998-99
1992-00
1993-97
FWY-G
FWY-HI
IM240
LA92
NYCC
RAMP
ST01
US06
ART-CD
FTP
FWY-D
FWY-E
FWY-F
FWY-G
FWY-HI
IM240
LA92
NYCC
ST01
US06
ART-CD
F505
FTP
FWY-D
FWY-E
FWY-F
FWY-G
FWY-HI
LA92
NYCC
ST01
US06
F505
FTP
IM240
US06
ART-EF
F505
FTP
FWY-D
FWY-F
FWY-G
FWY-HI
IM240
LA92
NYCC
RAMP
ST01
US06
FTP
IM240
FTP
FWY-HI
NYCC
US06
FTP
FWY-HI
NYCC
US06
FTP
LA4
US06
ART-EF
F505
FTP
FWY-F
FWY-HI
IM240
73,489
365,572
7,949
792,700
105,573
15,962
59,031
40,509
289,768
73,196
365,164
7,877
791,659
105,416
15,950
59,016
40,479
287,161
71,814
357,842
7,647
773,893
100,852
15,512
57,272
38,716
278,764
14
-------
TIER IB
Investigation on gasoline vehicles/trucks/SUVs over
different driving cycles including SFTP
15
1996-00
LA92
NYCC
ST01
US06
FTP
FWY-HI
LA92
NYCC
SC03
ST01
US06
93,585
93,359
90,159
The EPA programs were performed in the Ann Arbor, Michigan EPA laboratory
or at one of two contractor labs: Automotive Testing Laboratory or Southwest Research
Institute. The vast majority of vehicles in the program are in-use vehicles recruited from
the public. Many of the LHDTs and both vehicles in the 'OEM_2100" test program
were leases from vehicle dealers or vehicle rental agencies. Several programs were
assessments of new methodologies to measure emissions with on-board instrumentation
("ROVER_A", "OEM_2100", and "SHOOT_OUTA") - although the data used in
MOVES was from the laboratory measurements from those programs. Two of the
programs were focused on the characterization of two emission components of interest in
SI vehicles N2O ("98N2O_A") and particulate matter ('CDHOT_PM_A') where the
principal and typical emission components were collected. All the tests followed typical
EPA test documentation procedure, resulting in complete and extensive documentation.
The EPA programs were not subject to the ERG data acquisition process, but
followed the data checking, loading, alignment, and delivery parts of the MOVES data
process. An EPA test's second by second data was excluded from the MOVES modeling
team delivery if any of the aggregated second by second constituents failed to be within
10 percent of the bag analysis values. The various programs reported as being delivered
to MOVES have statements of work describing how EPA thinks the work should be
done, a work plan delivered by testing contractor that describes how the work will be
done, and a final report describing what work was done. These reports are all available in
electronic form from the Data Acquisition and Management Team of the Assessment and
Standards Division, and can be obtained upon request to mobile@epa.gov.
4.3.1.2 Non-EPA Test Programs
There were sixteen groups of data either delivered to EPA from ERG, or already
in the MSOD, that were originally targeted for MOVES. Thirteen were emission
characterization programs and the three were emission tests from state or regional
inspection and maintenance programs (these data were not used in MOVES2004). In all
there were 10,760 vehicles and 35,489 tests in the characterization programs, of which
9,161 vehicles and 34,901 tests loaded in MSOD; the excluded characterization data was
not considered to be of sufficient quality to be useful.
A summary of non-EPA programs used in MOVES2004 are shown in Table 4-5.
Table 4-5: Test programs supplying data to MOVES2004 not conducted by EPA
15
-------
MSOD
Program ID
CARBJJCC96
CECERT_NH3
CRC_E55_59
CRC_S_LDV1
GRANT97_NY
NCHRP
NYIPA/
NYIPA2002
WVU1-4
Description
California Air Resources Board (CARB)
developement of Unified Correction Cycles (UCC)
in 1996
University of California Riverside College of
Engineering Center for Environmental Research
and Technology (CE CERT) Emissions of
Ammonia for Light -duty vehicles
Coordinating Research Council (CRC) study in
2002 on Heavy-duty Vehicle Chassis Dynamometer
Testing for Emissions Inventory.
Coordinating Research Council (CRC) study in
1997 to determine the effects of sulfur levels in fuel
on vehicles.
NewYork/DEC-Characterization and Control of HD
Diesel Vehicle Emissions in the New York
Metropolitan Area
University of California Riverside College of
Engineering Center for Environmental Research
and Technology (CE_CERT) NCHRP 25-1 1
Comprehensive Modal Emissions Model and
Vehicle Emissions Database, Version 2.02
Second Iteration of the New York State
Instrumentation/Protocol Assessment Study which
compares the standard IM240 test procedure and
instrumentation with the New York Transient
Emissions Short Test (NYTEST).
West Virginia University testing of heavy duty
vehicles using their portable dynamometer.
No. of
Vehicles
Used
42
35
25
12
35
337
9900
149
MY
Range
1973-94
1983-01
1973-00
1997
1966-99
1965-99
1992-01
1991-99
Test Cycles
FTP
LA92
UCC 15
UCC20
UCC25
UCC30
UCC35
UCC40
UCC45
UCC50
FTP
FWY
NYCC
US06
AC5080
CARB-C
CARB-I
CARB-R
CARB-T
CARBCL
FTP
SPEAK
UDDS D
WVUCBD
FTP
MEC5
MEC6
MEC7
SMEC6
SMEC7
US06
IM240
2-5MIL
2CSHVR
2TESTD
3CBD
SMILE
CBD
CSHVR
DRT
HVDUTY
KERN
NYBUS
NYCCT
NYGT2
RT22
RT77
TEST D
UDDS W
VFAC
WHM
Seconds of Data
Raw
496,827
82,262
631,380
618,567
257,101
1,379,200
5,945,603
1,634,576
After
QA/QC
495,825
82,262
473,589
616,719
257,101
1,378,101
5,902,463
1,558,382
After
filtering
487,112
78,278
472,155
582,122
256,303
720,911
5,720,688
1,554,281
The "MEC" and "SMEC" cycles run as part of the NCHRP program were
ultimately removed from the data used for MOVES2004. They are cycles engineered for
populating UC Riverside's Comprehensive Modal Emission Model (CMEM). These
cycle contain a high proportion of very aggressive driving (e.g. wide open throttle) which
16
-------
skewed the bin results. A similar argument could be made for US06, but the same
analysis determined that the US06 was not as extreme an outlier as the MEC and SMEC
cycles - and the US06 is based on real in-use driving. This analysis is documented in a
presentation made at the 2004 CRC On-Road Emission Workshop.10
4.3.2 Generating Rates from Available Data
Total Energy Rates were developed by binning second-by-second data from the
test programs described in Section 4.3.1 for each intersection of source bin and operating
mode bin (termed "cells"). "Binning" means computing the average energy rate (energy
use per time - e.g. KJ per hour) for each cell across all the seconds of data falling within
that cell.
To do this in an automated way, a MySQL script termed the "binner program"
was written that performed the following steps, starting with an input of second-by-
second data compiled from the data sources listed in Section 4.3.1: 1) calculate vehicle
specific power (VSP) for each second of data; 2) determine the source bin and operating
mode "cell" for each second of data; 3) calculated second-by-second energy
consumption from HC, CO and CC>2 mass emissions based on carbon balance and energy
content calculations, and 4) calculate, for each cell, the average energy rate, coefficient of
variation, and a number of diagnostic statistics. The output of the binner program was a
MySQL table containing energy rates (in KJ per hour) and coefficient of variation (CV)
by source bin and operating mode bin, in the same format of the MOVES emission rate
database table. Detailed documentation on the binner program are in Appendix B,
including the algorithms used for performing these three steps.
4.3.3 Hole Filling
The binner program directly populated a significant number of source bin and
operating mode cells with data. However, a number of cells were not populated because
either no data existed to populate them with, or the data which did exist failed the data
quality objectives set out in the binner process - i.e. a cell must have data from at least 3
tests and have a coefficient of variation (CV) of less than 0.50. The cells left unpopulated
are dubbed "holes" which required supplemental methods to populate.
The two methods used for hole-filling were 1) employing the Physical Emission
Rate Estimator (PERE) to fill holes directly, and 2) interpolation or copying of
neighboring cells populated with data. These methods were chosen based on a proof-of-
concept evaluation which showed the efficacy of these methods (Appendix C). As
discussed in Appendix C, an additional method initially evaluated for hole-filling but not
ultimately used for MOVES2004 was to derive modal rates from aggregate bag data.
After the initial binner run, an analysis was performed which identified the
fraction of the fleet covered by the binned data, and the relative importance of the
remaining holes. Table 4-6 shows the estimated fraction of the fleet covered by just the
binned second-by-second data, by source use type.
17
-------
Table 4-6: Fraction of 1999 on-road fleet covered by MSOD data
Source Use Type
Motorcycle
Passenger Car
Passenger Truck
Light Commercial Truck
Intercity Bus
Transit Bus
School Bus
Refuse Truck
Single-Unit Short Haul
Single-Unit Long Haul
Motorhome
Combination Short-Haul
Combination Long-Haul
Fraction Covered
0
0.98
0.93
0.87
1.0
0.99
0.84
0.86
0.65
0.65
0.58
0.36
0.24
As shown, the light-duty source types are covered fairly well (over 90 percent) by
the existing data, with coverage dropping off for the heavier-duty classes - single-unit
and particularly combination trucks, where for long-haul trucks about a quarter of the
fleet is covered.
From this analysis a list of source bins holes was created, ranked according to
their importance in terms of the relative fraction of the fleet - in essence a ranking of the
most important holes to fill. The highest priority holes were defined as those which a)
were needed to bring the fleet coverage up to 95 percent within each source use type, and
b) represented at least 2 percent of the fleet within a source use type. These priority
holes were generally filled by PERE, with remaining holes left for interpolation. As
expected from the results in Table 4-6, the most important holes to fill were the heaviest
weight bins making up the combination trucks. In particular, source bins representing
loaded weights of 80,000 pounds and higher are a significant portion of heavy-duty
activity (according to the U.S. Census Bureau's Vehicle Inventory and Use Survey, or
VIUS, as discussed in the Fleet and Activity report); but because available heavy-truck
emission data was limited to approximately 60,000 Ibs, no in-use data were available for
the heaviest trucks.
4.3.3.1 Hole Filling with PERE
The Physical Emission Rate Estimator (PERE) is a stand-alone spreadsheet model
developed to fill data gaps in MOVES and to help it extrapolate to future projections of
energy and emissions. The details of the PERE model are in a separate document.n Its
inputs are vehicle parameters and second-by-second driving traces, and it outputs second-
by-second fuel consumption rates for the running operating (e.g. it does not yet include
starts). PERE uses physical principles to model propulsion systems in the vehicle. The
model is based on a simple model for the internal combustion engine. Simulation of
18
-------
hybridization is achieved by inserting a secondary power source and energy storage
device (usually a battery/motor combination).
Aside from advanced technologies, PERE was used to fill several holes in the
current fleet for MOVES2004, including most of the top-priority holes where test data
was lacking or deemed insufficient. To fill a source bin hole, PERE was run once for
each source bin over a series of representative driving schedules. This required choosing
vehicle parameter inputs which represented an entire source bin. The parameter selection
for PERE is a somewhat involved process. The decision tree is shown in Figure 4-2. The
parent variables (underlined) are above the dependent variables. The parent variables
define the decision source of the parameters. The dependent variables are the parameters
that are determined from the parent variable. Some dependent parameters vary with
multiple parents (e.g. engine friction). Some other parameters are fixed.
Figure 4-2: Decision Tree for Determine PERE Inputs for a Source Bin
SOURCE BIN
Weight
T nflSTnfrm
- Road load: Cr, Cd:
AF, orA,B, C
Enrichment
Threshold: FRth
i
r
Model Year
- Engine Friction:
kO, kl
- peak bmep fit
- Enrichment
Threshold: FRth
Engine Size
- Peak torq fit
- Enrichment
Threshold: FRth
Gas/Diesel
- Ind Efficiency: r|
- Engine Friction: kO, kl
- peak bmep fit
- fuel: LHV, pf
LD/MD/HD/Motorcycle
- Driving cycle, Transmission: N/v, shift points, gear ratio, manual/auto, accessory: Pace
FIXED PARAMETERS
Auto transmission efficiency (by gear), manual transmissions efficiency, density of air, fuel
enrichment slope (gasoline only), inertial rotational mass term.
The determination of general vehicle weight classification is the very first step in
the parsing process. Since MOVES does not have explicit splits for light, medium and
heavy duty, PERE assumes that the light to medium split occurs at weighclassID 80
(7,000-8,000 Ibs), where above this weight bin is medium duty (single unit delivery
trucks, buses, etc). The medium to heavy duty split occurs at weightclassID 400 (33,000-
40,000 Ibs). This in turn defines whether the transmission will be a 5-speed automatic
(LD), 6-speed automatic (MD), 12-speed manual (HD), or 5-speed manual (motorcycle).
Motorcycles have their own weight and engine size categories as well as a different
accessory loading (0.25 vs 0.75kW).
The weight determines the road-load coefficients for the vehicles directly. For
model years 2000 to present, A, B, and C track coefficients are provided by
19
-------
manufacturers. Older vehicles only have the single Tractive Road Load Horsepower
(TRLHP) figure. The TRLHP for a typical light-duty vehicle (and truck) in the source bin
can be approximated from weight using the relation used in the binner program, detailed
in Appendix B.
The weight class also defines the driving cycles that PERE will run. Before
model year 1999, weight also is a term in the enrichment equation. After 1999, the
vehicles are assumed to go into enrichment so rarely that it is ignored by PERE.
The model year also affects the performance of the engine, both in terms of
friction and peak bmep (which mainly influences power downshifting). The engine/fuel
type determines the efficiency and friction characteristics of the engine, and it also
defines the fuel parameters since gasoline and diesel have different physical properties.
Table 4-7 show an example source bin filled by PERE
Table 4-7: Example Source Bin Filled by PERE
SourceBinID: 1020100045099080000
Digit
1
2-3
4-5
6-7
8-9
10-13
14-17
18-19
Field
Leading digit
Fuel type
Engine Technology
Regulatory Class
Model Year Group
Engine Size
Loaded Weight
Trailing zeros
Code
1
02
01
00
04
5099
0800
00
Meaning
-
Diesel
Conventional Internal Combustion
- (not used for energy)
1991-1999
> 5.0 Liters
60,000 - 80,000 Ibs
-
PERE input
1995
6.9 Liters
70,000 Ibs
Using the information above and the descriptions in the PERE documentation, the
PERE input selected to fill this source bin are shown in Table 4-8.
20
-------
Table 4-8: PERE Inputs For Filling Source Bin 1020100045099080000
Vehicle
Model Year
Vehicle wgt (kg)
CrO (rolling resistance)
Cd (drag coeff)
A (frontal area mA2)
A(N)
B (N/mps)
C (N/mpsA2)
Pace (accessory - kW)
Engine
Engine Displ (L)
kO (N indep friction kJ/Lrev)
k1 (N dependent fric)
P/T indicated eff (eta)
Transmission
N/v (rpm/mph)
Nidle (rpm)
trans eff
Shift point 1-2 (mph)
Shift point 2-3
Shift point 3-4
Shift point 4-5
Shift point 5-6
Shift point 6-7
Shift point 7-8
Shift point 8-9
Shift point 9-10
Shift point 10-11
Shift point 11 -12
Shift point 12-13
g/gtop 1
g/gtop 2
g/gtop 3
g/gtop 4
g/gtop 5
g/gtop 6
g/gtop 7
g/gtop 8
g/gtop 9
g/gtop 1 0
g/gtop 1 1
g/gtop 12
g/gtop 1 3
Fuel
LHV (kJ/g)
density gas (kg/L)
1995
31752
N/A
N/A
N/A
2098.81
0.000
4.2268
0.75
6.9
0.0605
0.00333
0.48
26.7
700
0.95
2.48
4.75
7.75
13.5
17.5
23
34
50
56
57
64
64
13
9.4
6.9
5
3.7
2.7
2
1.5
1.2
1.1
0.93
0.86
0.86
43.2
0.8114
21
-------
Once the vehicle inputs were determined, PERE was run over a series of
driving schedules (cycles) to generate second-by-second energy consumption.
Binned energy consumption rates for MOVES were generated from the second-by-
second PERE results using the same process the binner program used with test data,
as described in Section 4.3.2.
The binned energy consumption and emissions rates are dependent on the driving
cycle input. For PERE, the driving cycle is (in turn) dependent on the weight of the
vehicle. It is important to capture a representative sampling of real-world driving.
Fourteen driving cycles for light duty applications have been selected for this
purpose from MOVES, shown in Table 4-9. These are a subset of the schedules
used by MOVES to generate operating mode distributions for the calculation of total
energy consumption (the cycles themselves can be queried in the
Drive Schedule Second table of the MOVES default database). These cycles
represent a broad spectrum of driving, from very low to very high speeds. When
merged, the cycles run 6,981 seconds. In the future it is likely that similar binned
rates could be obtained from a smaller sample of driving cycles, thus making the
execution of PERE easier to manage.
Table 4-9: Light-duty driving cycles used as input to PERE
MOVES
SchedulelD
156
155
154
153
152
151
199
105
104
103
102
157
158
101
Cycle
FWYHI1
FWYAC
FWYD
FWYE
FWYF
FWYG
RAMP
ARTAB
ARTCD
ARTEF
NYCC
FWYHI2
FWYHI3
LOWSPEED1
Avg Spd (mph)
63.2
59.7
52.9
30.5
18.6
13.1
N/A
24.8
19.2
11.6
7.1
68.2
76.0
2.5
For medium and heavy-duty applications, the driving cycles where drawn from
the medium and heavy-duty schedules used to populate the MOVES default database,
described in the Fleet and Activity report (and also contained in the
DriveScheduleSecond table). The medium duty drive cycles are listed in Table 4-10 and
totaled 6,050 seconds.
22
-------
Table 4-10: MOVES Medium duty driving cycles used as input to PERE
MOVES
SchedulelD
201
202
203
204
205
206
251
252
253
254
Cycle
MD 5mph Non-Freeway
MD lOmph Non-Freeway
MD 15mph Non-Freeway
MD 20mph Non-Freeway
MD 25mph Non-Freeway
MD 3 Omph Non-Freeway
MD 3 Omph Freeway
MD 40mph Freeway
MD 50mph Freeway
MD 60mph Freeway
Avg Spd (mph)
1.8
10.5
15.6
20.4
24.4
30.8
37.4
45.3
55.5
60.1
The heavy-duty cycles (Table 4-11) are also a subset from the MOVES drive
schedules. Together they total 28,313 seconds of driving, weighted mainly with freeway
driving. The length of this file makes the execution of PERE cumbersome for heavy-
duty, and in the future, the number of cycles will be reduced.
Table 4-11: Heavy-duty driving schedules used as input to PERE
MOVES
SchedulelD
301
302
303
304
305
306
351
352
353
354
Cycle
HD 5mph Non-Freeway
HD 1 Omph Non-Freeway
HD 15mph Non-Freeway
HD 20mph Non-Freeway
HD 25mph Non-Freeway
HD 3 Omph Non-Freeway
HD 3 Omph Freeway
HD 40mph Freeway
HD 50mph Freeway
HD 60mph Freeway
Avg Spd (mph)
1.2
10.8
15.2
19.8
24.9
30.8
34.9
46.9
54.3
59.5
The motorcycles are run only on the FTP and HWY driving cycles; small engine
scooters (under 500 pounds) were run on only the reduced speed FTP scooter cycle.
With the driving cycles entered, a binning macro (on the PERE spreadsheet) is
run in order to bin the fuel consumption rates according the 17 VSP / speed bins used by
MOVES, which were in turn converted to MOVES energy rates.
Tables 4-12 and 4-13 show the source bin holes filled by PERE for all vehicle
classes except motorcycles. The source bins in Table 4-12 were chosen based on their
relative importance in the fleet (the "rank" category signifies the rank of importance in
23
-------
terms of fraction of the overall in-use fleet). Table 4-13 are less important "leftover" bins
which where not filled by the interpolation / copying process described in Section
4.3.3.2, generally because there were enough neighbor bins present to satisfy the criteria
for this method.
Table 4-12 Priority Holes Filled by PERE
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
25
26
27
28
29
32
35
36
37
38
40
42
43
44
62
65
SourceBin ID
1020100045099080000
1020100045099100000
1020100055099080000
1010100045099016000
1010100055099016000
1020100055099100000
1010100045099019500
1020100055099014000
1020100055099016000
1020100045099016000
1020100045099004500
1020100045099005000
1020100045099006000
1010100055099026000
1010100032025006000
1020100043540014000
1020100055099009000
1010100045099026000
1020100045099003500
1010100055099019500
1020100045099004000
1010100055099033000
1020100055099008000
1020100055099007000
1020100045099130000
1020100055099010000
1010100044050002500
1010100053035014000
1010100044050009000
1010100044050007000
1010100045099033000
1010100055099004000
1010100055099003500
1010100055099002500
1010100055099003000
1020100043540019500
1020100043540007000
Fuel
Diesel
Diesel
Diesel
Gas
Gas
Diesel
Gas
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Gas
Gas
Diesel
Diesel
Gas
Diesel
Gas
Diesel
Gas
Diesel
Diesel
Diesel
Diesel
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Diesel
Diesel
MY
Group
91-00
91-00
<90
91-00
01-10
<90
91-00
<90
<90
91-00
91-00
91-00
91-00
01-10
86-90
91-00
<90
91-00
91-00
01-10
91-00
01-10
<90
<90
91-00
<90
91-00
01-10
91-00
91-00
91-00
01-10
01-10
01-10
01-10
91-00
91-00
Engine Size
Class (L)
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
2.0-2.5
3.5-4.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
>5.0
4.0-5.0
3.0-3.5
4.0-5.0
4.0-5.0
>5.0
>5.0
>5.0
>5.0
>5.0
3.5-4.0
3.5-4.0
Loaded Weight
Class (Ibs)
60-80K
80-100K
60-80K
14-16K
14-16K
80-100K
16-19.5K
10-14K
14-16K
14-16K
4-4.5K
4.5-5K
5-6K
19.5-26K
5-6K
10-14K
8-9K
19.5-26K
3-3. 5K
16-19.5K
3.5-4K
26-3 3K
7-8K
6-7K
100-130K
80-100K
2-25K
10-14K
8-9K
6-7K
26-33K
3.5-4K
3-3. 5K
2-2.5K
2.5-3K
16-19.5K
6-7K
24
-------
Table 4-13: "Leftover" Holes Filled by PERE
SourceBinID
1010100012025003500
1010100013035003500
1010100013540003500
1010100013540004000
1010100013540004500
1010100014050004500
1010100015099006000
1010100020020002000
1010100022025004500
1010100023035004500
1010100024050003000
1010100024050005000
1010100024050006000
1010100025099002500
1010100030020002000
1010100032530002500
1010100033540003000
1010100034050006000
1010100035099007000
1010100043035002500
1010100043540002500
1010100043540003500
1010100044050002000
1010100045099003000
1010100995099004500
1010100045099009000
1010100052530004000
1010100053540004000
1010100984050005000
1010100985099004500
Fuel
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
Gas
MY
Group
<80
<80
<80
<80
<80
<80
<80
81-85
81-85
81-85
81-85
81-85
81-85
81-85
86-90
86-90
86-90
86-90
86-90
91-00
91-00
91-00
91-00
91-00
91-00
91-00
01-10
01-10
01-10
01-10
Engine Size Class
(L)
2.0-2.5
3.0-3.5
3.5-4.0
3.5-4.0
3.5-4.0
4.0-5.0
>5.0
<2.0
2.0-2.5
3.0-3.5
4.0-5.0
4.0-5.0
4.0-5.0
>5.0
<2.0
2.5-3.0
3.5-4.0
4.0-5.0
>5.0
3.0-3.5
3.5-4.0
3.5-4.0
4.0-5.0
>5.0
>5.0
>5.0
2.5-3.0
3.5-4.0
4.0-5.0
>5.0
Loaded Weight
Class (Ibs)
3-3.5K
3-3.5K
3-3.5K
3.5-4K
4-4.5K
4-4.5K
5-6K
<2K
4-4.5K
4-4.5K
2.5-3K
4.5-5K
5-6K
2-2.5K
<2K
2-2.5K
2.5-3K
5-6K
6-7K
2-2.5K
2-2.5K
3-3.5K
<2K
2.5-3K
4-4.5K
8-9K
3.5-4K
3.5-4K
4.5-5K
4-4.5K
In addition to the bins in Table 4-12 and 4-13, PERE was used to fill all the bins
for motorcycles, since we did not have any test data to bin directly. In MOVES,
motorcycles are divided into three weight categories (<500 Ibs, 500-700 Ibs and >700
Ibs) and three engine size categories (<170 cc, 170-280 cc, and >280cc). Energy rates
were calculated for each combination of engine size and weight that exists in the fleet (as
captured in the MOVES default database) by running PERE using midpoint values for
each category. Detailed methodology and specifications for motorcycles are described in
the PERE report.
25
-------
4.3.3.2 Hole Filling with Interpolation and Copying
It was not feasible to use PERE to fill all holes, so a systematic approach for
interpolation or copying rates for lower priority holes based on the rates of neighboring
bins was developed. Since many dimensions were available to interpolate, a hierarchical
set of rules were developed to perform interpolation based on weight, engine size or
model year; and barring adequate data to perform these interpolations, to copy directly
from neighboring bins by model year, weight or engine size. This process was executed
using a SAS script. A full discussion of this interpolation decision algorithm is contained
in Appendix D.
4.3.4 Advanced Technologies & Alternative Fuels
Advanced technologies, alternative fuels and future model year vehicles are an
important component of MOVES, in order to generate estimates into the future, and
evaluate the effect of new technologies penetration into the fleet. Unfortunately little to
no data exists advanced technology and alternative fuel vehicles (not to mention future
vehicles) to allow direct binning, or even interpolation; hence a process for generating
these rates was required. Advanced technologies, alternative fuels and future model year
rates are not included in the emission rate table in the MOVES default database; rather, a
separate database table is created by the user via a pre-processing step known as the
Future Emission Rate Creator, or FERC. Details on executing the FERC can be found in
the MOVES2004 User Guide; details on the design can be found in the MOVES2004
Software Design Reference Manual. This section documents the methodology used by
the FERC to generate these rates, focusing on how PERE was used to generate the default
technology ratios employed in the FERC.
A matrix of fuel type and engine technology combinations available for inclusion
in MOVES2004 via the FERC is shown in Table 4-14, shown with checks. Hydrogen-
related technologies (internal combustion, fuel cell and fuel cell hybrid) are not included
in the initial release of MOVES2004 but we plan to add them to the model as soon as
they are available.
26
-------
Table 4-14: Advanced Technologies and Alternative Fuels in MOVES2004
Conventional Internal
Combustion (CIC)
Advanced Internal
Combustion (AIC)
Moderate Hybrid - CIC
Moderate Hybrid - AIC
Full Hybrid -CIC
Full Hybrid -AIC
Electric
Fuel Cell
Hybrid -Fuel Cell
Gas
/
/
/
/
/
/
Diesel
/
/
/
/
/
/
Alt Fuels*
/
Electricity
/
Liquid H2
/
/
Gaseous H2
/
/
/
* Alternative Fuels in MOVES are CNG, LPG, E85 and M85 - these are separate in the model, and are
only combined here for presentation
The rates for advanced technologies and alternative fuels for model years 2001
through 2010 are generated in the FERC by applying ratios to the rates for "base" fuel
and engine technologies for which rates already existed. This is done for all source bins
in the base technology, so that advanced technologies and alternative fuels could be
modeled for all classes of vehicles. For most cases, the technologies used as the base
rates to which ratios were applied were gasoline conventional internal combustion. The
diesel-based technologies (e.g. advanced 1C and hybrids) used diesel conventional
internal combustion for a base. Because the FERC generates rates for all permutations
of source bin, operating mode, fuel type and engine technology, the database table
produced by the FERC is about four times larger than the emission rate table included in
the MOVES default database (nearly 100,000 records in the default configuration).
The source of the ratios used to generate advanced technology rates depend on
technology and fuel type. PERE was used to generate ratios for gas and diesel advanced
internal combustion and hybrid configurations. The PERE ratios varied by operating
mode bin, which allows more flexibility than benefits expressed as aggregate ratios over
standard test procedures that are common in current literature. Ratios for E85, M85,
CNG and LPG were derived from the relative benefits used in GREET, and did not vary
by operating mode bin. For a given fuel/technology combination, one set of ratios (by
operating mode bin, if applicable) are applied to all base source bins, meaning that the
relative benefit of a certain technology is assumed to be the same across all weight
classes and engine sizes. This is a broad assumption that could be refined with more
investigation; we did investigate applying different ratios depending on vehicle weight,
but lacked enough information at this point to determine how the ratios should change by
weight or other vehicle characteristics.
27
-------
Model years beyond 2010 are characterized by two broad model year groups:
2011 thru 2020 and 2021 thru 2050. Rates for these model year groups were generated
by a direct ratio to the 2001 thru 2010 rates for the same technology and fuel type. This
provides some opportunity to model evolutionary improvements in fuel economy and
emissions technology into the future. For the default case, however, a ratio of 1 was
applied across the board for all technologies and fuel types, meaning that the 2011 and
later rates are assumed to have the same performance as 2001 through 2010 rates. While
evolutionary scenarios could certainly be modeled in MOVES through changes to FERC
default inputs, we chose the most conservative future scenario for the default case based
on the trend of minimal evolutionary improvement in fuel economy which has occurred
over the past decade.
The FERC is a MySQL script which employs input files containing the advanced
technology and alternative fuel ratios by fuel and engine technology and the base
technologies to which the ratios should be applied. The default 2001 through 2010 model
year ratios shown in Table 4-15 are contained in the FERC input file named
"ShortTermFERC_PERE" included with the MOVES2004 installation package. The
2011 and later model year ratios (all equal to 1) are contained in the FERC input file
named "LongTermFERC" included with the MOVES2004 installation package. Users
wishing to use alternate advanced technology assumptions would replace the default
ratios in these files prior to executing the FERC. This is designed as an automated
process for customized "what-if' sensitivity analysis.
4.3.4.1 Modeling Advanced Technology Vehicles Using PERE
There are actually a number of ways that PERE could be used to generate
advanced technology rates in MOVES. It can be run many times for each source bin
to generate rates directly, but this would amount to hundreds of runs. Alternatively,
a select few of the engine size and weight categories could be run and the rest
interpolated. The approach used for MOVES2004, however, was to model a single
representative vehicle, bin the results by operating mode, and then ratio these rates to
the corresponding "base" conventional vehicle rates across all source bins. This
approach is simpler and less time consuming, but requires broad assumptions and is
considered a first step. The process for generating these ratios is described in the
following steps:
Step 1: Choose the Representative Source Bin
Due to its relative frequency in the fleet, the "representative" vehicle was
chosen to lie in a MOVES source bin with weight ranging from 3,500 - 4,000 Ibs
(test weight), and engine size ranging 3.0-3.5 Liters (automatic transmission).
Step 2: Define the Vehicle Specifications
Based on the engine technology source bin, the vehicle parameters such as
weight, engine size, vehicle shape etc, can be defined. The weight and engine size
28
-------
values are simply the central value of the bin. Thus PERE effectively models an
"average" vehicle in the source bin. The corresponding values for the representative
source bin described under Step 1 to a PERE conventional vehicle of 3,750 Ibs,
3.25L, and 155kW (peak power).
Since engine size has limited meaning for hybrids and no meaning for
electric vehicles, it is necessary to define a power surrogate for engine size.12
Though it is not perfect, we take the convention that motor power + engine peak
power = total peak power. Taking this into account, the PERE specifications for the
representative source bin are and engine size of 1.7L for a full hybrid with a 72kW
motor, and 2.8L for a moderate hybrid with a 22kW motor (gasoline and diesel).
Because the source bins do not have a dimension for body type, the road load
coefficients are estimated based on the weight. Lighter light duty weights tend to be
(compact) passenger cars, then midsize cars, luxury, compact pickups, SUVs,
minivans on up to medium-duty trucks. The variation in body types will certainly
lead to variation (or uncertainty) in the emission rates.
Step 3: Define the Driving Cycles
The output to PERE is second-by-second energy consumption, therefore the
next step is to input the driving cycles. The driving cycles input into PERE help
determine the "binned" energy consumption rates for MOVES. The binned energy
consumption and emissions rates are dependent on the driving cycle input. It is
important to capture a representative sampling of real-world driving. The number of
cycles are reduced for advanced technologies compared to conventional vehicle
cycles (described in Section 4.3.3), to speed up the model runs.
Three driving cycles were used for the advanced technology runs: FTP
(urban), FTP (highway), and LA92. The latter is a self-weighting cycle and includes
harder accelerations and higher speeds. For hybrids, the state-of-charge was
maintained over the total of the three cycles.
Step 4: Run PERE and Bin Output
After PERE was run, the second-by-second results were binned into the 17
operating modes described in Section 4.1.3. Figure 4-3 shows a sample of predicted
fuel consumption as a function of VSP for a fictional hybrid passenger car. This is an
indication of how an output from PERE would be translated into an emissions rate in
MOVES. The uncertainty bars are from PERE generated variations within a bin, and
do not adequately reflect the true uncertainty of the emission rates, which will be
added in future versions of PERE.
29
-------
Figure 4-3: Hybrid Fuel Consumption by Operating Mode Bin from PERE (g/s)
CO
-z.
O
o
LJJ
o
LO
CD
2.0
1.5
1.0
.5
0.0
r i
N= 258 357 111 171 121 85 48 15 151 283 281 144 62 46 173 301 34
.00 11.00 13.00 15.00 21.00 23.00 25.00 33.00 36.00
1.00 12.00 14.00 16.00 22.00 24.00 26.00 35.00
VSPBIN3
The binned results for each advanced technology model by PERE were
divided by the binned results for the baseline conventional technology runs to
generate the ratios. As PERE produces fuel consumption rates which reflect
differences in energy content, PERE results for non-gasoline fuels were corrected to
gasoline equivalent so that the ratios reflect differences in energy consumption, not
fuel consumption.
The ratios generated by PERE for all fuel types and engine technologies included
in MOVES are shown in Table 4-15. These ratios are the same as included in the
"ShortTermFERC_PERE" file, except for diesel technology ratios, which are shown
relative to gasoline conventional internal combustion here but are expressed relative to
diesel conventional internal combustion in this file.
30
-------
Table 4-15: Advanced Technology Running Energy Consumption Ratios (relative to Gasoline Conventional 1C)
Fuel/Technology
Gasoline Conventional 1C (CIC)
Gasoline Advanced 1C (AIC)
Gasoline CIC Hybrid Mild
Gasoline CIC Hybrid Full
Gasoline AIC Hybrid Mild
Gasoline AIC Hybrid Full
Diesel Fuel Conventional 1C
Diesel Fuel Advanced 1C
Diesel CIC Hybrid Mild
Diesel CIC Hybrid Full
Diesel AIC Hybrid Mild
Diesel AIC Hybrid Full
Electric
Operating Mode Bin
0
i
0.88
0.00
0.00
0.00
0.00
0.52
0.48
0.00
0.00
0.00
0.00
-0.54
1
1
0.86
0.00
0.00
0.00
0.00
0.47
0.41
0.00
0.00
0.00
0.00
0.08
11
1
0.88
0.00
0.00
0.00
0.00
0.54
0.50
0.00
0.00
0.00
0.00
-0.04
12
1
0.89
0.33
0.18
0.29
0.15
0.66
0.62
0.15
0.10
0.14
0.10
0.17
13
1
0.90
1.02
0.87
0.92
0.78
0.72
0.69
0.57
0.52
0.54
0.50
0.31
14
1
0.90
1.08
0.97
0.97
0.87
0.75
0.71
0.78
0.73
0.75
0.69
0.36
15
1
0.90
1.13
1.06
1.00
0.95
0.77
0.74
0.84
0.81
0.80
0.77
0.39
16
1
0.90
1.16
1.22
1.03
1.14
0.79
0.76
0.90
0.96
0.86
0.92
0.44
21
1
0.88
0.00
0.00
0.00
0.00
0.51
0.47
0.00
0.00
0.00
0.00
-0.08
22
1
0.89
0.36
0.11
0.32
0.07
0.64
0.61
0.45
0.19
0.43
0.18
0.20
23
1
0.90
0.98
0.85
0.88
0.76
0.71
0.67
0.70
0.63
0.67
0.61
0.32
24
1
0.90
1.05
0.95
0.95
0.85
0.73
0.70
0.79
0.74
0.75
0.71
0.40
25
1
0.90
1.13
1.10
1.02
0.99
0.77
0.74
0.87
0.86
0.83
0.83
0.47
26
1
0.90
1.18
1.27
1.07
1.21
0.81
0.77
0.94
1.02
0.90
0.98
0.50
33
1
0.89
0.82
0.63
0.70
0.55
0.63
0.60
0.54
0.44
0.52
0.42
0.23
35
1
0.90
1.05
0.93
0.91
0.82
0.70
0.67
0.72
0.67
0.69
0.64
0.36
36
1
0.90
1.14
1.09
1.01
1.01
0.74
0.71
0.84
0.83
0.81
0.80
0.47
31
-------
Figures 4-4 and 4-5 show the ratios for gas hybrid and diesel hybrid
technologies. Note that it is possible for the ratios to exceed 1 in some operating
mode bins. This is due to the fact that the hybrids are heavier vehicles, driven with
smaller engines, yet still follow the same driving trace. Thus, it is quite possible for
the fuel consumption to be higher than the conventional vehicle during certain
modes of driving. However, over an entire driving cycle, the fuel consumption is
significantly lower. In most driving cycles, the time spent in modes where the ratio is
greater than 1 is usually quite small (high acceleration events). These results support
the well-known evidence that the advantage of hybrids is seen in stop-and-go
driving.
Modeling hybrid performance by operating mode greatly increases the power
of the model to assess real-world performance, compared to the use of a single fixed
fuel economy ratio (over an FTP cycle for example) as other studies have proposed.
This is due to the fact that hybrid fuel economy depends in large part on the type of
driving, idle time, decelerations etc., which are washed out in a single cycle average
number.
Figure 4-4. The ratio of energy (or fuel consumption) for gasoline moderate and
full hybrid vehicles, relative to gasoline conventional 1C
Gasoline Hybrid
Ru
ng
En
y
Ra
to sol
Coine
ent
ion
n
•Conv Mod
DConv Full
>1 ratio due to
rpm/smaller engine
combinatio
-
n
I
1
h
r
1
•
~
n
h
-
n
•
i
-.
11 12 13 14 15 16 21 22 23 24 25
Operating Mode bin
26 33
35
32
-------
Figure 4-5. The ratio of energy (or fuel consumption) for diesel moderate and
full hybrid vehicles, relative to gasoline conventional 1C
Diesel Hybrid
Ru
nni 1
ng
En
RaGa
I0sol
nJne
CO
nv
ent
ion 0-4-
al
c
c
D Diesel
• Diesel Mod
nbieseLEull_
esel ratio is set
onstant
(gasoline
i-i
-
_.
]l
rh
(
]
r
-i
-i
'
0 1 11 12 13 14 15 16 21 22 23 24 25 26 33 35 36
Operating Mode bin
4.3.4.2 Investigation of Weight-Eased Ratios
The development of ratios based on a single source bin has limitations, which are
presently explored for the purpose of considering next steps for modeling advanced
technology vehicles in the evolution of MOVES. The ratios are based on a single (though
common) representative vehicle class. We might expect vehicles of different weight but
same power-to-weight ratio to give similar rates. However, vehicles with lower power-to-
weight ratios would (for example) require more assistance from the engine, both during
launch and assist (power boost). This would necessarily change the shape of the curves.
An example of this is shown in Figure 4-7. The varying power-to-weight ratios having
different shapes. The effect is mainly pronounced in the "2" bins, i.e. bins 12, and 22.
These correspond to moderate acceleration from low to medium speed (launch).
33
-------
Figure 4-7. Fuel consumption ratios for a series of power-to-weight ratio (kW/kg)
light-duty gasoline hybrid vehicles.
LOG Moderate Hybrids
1 9
o
** A
||08
•S 206
o gu.o
0 0
'f UQ4
no
OH
(
B i !
s
i
X
rs
i
* i
i j{S
! -
*
1
i |
i -
i
+
H
S
•
) 5 10 15 20 25 30 35
vsp bin
» 0.043
• 0.05
0.061
0.072
*0.08
n 0.091
+ 0.1
-0.14
-0.21
The shape difference is even more pronounced in medium-duty vehicles
where the power-to-weight ratio can fall below 0.01. This is shown in Figure 4-8 for
a series of medium-duty diesel hybrid vehicles. Note that the fuel consumption drops
at higher loads for the underpowered vehicles. This is due to the fact that the engine
cannot keep up with the cycle by itself and more of the total power is supplied by the
battery/motor. It is also likely that these vehicles are unable to follow the driving
cycles. These hybrids were run on the following medium-duty driving cycles: CBD
(Central Business District), NYBUS, NYCCT, and UDDS-D. Although weight-
based ratios were not used in MOVES2004, these results merit further consideration
of how to improve the resolution of advanced technology modeling in MOVES.
Figure 4-8. Fuel consumption ratios for a series of power-to-weight ratio (kW/kg)
medium duty diesel hybrid vehicles.
Moderate MOD Hybrids
15
0 1
'c
? n ft
o
o
° 06
2
-*1 n4
"o
2
0 ^
i
!• ,
r1
* i
i
i *
h"
i
1 9
«
l
"
» 1
|
0
1 i
*
0 5 10 15 20 25 30 35
vsp bin
• 0.0113
» 0.0214
• 0.0302
A 0.0411
0.0494
x 0.0757
o 0.0677
34
-------
For future versions of MOVES, the rates may likely be separated by power-
to-weight (P/Wt) ratio bins, rather than engine displacement or weight separately in
order to differentiate the energy ratios for the different vehicle classes. An example
of such a split is shown in Table 4-16. The bands of color represent possible P/Wt
ratios to group together into bins. The boxed cells are modeled using PERE.
Table 4-16. Possible Power-to-weight ratio bins (kW/kg) for turbo diesel hybrids.
lo wt bin
Ibs
2000
2500
3000
3500
4000
4500
5000
6000
7000
8000
9000
10000
14000
16000
19500
26000
avg wt in bin
kg
1021
1247
1474
1701
1928
2155
2495
2948
3402
3856
4309
5443
6804
8051
10319
13381
Engine displacement
2 2.5
95.2 116.3
0.0933
0.0763
0.0646
0.0560
0.0494
0.0442
0.0382
0.0323
0.0280
0.0247
0.0221
0.0175
0.0140
0.0118
0.0092
0.0071
0.1140
0.0932
0.0789
0.0684
0.0603
I 0.0540
0.0466
0.0394
0.0342
0.0302
0.0270
0.0214
0.0171
0.0144
0.0113
0.0087
or avg pwr
3
137.5
0.1347
0.1102
0.0933
0.0808
0.0713
0.0638
0.0551
0.0466
0.0404
] 0.0357 [
0.0319
] 0.0253
0.0202
0.0171
] 0.0133
0.0103
3.5
158.6
0.1554
0.1271
0.1076
0.0932
0.0823
0.0736
0.0636
0.0538
0.0466
0.0411 |
0.0368
0.0291
0.0233
0.0197
0.0154
0.0119
4
190.3
0.1865
0.1526
0.1291
0.1119
0.0987
0.0883
0.0763
0.0645
0.0559
0.0494
0.0442
0.0350
0.0280
0.0236
0.0184
0.0142
5
291.9
0.2860
0.2340
0.1980
0.1716
0.1514
0.1355
0.1170
0.0990
0.0858
0.0757
0.0677
0.0536
0.0429
0.0363
0.0283
0.0218
4.3.4.3 Modeling Alternative Fuels Using GREET
Energy ratios for CNG, LPG, E85 and M85 were taken directly from GREET,
which in turn are based on analysis of available certification data on vehicles using these
fuels.13 GREET uses aggregate ratios of energy consumption, so no operating mode
split was available. However, for these fuels, the relative benefit of the technology would
be expected to be more uniform across operating mode, and the need to split by operating
mode is less important than for advanced technologies, particularly hybrids. The ratios
for these fuels derived from GREET are shown in Table 4-17, and are in the
"ShortTermFERC_PERE" file used by the FERC to generate future rates.
Table 4-17: Alternative Fuel Energy Consumption Ratios
(relative to Gasoline Conventional 1C)
Fuel
CNG
LPG
E85
M85
Ratio
1.05
1.0
0.95
0.95
35
-------
4.4 Start Energy Rate Development
The definition of start energy consumption for MOVES follows the approach
initially developed and documented with MOBILE6. With this approach, "start" energy
is defined as the energy consumed at startup over and above the energy which would be
consumed had the vehicle followed the same trajectory during running (warmed-up)
operation. Start energy rates are therefore the incremental amount of energy consumed at
start-up, and start rates in the model are in the units of KJ per start. Starts were not
separated into operating mode in MOVES2004. The main ramification of this is that
MOVES2004 does not differentiate between hot start and cold start energy consumption
(e.g soak time). We expect to expand the approach used for start emissions to include
soak time and load-based effects when the model is developed for criteria pollutants.
4.4.2 Data Sources
Data used in the development of start rates analysis came from EPA's Mobile
Source Observation Database (MSOD) as of April 2003. The initial analyses were
limited to the FTP tests that were performed within the temperature range of 68 degrees
to 86 degrees Fahrenheit (i.e., at a nominal temperature of 75° F). This restriction
produced a database of 18,676 FTPs performed on 10,422 vehicles. Only FTP tests were
used, since the basis of start emission rates is the difference between Bag 1 cold start and
Bag 3 hot start. Whereas running energy rates are based entirely on second-by-second
data, start energy rates are based entirely on aggregate FTP bag results.
4.4.3 Methodology for Rate Development
4.4.3.1 Assessing Hot Start vs. Hot Running
The approach used to estimate the amount of fuel consumed during an engine
start requires identifying vehicles that were tested over driving cycles that differed only
by the presence of an engine start. What we found were 192 vehicles (in the MSOD) that
were tested over a Federal Test Procedure (FTP) driving cycle followed by an immediate
"Hot-Running 505" (HR-505).
The FTP driving cycle consists of three individual operating modes:
• The first mode (Bag-1) is a 505-second driving cycle that begins with a cold start
(i.e., following a soak of 12 to 36 hours). This mode is referred to as a "Cold-
Start 505" (CS-505).
• The second mode (Bag-2) is an 867-second driving cycle that involves no engine
start.
• The third mode (Bag-3) is a 505-second driving cycle identical to the first mode
but begins with a hot start (i.e., following a soak of only 10 minutes). This mode
is referred to as a "Hot-Start 505" (HS-505).
36
-------
Following the completion of the FTP (but without shutting off the engine), 192
vehicles then drove another 505-second driving cycle identical to both the first and third
modes. Since this cycle involved NO engine start, it is referred to as a "Hot-Running
505" (HR-505).
Subtracting the fuel consumed over the HR-505 from the fuel consumed over the
corresponding CS-505 should provide an estimate of the fuel consumed during an engine
cold-start itself. However, this data set is relatively small considering the analyses we
wish to perform. Hence, our approach will be to first develop a substantially larger data
set for estimating fuel consumed during engine start. Specifically, we will consider
whether we can use the fuel consumed over the HS-505 as a surrogate for the fuel
consumed over the HR-505.
To compare the fuel consumed over the "Hot-Start 505" with the fuel consumed over
the corresponding "Hot-Running 505" cycle, we used 244 test pairs (FTP and "Hot-
Running 505") on those 192 vehicles. In Figure 4-9, we plotted the fuel consumed over
the HR-505 versus the fuel consumed over the corresponding HS-505 cycle. Even the
most cursory visual inspection of that graph reveals a very strong match between these
two quantities. A more rigorous approach using a linear regression produces the equation
for the fuel consumed (in gallons), with an R-squared value of 0.99:
Eqn 4-1 HR505 = (1.0095 *HS505) - 0.002
The slope in the preceding equation has a 95 percent confidence interval of 0.999 to
1.019 (with a P-value less than 0.00001), and the intercept has a 95 percent confidence
interval of -0.0041 to 0.00015 (with a P-value of 0.069). This strong linear correlation
and the virtual equality implied by this regression equation suggests that fuel consumed
during the HS-505 is a reasonable estimate of the fuel used during the HR-505.
(Additionally, in 107 of those 244 test pairs, the fuel consumed during the HR-505
exceeded the amount consumed over the HS-505. In the remaining 137 pairs the reverse
was true. This nearly equal split also argues for the HS-505 being a reasonable estimate
of the HR-505.)
Therefore, for the remainder of this analysis, we used the difference of the fuel
consumption on Bag-1 of the FTP minus the fuel consumption on Bag-3 of the FTP as an
estimate of fuel consumed during a cold engine start.
37
-------
Figure 4-9: Comparison of Fuel Consumed Over HS-505 and HR-505
0.50
0.40
0.30
0.20
0.10
0.00
0.00 0.10 0.20 0.30 0.40 0.50
Fuel Consumed (gallons) over HS-505 (Bag_3)
4.4.3.2 Determining Source Bin Variables
From the MSOD, we identified 20,156 FTPs that were performed on 10,516
vehicles. The vehicle and test result data from those tests were analyzed using a stepwise
regression process in order to determine the most important variables to consider for
source bins.
The stepwise regression process first uses the Pearson Product-Moment to select
the independent variable that has the highest correlation with the "Fuel Consumed During
Start" (estimated by subtracting the Bag-3 fuel consumption from the corresponding Bag-
1 fuel consumption). The difference between the best linear estimate using that variable
and the "Fuel Consumed During Start" term (i.e., the residuals) is then compared with the
set of remaining variables to identify the variable having the next highest correlation.
This process continues as long as the "prob" values do not exceed (an arbitrary) five
percent, thus, creating a sequence of variables in descending order of statistical
correlation. The rank ordering produced by this process is dependent upon the
independence of the variables. In this instance, there is some collinearity among the
variables, which may reduce the usefulness of this statistical tool.
38
-------
For each of those 20,156 FTPs, the difference of the fuel consumed on Bag-1
minus the fuel consumed on Bag-3 was calculated as an estimate of fuel consumed during
an engine start. Those values (in gallons) were used as the dependent variable along with
the following 12 variables that were considered as independent variables in this stepwise
analysis:
Age (estimated as "Model Year" minus test year)
Ambient Temperature
Engine Displacement (in cubic inches)
Number of Cylinders
Model Year (four-digit year ranging from 1965 through 2000)
Odometer
Test Weight
Car v. Truck
Fuel Delivery System (Carbureted v. Fuel Injected)
Fuel Injection System (TBI v. PFI)
Gasoline v. Diesel
Light-Duty (cars & trucks) v. Heavy-Duty
The last five variables on the preceding list are categorical variables (i.e., ones and zeros).
Some of these variables are potentially collinear, and that collinearity could affect
the usefulness of the regression process. Thus, we were cautious about using (in the
same regression) those pair of potentially collinear variables. Those pairs are:
Engine Displacement / Number of Cylinders,
Model Year / Fuel Delivery System, and
Odometer / Age.
The stepwise regression process continued for eight steps, the last of which is shown in
Table 4-18.
39
-------
Table 4-18: Stepwise Regression of
Dependent variable is:
No Selector
20156 total cases of which
3960 are missing
R squared = 38.5% R squared (adjusted) =
s= 0.0144 with 16197-9
Source Sum
Regression
Residual
Variable
Constant
Disp (CID)
Avg_Temp
Model_Year
Carb / FI
LD/HD
Estimated_Age
No._of_ Cylinders
Odometer (miles)
= 16188 degrees
of Squares
2.08917
3.3355
Coefficient
1.40283
0.000072
-0.000710
-0.000674
-0.003503
0.007635
0.000312
0.001212
0.000000
Start (Bag 1 - Bag 3) Fuel Consumption
38.5%
of freedom
df
8
16188
s.e. of Coeff
0.0786
0.0000
0.0000
0.0000
0.0003
0.0011
0.0000
0.0002
0.0000
Start
Mean Square
0.261146
0.000206
t-ratio
17.9
17.7
-64.1
-17.0
-11.1
7.2
4.45
6.11
-3.12
(Bag-1 minus Bag-3)
F-ratio
1267
prob
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.0018
From this analysis we concluded that engine displacement and model year were
important variables to bin by. The importance of average temperature also indicated the
need for a separate temperature effect, discussed in Section 9. The coefficients values
also suggest that the difference in actual fuel consumption during engine starts may be
substantial between light-duty and heavy-duty vehicles.
4.4.4 Emission Rate Development
With no operating mode bins and only aggregate bag data to work with, the
development of emission rates was straightforward. Energy rates, in terms of KJ per
start, were calculated for each combination of engine displacement and model year group
from Table 4-1 by averaging the start (Bag 1 minus Bag 3) results for all FTP tests from
vehicles falling in those bins. Vehicle weight did not appear as an important variable in
the stepwise regression analysis, so test data was not segregated by weight for this
analysis. However, since total energy source bin definitions include the vehicle weight
class, the start energy rates were set at the same level for all weight classes within an
engine displacement / model year group bin.
The start emission rates by source bin category which resulted from this analysis
are presented in Appendix E.
40
-------
4.4.5 Hole Filling
Despite the large number of tests, a number of source bins lacked adequate data to
fill directly - mostly for heavy-duty and/or diesel vehicles. We filled these holes mostly
by copying from other bins where data existed, or applying ratios to existing rates based
on trends observed in the data. This section provides details for this process. The
calculations as described here were performed in fuel consumption space first, then
converted to energy consumption for use in MOVES. The rates developed from this
process are presented in Appendix E.
In comparing the differences between the fuel consumed during engine starts for
the light-duty gasoline cars and trucks (LDGs) and for the heavy-duty gasoline-fueled
vehicles (HDGVs), we noted:
• For the HDGVs, there were data only in the 1986-90 and 1991+ model year bins
and in only the 4.1-5.0 and 5.1+ displacement bins.
• Within each of the two displacement bins, the effect of changing model year
ranges was not statistically significant.
• After combining all model year data, the resulting (two) displacement bins were
not statistically different from the corresponding displacement bins for the Pre-
1981 light-duty trucks & cars. (Probably due to lack of CAFE standards for the
HDGVs.)
Therefore, we used the results (i.e., means) from the Pre-1981 light-duty gasoline-fueled
cars and trucks for the HDGVs.
In comparing the differences between the fuel consumed during engine starts for
the heavy-duty gasoline-fueled vehicles (HDGVs) and for the heavy-duty diesel-fueled
vehicles (HDDVs), we noted:
• For the HDDVs, there were data only in the 1986-90 and 1991+ model year bins
and in only the 5.1+ displacement bin.
• Although the difference in average engine-start fuel consumption was statistically
significant for those two model year bins, the fact that the 1986-90 bin contained
data on only two vehicles makes it difficult to justify splitting by model year.
• Combining results on the 2 tests from the 1986-90 model year range with the 29
tests from the 1991 and newer range, produces a mean of 0.0277 gallons per start
with a standard deviation of 0.0247.
• Comparing the mean and standard deviation with those from the HDGVs (over
5.0 liter displacement), we note the standard deviations are close and that the fuel
consumption (in gallons) was reduced by 46 percent.
41
-------
Therefore, we calculated the average start fuel consumption of the HDDVs as the number
of gallons of gasoline used by the corresponding HDGVs reduced by 46 percent.
In comparing the differences between the fuel consumed during engine starts for
the light-duty gasoline-fueled vehicles (LDGVs) and for the light-duty diesel-fueled
vehicles (LDDVs), we noted:
• Except for the 2.6-3.0 liter displacement bin, the amount of diesel fuel consumed
(in gallons) for each engine start is only 60 to 80 percent of the number of gallons
of gasoline consumed for each corresponding engine start.
• Although there are no results for the 1986-1990 model year diesel-fueled vehicles,
we note that, for the corresponding gasoline-fueled vehicles, the amount of
gasoline consumed for an engine start is about 3 percent higher than for the
corresponding 1991 and newer vehicles. Applying this factor to the diesel fuel
consumed (by that single test) of the 1991+ vehicle permits us to obtain an
estimate for the 1986-90 diesel light-duty cars and trucks.
Therefore, we propose to estimate the average engine-start fuel consumption of the
LDDVs, by:
• Multiplying by 70 percent (the mid-point from the first observation) the number
of gallons of gasoline consumed by the light-duty vehicles to estimate the number
of gallons of diesel fuel consumed (for an engine start) for all but the 2.6-3.0 liter
displacement bin.
• Using the calculated mean start fuel consumption for the LDDVs with 2.6-3.0
liter engines for all model year ranges except the 1986-1990.
• For the 1986-90 model year diesels with 2.6-3.0 liter engines, use the fuel
consumption of the corresponding 1991+ vehicles increased by 2.91 percent
(where 2.91 percent was the increased start fuel consumption of the 1986-1990
LDDVs compared to their 1991+ counterparts).
The final category of holes to fill were motorcycles. In MOVES, motorcycles are
stratified into two groups, "uncontrolled" and "controlled". The uncontrolled
motorcycles have engines similar to but usually smaller than those used in the pre-1981
model year passenger cars and light-duty trucks. The controlled motorcycles have
engines similar to but usually substantially smaller than those used in the 1991-2000
model year passenger cars and light-duty trucks. Performing linear regressions on the
fuel consumption estimates for those two model year ranges produce the following
equations for fuel consumed during an engine start:
Eqn 4-2 Uncontrolled Motorcycle Energy Consumption (KJ)
= 2048.1 + (717.16 * Displacement)
42
-------
Eqn 4-3 Controlled Motorcycle Energy Consumption (KJ)
= 1402.15+ (477.97* Displacement)
Using these regression equations, we are able to estimate the fuel consumed (as functions
of engine displacement) during an engine start of each type of motorcycle.
4.4.6 Advanced Technologies and Alternative Fuels
Advanced technology and alternative fuel start rates were developed in the
same manner as for running energy rates; by applying ratios to base conventional
technology via the Future Emission Rate Creator (FERC). There is a dearth of data
on cold start factors for advanced technology vehicles, so some assumptions were
required to develop these ratios. A comparison of cold start energy consumption
was conducted on two hybrid vehicles (details available in the PERE report), from
which it was decided that hybrid cold start factors may be consistent with those for
conventional vehicles. For gasoline and diesel-fueled vehicles, we therefore based
the ratios solely on the changes in engine displacement assumed in the development
of the running energy ratios. Advanced internal combustion engines were therefore
given no additional benefit, but hybrid vehicles are assigned benefit based on the
expectation that engine size will be reduced.
Cold start factors for hybrids were approximated by developing a linear
function of start fuel consumption as a function of engine displacement on the
conventional vehicles, plugging in the smaller engine displacement assumed for
hybrids, and calculating the ratio of conventional start fuel energy to hybrid start
energy for use in the FERC. Based on an analysis of start fuel consumption rates
versus engine displacement from the data detailed in Section 4.4.1, the following
regression equations were used for gasoline and diesel:
Eqn 4-4 Gasoline Consumed (gallons) = 0.0114 + (0.0039 * Displacement)
Eqn 4-5 Diesel Consumed (gallons) = 0.0072 + (0.0028 * Displacement)
Alternative fuel ratios were derived from GREET, and are the same as the running
ratios presented in Section 4.3. The ratios used for start are shown in Table 4-19. These
ratios are the same as included in the "ShortTermFERC_PERE" file, except for diesel
technology ratios, which are shown relative to gasoline conventional internal combustion
here but are expressed relative to diesel conventional internal combustion in the FERC
file.
43
-------
Table 4-19: Advanced Technology & Alternative Fuel
Start Energy Consumption Ratios (relative to Gasoline Conventional 1C)
Fuel / Technology
Gasoline Advanced 1C
Gasoline CIC Hybrid Moderate
Gasoline CIC Hybrid Full
Gasoline AIC Hybrid Moderate
Gasoline AIC Hybrid Full
Diesel Fuel Conventional 1C
Diesel Fuel Advanced 1C
Diesel CIC Hybrid Moderate
Diesel CIC Hybrid Full
Diesel AIC Hybrid Moderate
Diesel AIC Hybrid Full
Compressed Natural Gas (CNG) CIC
Liquid Propane Gas (LPG) CIC
Ethanol (E85) Conventional 1C
Methanol (M85) Conventional 1C
Electric
Start Energy Ratio
1.0
0.93
0.75
0.83
0.67
0.77
0.73
0.71
0.57
0.67
0.54
1.05
1.0
0.95
0.95
1.0
4.5 Extended Idle
The extended idle process was added to MOVES primarily to account for
"hoteling" of long-haul heavy-duty trucks, whose activity isn't explicitly accounted for in
VMT estimates or in the driving schedules used to characterize on-road operation.
Extended idle for long-haul trucks generally occurs at truck stops, to allow drivers to heat
or air condition the truck cab overnight. Energy and emission rates tend to be higher for
extended idle relative to on-road idle operation, since RPM is usually set higher (e.g.
1200 RPM versus 600 RPM for on-road idle) to provide adequate power for the auxiliary
systems.
Limited emission data exists for extended idle conditions, so direct binning of test
data was not an option across the range of source bins in MOVES. However, EPA has
conducted a test program to assess the relative change in energy and some emissions on
short periods of idle vs. longer periods.14 To develop energy rates for extended idle, this
work was used to generate an adjustment factor to be applied to the "idle" bin data for
running total energy (Bin 1). A series of tests conducted in May 2002 on 5 trucks at
paired varying idle speeds, with and without air conditioning. The most common low-
high idle speed pair was 600 RPM (simulating on-road idle) and 1200 RPM (simulating
extended idle), so from the original sample, we analyzed the difference in fuel
consumption between all points at 600 RPM and 1200 RPM. The results are shown in
Table 4-20.
44
-------
Table 4-20: Fuel Consumption Rates At Varying Idle Speeds
RPM
600
1200
600
1200
Relative Increase
Relative Increase
A/C Status
Off
Off
On
On
Off
On
Fuel Rate (gallons/hour)
0.503
1.112
0.565
1.373
2.22
2.43
For the A/C off case, fuel consumption increased by a factor of 2.22 between 600 and
1200 RPM.
The default case for MOVES2004 only includes extended idle for combination
long-haul trucks, so that emission rates are only needed for those source bins which are
mapped to this category. However, to provide the possibility for modeling extended idle
for all source types, we generated extended idle rates for all total energy source bins (this
would allow extended periods of idling in fast-food drive-thrus, school zones, etc. to be
accounted for if desired). To generate extended idle rates for all source bins, we applied
the 2.22 multiplier from Table 4-20 to the Bin 1 on-road idle rates within a given source
bin, for all loaded weight categories greater than or equal to 33,000 pounds (the de facto
"heavy duty" cutpoint). Since we did not have data for the lighter vehicle weight range,
we set extended idle rates equal to the Bin 1 on-road idle rates for loaded weights below
33,000 Ibs.
Advanced technology and alternative fuel rates for extended idle were generated
using the same adjustment process described for running and start in Sections 4.2 and 4.3.
The adjustment ratios used for extended idle were estimated based on the adjustments for
the idle bin under running energy consumption, although some modifications were made
to account for the fact the length of idle period would preclude the battery as a power
source for hybrid vehicles. The ratios used for extended idle are presented in Table 4-21,
and are included in the FERC input table "ShortTermFERC_PERE".
45
-------
Table 4-21: Advanced Technology & Alternative Fuel
Extended Idle Energy Consumption Ratios
(relative to Gasoline Conventional 1C)
Fuel/Technology
Gasoline Conventional 1C (CIC)
Gasoline Advanced 1C (AIC)
Gasoline CIC Hybrid Moderate
Gasoline CIC Hybrid Full
Gasoline AIC Hybrid Moderate
Gasoline AIC Hybrid Full
Diesel Fuel Conventional 1C
Diesel Fuel Advanced 1C
Diesel CIC Hybrid Moderate
Diesel CIC Hybrid Full
Diesel AIC Hybrid Moderate
Diesel AIC Hybrid Full
Electric
Running Idle
1
0.86
0.00
0.00
0.00
0.00
0.47
0.41
0.00
0.00
0.00
0.00
0.08
Extended Idle
1
0.86
1.00
1.00
1.00
1.00
0.47
0.41
1.00
1.00
1.00
1.00
0.08
Comments
assumed to run off engine
assumed to run off engine
assumed to run off engine
assumed to run off engine
5. Petroleum and Fossil Energy Calculations
In MOVES2004, petroleum and fossil energy are the quantities of energy
consumption derived from petroleum or fossil-based sources. Petroleum energy is a
subset of fossil energy, the latter including all of fuel subtypes CNG, LPG and M85 as
well as gasoline and diesel. Well-to-pump petroleum and fossil energy rates are
generated directly by GREET, and passed into the GreetWellToPump table. Estimates
of petroleum and fossil-based energy are calculated for the pump-to-wheel processes
(running , start, extended idle) in MOVES by multiplying total energy consumption
results by the fraction of energy which is either petroleum or fossil-based. With this
approach, direct rates of petroleum and energy consumption are not required in the
emission rate database.
The petroleum and fossil energy fractions for the pump-to-wheel processes were
derived from GREET values for on-road energy consumption. Specifically, Table 2 of
the "Results" sheet within GREET (spreadsheet version) contains gram/mile total,
petroleum and fossil energy results for the range of fuels and vehicles in GREET. To
generate the petroleum and fossil fractions for MOVES, we divided the petroleum and
fossil energy gram/miles results attributed to "vehicle operation" by the corresponding
total energy value. The results for each MOVES fuel subtype are shown in Table 5-1,
including the corresponding GREET fuel / vehicle category used to derive the results.
46
-------
Table 5-1; Pump-To-Wheel Petroleum and Fossil Energy Fractions by Fuel Subtype
Fuel Subtype
Conventional Gasoline
Reformulated Gasoline
E10
Conventional Diesel
Biodiesel (BD20)
Fisher-Tropsch(FTlOO)
CNG
LPG
Ethanol (E85)
Methanol (M85)
Gaseous Hydrogen
Liquid Hydrogen
Electricity
GREET Fuel / Vehicle
Category
Baseline Gasoline
Baseline Gasoline
Low-level EtOH Blend
Conventional Diesel
BD20
FT100
Dedicated CNGV
LPGV: Dedicated
EtOH: FFVE81 Corn
MeOH FFV: M85 nNA NG
FCV: GH2
FCV: LH2
Electric
Petroleum Fraction
0.95
0.95
0.94
1.0
0.81
0
0
0.4
0.26
0.26
0.01
0.01
0.02
Fossil Fraction
1.0
1.0
0.94
1.0
0.81
1.0
1.0
1.0
0.26
1
0.94
0.94
0.87
6. Carbon Dioxide (CO2) Calculations
MOVES2004 does not currently estimate CC>2 emissions, but is planned for future
releases using the methodology presented here. CC>2 will be calculated from total energy
consumption results, rather than through the direct use of CC>2 emission rates. CC>2 will
be calculated in this way according to equation 6-1:
Eqn 6-1 CO2 = Total Energy Consumed * Carbon Content * Oxidation Fraction * (44/12)
This equation methodology is consistent with methods to calculated CC>2
inventories used in the U.S. Inventory of U.S. Greenhouse Gases Emissions and Sinks
(hereafter referred to as Emissions & Sinks), in line with International Panel on Climate
Change (IPCC) guidelines.15 In this equation, carbon content is in terms of energy,
expressed in grams per KJ, and the oxidation fraction is the percent of carbon which
winds up as CC>2 in the atmosphere - the portion which remains unoxidized is generally
black carbon (particulate matter) emissions.
The Fuel Sub Type table in the MOVES default database contains the carbon
content and oxidation fraction values used in equation 6-1, by fuel subtype. Carbon
content values were derived from GREET estimates of carbon weight fraction and Lower
Heating Values (LHVs) from Heywood.16 The Heywood LHVs were used for
consistency, since they were used in the binner program to convert fuel rates to energy
rates. The oxidation fractions were taken directly from the Emissions & Sinks report.
Carbon content, the inputs to carbon content and oxidation fraction are shown by fuel
subtype in Table 6-1:
47
-------
Table 6-1: Carbon Content and Oxidation Fraction by Fuel Subtype
Fuel Subtype
Conventional Gasoline
Reformulated Gasoline
E10
Conventional Diesel
Biodiesel (BD20)
Fisher-Tropsch (FT 100)
CNG
LPG
Ethanol (E85)
Methanol (M85)
Gaseous Hydrogen
Liquid Hydrogen
Electricity
M100**
E100**
BD100**
Lower
Heating
Value (KJ/g)
44.0
42.9
-
43.2
-
41.6
45.0
46.4
-
-
120
120
-
20.0
26.9
40.2
Carbon
Weight %
-
-
85.3%
72.4%
82.0%
-
-
0%
0%
-
37.5%
52.2%
77.6%
Carbon
Content
(g/KJ)
0.0196
0.0196
0.0196*
0.0200
0.0199*
0.0205
0.0161
0.0177
0.0194*
0.0189*
0.0
0.0
0.0
0.0188
0.0194
0.0193
Oxidation
Fraction
0.99
0.99
0.99
0.99
0.99
0.99
0.995
0.995
0.99
0.99
0
0
0
-
-
-
* weighted average of blended fuels
** reference fuels only
48
-------
7. Methane (CH4) and Nitrous Oxide (N2O) Rates
7.1 Source Bin Definitions
Unique source bin categories were defined for CH4 and N2O since the factors
important for their formation differ from energy consumption, and because the data used
to generate rates were much more limited than for energy. Important vehicle
characteristics for these pollutants are likely in line with those important for HC and
NOx, e.g. emission standards, vehicle age, emitter category, and vehicle class. The
limited test data on CH4 and N2O restricts the ability to split source bins by all of these
categories.
Emission rates for CH4 and N2O are reported in the Emissions & Sinks report
according to fuel type, broad vehicle class (e.g. light-duty and heavy-duty) and model
year. We decided to stay consistent with this classification, but added the full range of
fuel types and advanced technologies to allow modeling of these bins alongside energy
consumption. The source bin classifications for CH4 and N2O are shown in Table 7-1.
The changes from energy consumption source bin categories are: finer definitions of
model year group to account for changes in standards, elimination of loaded weight and
engine size, and addition of a new category, "regulatory class", meant to capture
differences due to vehicle class-based emission standards.
Table 7-1: Source Bin Categories for CH4 & N2O
Fuel Type
Gas
Diesel
CNG
LPG
Ethanol (E85)
Methanol (E85)
GasH2
Liquid H2
Electric
Engine Technology
Conventional 1C (CIC)
Advanced 1C (AIC)
Moderate Hybrid - CIC
Full Hybrid - CIC
Moderate Hybrid - AIC
Full Hybrid - AIC
Fuel Cell
Hybrid - Fuel Cell
(See Table 4- 14 for
combinations of fuel type
and engine type used in
MOVES2004)
Model Year
Group
1972 and earlier
1973
1974
1999
2000
2001-2010
2011-2020
2021-2050
Regulatory Class
Motorcycle
Light Duty Vehicle
Light Duty Truck
Heavy Duty Vehicle
7.2 Operating Mode Definitions
Virtually no second-by-second data were available for CH4 and N2O, making it
unfeasible to split emissions by the 17 bins devised for total energy. Much of the test
data was conducted over the standard Federal Test Procedure, which limited to ability to
develop emission rates at varying operating conditions. We therefore decided to define a
single operating model for running and a single operating mode for start, with FTP bag
data used as the basis to develop rates for these modes.
49
-------
7.3 Data Sources
EPA has published estimated on-road rates for both CH4 and N2O in the report
entitled "Direct and Indirect Emissions from Mobile Combustion Sources".17 Those
values were based on estimates from Annex E of the Emissions & Sinks report. In both
documents, the rates represented emissions produced (in grams per mile) over the FTP
test.
Under contract, EPA had ICF Consulting revise those estimates of CFLt and N2O
emissions: 1) making use of all available test results, 2) splitting the emissions associated
with engine start from the running emissions, and 3) converting the running emissions
into units of grams per hour (rather than grams per mile). The ICF analyses are
documented in the report entitled "Update of Methane and Nitrous Oxide Emission
1 8
Factors for On-Highway Vehicles".
EPA provided ICF with test results containing methane measurements over
13,277 FTP tests on 6,950 vehicles and 14,636 non-FTP tests on 2,963 vehicles; and with
test results containing nitrous oxide measurements on 95 FTP tests on 64 vehicles and
232 non-FTP tests on 74 vehicles. The non-FTP tests included a hot running 505 as well
as several other driving cycles not utilized in this report. Methane tests were performed
in various U.S. locations during the period between April 1982 and June 2000. Nitrous
oxide tests were performed in various U.S. locations during the period between June
1998 and January 2000. The analyses performed by ICF were limited to the FTP tests that
were performed within the temperature range of 68 degrees to 86 degrees Fahrenheit (i.e.,
at a nominal temperature of 75° F).
7.4 Running and Start Rate Development
Since the goal of the ICF analysis was to develop separate emission rates for both
the running operation and engine starts, the analyses focused on the FTP tests which
contained both of those two types of vehicle operation. The first step was to determine
how to split the FTP emissions into start and running emission rates.
7.4.1 N2O Emission Rates
The FTP test results were reported separately for each of three "bags" (or modes).
Bags 1 and 3 each begin with an engine start, and they have identical driving cycles (3.59
miles requiring 505 seconds of vehicle operation). The Bag 2 mode has no engine start
and requires 867 seconds of vehicle operation to travel 3.86 miles. A relatively small
number of those vehicles had a fourth mode performed following the FTP. This
additional mode was identical to the first bag/mode (hence, identical to the third
bag/mode) with the exception that it contained no engine start. (The fourth mode is
referred to as a HR-505 for "hot running 505.") A comparison among those three 505-
second modes could, therefore, lead to estimates of start emissions. However, the small
50
-------
number of vehicles receiving this additional testing limited the usefulness of such a direct
comparison.
For the FTPs for which N2O was measured, ICF identified a total of 21 for which
that additional 505-second additional mode was performed. (Those 21 vehicles consisted
of 9 gasoline-fueled passenger cars and 12 gasoline-fueled light-duty trucks, all 21 were
determined to be Tier-1 vehicles.) Within each of these two vehicle classes, ICF found
that the difference in N2O emission rates (measured in units of grams per mile) between
the Bag-2 mode and this fourth mode were not statistically significant (at the 95 percent
confidence level). A similar comparison was performed using emission rates in units of
grams per hour. In this comparison, the differences between the emissions of the two
modes were found to be statistically significant. We therefore used the Bag-2 N2O
emission rates (in grams per mile) as estimates of the N2O "running" emissions.
The FTP test is actually a weighted average of those three modes. Thus, the FTP
simulates a driving cycle, nominally 7.44 miles in length, requiring 1,372 seconds of
vehicle operation, and with a single "generic" engine start. The "generic" start is a
weighted average of 57 percent of a hot-start and 43 percent of a cold-start.
Subtracting the Bag-2 N2O emission rate from the weighted FTP emission rate
(both in units of grams per mile) and then multiplying that difference by the actual
weighted FTP distance (varies with each test, but about 7.44 miles) yields an estimate of
the N2O emissions (in grams per start) associated with each generic start. A similar
analyses could have been performed separately for Bag-3 and for Bag-1 to estimate the
individual cold-start and hot-start emission, respectively. However, the MOVES2004
model will not take advantage of that level of precision.
In a few instances, this approach to estimating engine start emissions led to
estimates of the N2O start emissions being negative. That negative value is inconsistent
with the mechanism forming N2O emissions. Therefore, in those few cases, the estimates
of negative N2O start emissions were rounded up to zero.
Multiplying the FTP N2O emission rate (in grams per mile) by the actual
weighted FTP distance traveled and the subtracting the calculated grams per start
emissions produces the estimated total N2O emissions (in grams) from the "running"
operations. Dividing that value by the duration of the driving cycle, 0.38 hours, yields an
estimate of the N2O emissions (in grams per hour) associated with "running" operations
of the vehicle.
These calculations were performed for each of the vehicle type/control technology
groups used in the development of EPA's greenhouse gas inventory reports,19 shown in
Table 7-2. This grouping required conversion to the model year-based source bins
proposed for MOVES2004. Therefore, it was necessary to weight the ICF estimates
together for each model year bin in MOVES according to fractions of control technology
and emission standard by model year used in the Emissions & Sinks report. The resulting
51
-------
model year-based rates, which were used directly in the Emission Rate table of the
MOVES default database, are shown in Appendix F.
Table 7-2: Proposed Start and Running N2O Rates By Control Technology
Vehicle Type
Gasoline Passenger Cars (LDGV)
Gasoline Light-Duty Trucks (LDGT)
Gasoline Heavy-Duty Vehicles (HDGV)
Diesel Passenger Cars (LDDV)
Diesel Light Duty Trucks (LDDT)
Diesel Heavy Duty Vehicles (HDDV)
Motorcycles (MC)
Control
Technology
N2O Running
(g/hr)
LEVs
Tierl
TierO
Oxid Catalyst
Non-Catalyst
Uncontrolled
0.00841
0.28501
0.81124
0.63214
0.25054
0.25054
LEVs
Tierl
TierO
Oxid Catalyst
Non-Catalyst
Uncontrolled
0.02346
0.79287
1.35376
0.82198
0.28336
0.28686
LEVs
Tierl
TierO
Oxid Catalyst
Non-Catalyst
Uncontrolled
0.04791
1.61934
2.76667
1.71067
0.62240
0.65375
Advanced
Moderate
Uncontrolled
0.02303
0.02354
0.02790
Advanced
Moderate
Uncontrolled
0.03209
0.03144
0.03633
Advanced
Moderate
Uncontrolled
0.09598
0.09598
0.09598
Controlled
Uncontrolled
0.10315
0.13173
N2O Start
(g/start)
0.09015
0.11280
0.09183
0.07156
0.02836
0.02836
0.05891
0.20046
0.15324
0.09305
0.03208
0.03247
0.12032
0.40942
0.31318
0.19364
0.07045
0.07400
0
0
0
0
0
0
0
0
0
0.01168
0.01491
52
-------
7.4.2 CH4 Emission Rates
The approach used to split the running and the engine start CH4 emissions
paralleled the one used for N2O emissions. The contractor (ICF) first identified test pairs
(FTPs and FIR-505s) for which methane was measured. A total of 345 such pairs were
identified. Those 345 vehicles consisted of 89 gasoline-fueled passenger cars, 84
gasoline-fueled light-duty trucks, and 172 gasoline-fueled heavy-duty trucks.
As with the N2O analyses, ICF compared the methane emissions on the Bag-2
mode with the corresponding methane emissions on the Hot-Running 505 (HR-505)
mode. In the N2O analyses, ICF found that the Bag-2 emissions (in units of grams per
mile) were a better estimate of the HR-505 than using units of grams per hour; for the
methane emissions, the comparisons were more mixed. However, the Bag-2 methane
emissions in grams per mile, again, appeared to be the better surrogate for the HR-505
emissions.
For estimating methane emissions associated with the "generic" engine start (in
grams per start), we simply subtracted the Bag-2 rate from the weighted FTP rate (both
in grams per mile), multiplied that difference by the actual distance traveled, and round
ed any negative estimates up to zero.
We then estimated the "running" methane emissions by first multiplying the FTP
emissions (in grams per mile) by the actual distance traveled (to estimate total grams of
methane). We then subtracted the calculated methane emissions associated with the
engine start, and then divided that value by the elapse time of the driving cycle (i.e., 0.38
hours).
As with the N2O estimates, these calculations were performed for each of the
vehicle type/control technology groups used in EPA's Climate Leaders Greenhouse Gas
Inventory Protocol, with results shown in Table 7-3. This grouping again required
transformation to the model-year based MOVES bins according to the weighting of
control technology and emission standard by model year. The model year-based rates are
shown in Appendix F.
53
-------
Table 7-3: Proposed Start and Running CH4 Emissions By Control Technology
Vehicle Type
Gasoline Passenger Cars (LDGV)
Gasoline Light-Duty Trucks (LDGT)
Control
Technology
CH4 Running
(g/hr)
LEVs
Tierl
TierO
Oxid Catalyst
Non-Catalyst
Uncontrolled
0.17026
0.23927
1.20220
2.57799
3.02017
3.17229
LEVs
Tierl
TierO
Oxid Catalyst
Non-Catalyst
Uncontrolled
0.20721
0.45605
1.20728
2.54222
3.41579
3.63215
CH4 Start
(g/start)
0.03189
0.05521
0.03425
0.00882
0.05906
0.06203
0.04599
0.08235
0.07245
0.09948
0.06680
0.07103
Gasoline Heavy-Duty Vehicles (HDGV)
Diesel Passenger Cars (LDDV)
Diesel Light Duty Trucks (LDDT)
Diesel Heavy Duty Vehicles (HDDV)
Motorcycles (MC)
LEVs
Tierl
TierO
Oxid Catalyst
Non-Catalyst
Uncontrolled
0.42320
0.47775
3.78625
3.50183
7.50273
8.27757
Advanced
Moderate
Uncontrolled
0.01946
0.01989
0.02358
Advanced
Moderate
Uncontrolled
0.02713
0.02658
0.03070
Advanced
Moderate
Uncontrolled
0.08112
0.08112
0.08112
Controlled
Uncontrolled
1.24339
1.66798
0.09393
0.16309
0.18304
0.21549
0.14672
0.16187
0
0
0
0
0
0
0
0
0
0.02431
0.03262
54
-------
7.5 Advanced Technologies & Alternative Fuels
The Emissions & Sinks report contains estimates of CFLt and N2O emission rates for
five alternative fuels (i.e., methanol, ethanol, CNG, LNG, and LPG) for three vehicle
classes (light-duty cars and trucks, heavy-duty trucks, and buses). Those estimated rates,
reproduced from Table E-14 of Emissions & Sinks, are shown in Table 7-4:
Table 7-4: Emission Factors for CH4 and NiO for Alternative Fueled Vehicles
Vehicle Type
Light-Duty Vehicles
Heavy-Duty Vehicles
Buses
Fuel
Methanol
CNG
LPG
Ethanol
Methanol
CNG
LNG
LPG
Ethanol
Methanol
CNG
Ethanol
N2O (g/mi)
0.063
0.113
0.152
0.076
0.217
0.297
0.440
0.150
0.307
0.217
0.162
0.364
CH4 (g/mi)
0.014
0.914
0.609
0.043
0.646
9.629
6.857
0.108
1.975
0.646
12.416
2.079
To work with MOVES, these rates required separation into start and running
emissions, and conversion to grams per hour (running) and grams per start (start). To
split the start and running emissions, we referred to the comparable gasoline-fueled
vehicles. We split the emissions of the alternative fueled vehicles so that the ratio of the
start to running would be the same as for the corresponding gasoline-fueled vehicles.
The final rates are shown in Appendix F.
Advanced technology rates for CFLi and N2O were developed by applying ratios to
the conventional technology rates - the same method used for energy consumption.
PERE doesn't model CH4 and N2O, and we weren't aware of any data on which to base
ratios. We therefore developed ratios based on those used for energy consumption, under
the assumption that the CFLi: energy and N2O : energy ratios would stay the same
between conventional and advanced technologies. Start energy ratios were used directly
for CH4 and N2O. Running ratios were more involved. Since the running energy ratios
were broken down by operating mode, a composite ratio was needed to apply to the
single running mode used for CFLi and N2O. These were calculated by weighting the
running energy ratios by operating mode with the national default operating mode
distribution generated by MOVES. The advanced technology ratios for CH4 and N2O
55
-------
are shown in Table 7-5. These ratios are in the "ShortTermFERC" file used by the
FERC, although the ratios for diesel advanced technologies are relative to gasoline
conventional 1C in this table and relative to diesel conventional 1C in the FERC file.
Table 7-5: Advanced Technology CH4 and N2O Ratios
(relative to Gasoline Conventional 1C)
Fuel / Technology
Gasoline Advanced 1C
Gasoline CIC Hybrid Moderate
Gasoline CIC Hybrid Full
Gasoline AIC Hybrid Moderate
Gasoline AIC Hybrid Full
Diesel Fuel Conventional 1C
Diesel Fuel Advanced 1C
Diesel CIC Hybrid Moderate
Diesel CIC Hybrid Full
Diesel AIC Hybrid Moderate
Diesel AIC Hybrid Full
Electric
Running Ratio
0.89
0.60
0.53
0.53
0.48
0.64
0.60
0.38
0.35
0.36
0.33
0
Start Ratio
1.0
0.93
0.75
0.83
0.67
0.77
0.67
0.65
0.51
0.61
0.48
0
8. CO2 Equivalent Calculation
MOVES2004 does not currently estimate CC>2 Equivalent emissions, but is
planned for future releases using the methodology presented here. CO2 equivalent is a
combined measure of greenhouse gas emissions weighted according to the global
warming potential of each gas, relative to CC>2. Although the mass emissions of CH4 and
N2O are much smaller than CC>2, the global warming potential is higher, which increases
the contribution of these gases to overall greenhouse effect. CC>2 equivalent is calculated
from CC>2, N2O and CFLi mass emissions according to equation 8-1, meaning that there
will be no direct CC>2 equivalent rates in MOVES.
Eqn 8-1 CO2 Equivalent = CO2 * GWPC02 + CH4 * GWPCH4 + N2O * GWPN20
GWP is Global Warming Potential. The values used for this, shown in Table 8-1, are
contained in the Pollutant table of the MOVES Default Database, and are taken from the
Emission & Sinks report. The values used are for a 100-year time horizon, according to
the guidelines of the United Nations Framework Convention on Climate Change
(UNFCCC).20
Table 8-1: 100- Year Global Warming Potentials
Pollutant
C02
CH4
N20
Global Warming Potential
1
21
320
56
-------
9. Adjustments
The modal binning approach eliminates the need for many of the correction factors
used in MOBILE - for example speed correction factors and off-cycle correction factors.
However, sufficient data doesn't exist to bin other important effects such as temperature,
air conditioning and fuel effects. As in MOBILE, correction factors are necessary to
account for these effects. This section discusses the development of correction factors in
MOVES2004.
9.1 Temperature
Temperature effects were only generated for the start process for total energy only, to
capture the effect of cold starts on energy consumption.
9.1.1 Adjustments for Gasoline Vehicles
9.1.1.1 Data Sources
The vast majority of the FTPs in the MSOD (i.e., 18,676 of 20,156) were
performed within a narrow temperature range (68° to 86° Fahrenheit). This concentration
of test results can result in the analyses being skewed. Therefore, we limited the analyses
of temperature effects to the 580 vehicles (from the MSOD) that had been tested at both
ambient temperatures outside the FTP temperature range as well as within the FTP
temperature range. There were a total of 2,818 FTPs performed on these 580 vehicles.
9.1.1.2 Start Analysis
We used a regression analysis to fit the start fuel consumption (Bag 1 minus Bag
3) using a quadratic curve. We then repeated the analysis using a logarithmic fit. Both
approaches fit the accepted physical model in which, as the ambient temperature
increases: 1) the fuel consumption improves (i.e. decreases) and 2) the rate of
improvement also decreases (i.e., the changes at colder temperatures are more dramatic
than at higher temperatures).
Mathematically, this suggests that we want an equation that is decreasing but
"concave up" (at least between zero and 110° F). A relatively simple approach involves
two linear equations with a common point (which turned out to be 73.9° F). The linear
equations are:
Eqn 9-1 Fuel Consumed During Engine Start (gallons) =
= 0.0919 - (0.0009 * Temperature ), where Temp < 73.9° F
= 0.0402 - (0.0002 * Temperature ), where Temp > 73.9° F
This piecewise linear fit, which is illustrated in Figure 9-1, indicates improving fuel
consumption (with increasing temperature) of 0.0009 gallons per each degree Fahrenheit
57
-------
up to 73.9° F. Then, the improvement continues but at a reduced rate of 0.0002 gallons
per degree above that temperature.
Figure 9-1: Comparison of Fuel Consumed During Engine Start
versus Temperature for MSOD sample
0.30
-0.10
20 40 60 80 100
Ambient Temperature (degrees Fahrenheit)
120
In a series of recent testing programs, EPA's Office of Research and
Development measured the emissions of vehicles with simulated malfunctions operating
on a variety of test fuels. In these studies, nine gasoline-fueled, fuel injected, passenger
cars were tested over the FTP at nominal temperatures of 75°, 20°, 0°, and -20° F.
Additionally, five of those nine cars were also tested at an intermediate temperature of
40° F.
The fuel consumed during engine starts was calculated for all of the tests.
Restricting the analyses to the tests using standard fuel and with no induced malfunctions,
and then regressing the fuel consumed versus the ambient temperature, we obtained the
following linear regression equations:
Eqn 9-2 Fuel Consumed During Engine Start (in gallons) =
= 0.0771 - (0.0011 * Temperature ), where Temp < 20° F
= 0.0669 - (0.0006 * Temperature ), where Temp > 20° F
These two line segments meet at a temperature of 20.4° F. This piecewise linear fit is
illustrated in Figure 9-2.
58
-------
Figure 9-2:_Comparison of Fuel Consumed During Engine Start
versus Temperature for EPA/ORD Dataset
0.15
•c
S
tn
9)
0.10
D)
D
Q
u
D
in
E
3
o
"(5
D)
0.05 o
0.00
-20
0 20 40 60
Ambient Temperature (degrees Fahrenheit)
80
This piecewise linear fit indicates improving fuel consumption (with increasing
temperature) of 0.0011 gallons per each degree Fahrenheit up to 20.4° F. Then, the
improvement drops to 0.0006 gallons per degree from that temperature up to 75° F.
Using Equation 9-1 and 9-2, we can predict the cold-start fuel consumption (for
each vehicle sample) over the temperature range from -20° through 120° Fahrenheit.
Normalizing those predictions (so that the fuel consumption at 75° F is 1.0) for each of
the two analyses produces multiplicative adjustment factors that are illustrated in Figure
9-3.
59
-------
Figure 9-3 Normalized Fuel Consumed During Engine Start versus Temperature
With Quadratic Curve Fit
MSOD Temperature Studies
Low Temperature Studies
-20 0 20 40 60 80 100 120
Temperature (degrees Fahrenheit)
We then developed a single quadratic function that would approximate those
piecewise linear segments. Rather than simply performing a regression analysis on the
normalized test results, we first established the following two criteria that the quadratic
function (which would become our temperature adjustment) must satisfy:
1. The quadratic function must have a value of 1.0 at the temperature that is the
nominal value for the baseline FTPs (i.e., 75° F).
2. The engine-start fuel consumption must improve (i.e., decrease) with increasing
temperatures, for all temperatures within the range of the test data (i.e., from -20C
F through 115° F). And, as we approach the maximum temperature, the
improvement in fuel consumption becomes very small.
The first criterion was met by using an equation of the form:
Eqn 9-3 Factor = A * (Temp - 75)2 - B* (Temp -75) + 1.0
60
-------
The second criterion was met by selecting the vertex of the curve (parabola) to occur at
the temperature of 120° F. This yielded the following relationship between those two
coefficients:
Eqn 9-4 B = 90 * A
The resulting quadratic curve that approximates the piecewise linear segments (pictured
as a dotted curve in the preceding figure) has as its equation:
Eqn 9-5 Factor = 0.000219 * (Temp - 75)2 - 0.01971 * (Temp - 75) + 1.0
Equation 9-5 was used in MOVES2004 to estimate the effects of engine-start
energy consumption of gasoline-fueled vehicles. The coefficients shown above are
stored directly in the table Temperature Adjust, as TempAdjustTermA (-0.01971) and
TempAdjustTermB (0.000219). A placeholder exists in the database for
TermAdjustTermC, but is not currently used.
9.1.1.3 Running Analysis
Subtracting the fuel consumed during engine starts from the FTP results leaves
the fuel consumed over a hot-running LA-4 driving cycle. Analyzing the calculated fuel
consumed over that hot-running LA-4 for the 2,818 FTPs performed on the 580 vehicles
(from the MSOD) that had been tested at both ambient temperatures outside the FTP
temperature range as well as within the FTP temperature range, we found virtually no
change in fuel consumption. Therefore, we are using a temperature adjustment factor of
1.0 for fuel consumed during "running" operations.
9.1.2 Adjustments for Diesel Vehicles
The difference in the process of vaporizing the fuel between gasoline-fueled
vehicles and diesel-fueled vehicles suggests that the diesel-fueled vehicles would be less
sensitive to the ambient temperature than would the gasoline-fueled vehicles (relative to
the amount of fuel consumed during engine start).To test this hypothesis, we examined
the database of 20,156 FTPs and identified 93 FTPs performed on 66 diesel-fueled cars
and trucks. Repeating the regression analysis from Section 2.2, we obtained the results in
Table 9-1 (on this much smaller subset of the test data).
61
-------
Table 9-1: Regression of Start (Bag-1 minus Bag-3) Fuel Consumption
(Diesel-Fueled Cars and Trucks)
Dependent variable is
No Selector
R squared = 13.8%
s= 0.0180 with 93-
Source
Regression
Residual
Variable
Constant
Avg_Temp
Disp (CID)
R squared (adjusted) = 11.8%
3 = 90 degrees of freedom
Sum of Squares df
0.004667
0.029274
Coefficient s.e.
0.055671 0
-0.000536 0
0.000043 0
2
90
ofCoeff
.0103
.0002
.0000
Start
Mean Square
0.002334
0.000325
t-ratio
5.38
-3.43
2.64
(Bag-1 minus Bag-3)
F-ratio
7.17
prob
< 0.0001
0.0009
0.0099
Comparing these two regressions to the previous analyses suggests that for the
diesel-fueled vehicles, the fuel consumption during engine start is substantially less
sensitive to changes in ambient temperature than are the corresponding gasoline-fueled
vehicles (at least within the range of the test data, between 29° and 70° F). To determine
the magnitude of the temperature effect on diesel fuel consumption (during engine starts),
we paralleled the approached used with the larger (gasoline-fueled) sample. That is, we
first identified (within this 66-vehicle sample), 12 diesel-fueled vehicles that were tested
(25 FTPs) at multiple temperatures (29° to 70° F). Then performing a regression analysis
on these tests produces Equation 9-6 (note: the coefficients are different than Table 9-1
because this is a different sample than that used to generate Table 9-1):
Eqn 9-6: Fuel Consumed During Engine Start (gallons) = 0.0564 - (0.0004 * Temperature )
This is illustrated by Figure 9-4.
62
-------
Figure 9-4: Comparison of Fuel Consumed During
Engine Start At Various Temperatures
Oin
10 20
0 00
-010
•
V
**
*
•*
** ^
^-
0 20 40 60 80 100 120
Ambient Temperature (degrees Fahrenheit)
This regression equation indicates improved fuel consumption (with increasing
temperature) of 0.0004 gallons per degree within the temperature range of the test data
(29° to 70° F). While the regression equation in Section 2.3 indicates improved fuel
consumption (with increasing temperature) of 0.0009 gallons per degree for that same
temperature range. This suggests that the change in fuel consumption during engine
starts resulting from change in the ambient temperature for diesel-fueled vehicles is about
0.44 (i.e., 0.0004 / 0.0009) times the corresponding change in the fuel consumption for
gasoline-fueled vehicles.
If the magnitude of the effect of ambient temperature on the engine-start fuel
consumption of diesels is 44 percent of the corresponding effect on gasoline-fueled
engines, then the temperature adjustment factor for diesel vehicles is derived by
multiplying the gasoline temperature adjustment coefficients from Equation 9-5 by 0.44,
resulting in Equation 9-7:
Eqn 9-7 Factor = 0.000096 * (Temp - 75)2 - 0.00867 * (Temp - 75) + 1.0
MOVES2004 employs these coefficients (in the Temperature Adjust table) to
adjust, for varying ambient temperatures, the engine-start fuel consumption of diesel-
fueled vehicles. As with the gasoline-fueled vehicles, we propose a temperature
adjustment factor of 1.0 for fuel consumed during "running" operations for the diesel-
fueled vehicles.
63
-------
9.2 Air Conditioning
MOVES2004 accounts for increased energy consumption due to air conditioning
usage through an air conditioning adjustment factor. The air conditioning adjustment is a
combined measure of air conditioning activity (the fraction of time the air conditioning
compressor is engaged) and the increase in energy consumption when the compressor is
engaged. This approach was first established in MOBILE6.21'22 In MOVES2004, air
conditioning adjustments are applied only to running total energy. Analysis performed
for MOBILE6 indicates that start energy and emissions rates are not highly impacted by
A/C usage, and there is not sufficient data of the effects of A/C usage on CH4 and N2O
emissions to merit an adjustment for these pollutants.
A/C adjustment factors stored in the MOVES Default database energy usage
reflect "full-usage", e.g. when the compressor is engaged. Within MOVES2004 these
are scaled back according to the fraction of time the compressor is engaged, the
penetration of A/C in the fleet, and an estimate of malfunctioning systems. The latter
three components are documented in the Fleet & Activity report.
Full-usage energy adjustment factors are stored in the Full AC Adjustment table in
MOVES Default. The are broken down by the 17 operating mode bins discussed in
Section 4.2 (Table 4-3). The adjustments were broken down by operating mode because
test data shows that the relative load increase caused by the A/C compressor varies
significantly depending on engine load - at low loads the relative increase is high, at high
loads the relative increase is low.
9.2.1 Data Sources
A subset of data from the MSOD contains a sample of second-by-second data on
a variety of test schedules tested with A/C on and off. The majority of these data are
from a program EPA conducted when gathering data from MOBILE6, which ran 38
LDVs and LDTs over several drive cycles at high A/C load conditions. Additional
details of this testing can be found in MOBILE6 documentation.
9.2.2 Analysis Methodology
Using the subset of vehicles with paired A/C on and off tests, the binner program
generated energy consumption rates with and without the A/C on, grouped by the source
bins for which the test vehicles fell into. The ratio of A/C-on to A/C-off was calculated
for each operating mode as the Full AC Adjustment, by source bin. Since the source bin
coverage of the 38-vehicle dataset wasn't broad enough to merit keeping the A/C
adjustment split by source bin, we developed a composite adjustment as a weighted
average across source bins, using the source bin fraction (i.e. relative occurrence of each
source bin in the fleet) as a weighting factor. The resulting adjustments are shown in
Table 9-2.
64
-------
Table 9-2 Full A/C adjustments by operating mode bin
Braking: 1.34
Idle: 1.36
VSP \ Speed
< 0 kw/tonne
Oto3
3 to 6
6 to 9
9 to 12
12 and greater
6 to 12
<6
0-25mph
1.31
1.25
1.19
1.17
1.15
1.13
25-50
1.29
1.22
1.19
1.17
1.16
1.13
>50
1.20
1.16
1.14
Since paired A/C on and off data is very limited, we decided to calculate a single
set of adjustment and apply them to all source use types in the model. This means light-
duty A/C effects would be applied to all vehicle classes, including heavy-duty.
10. Well-To-Pump Energy & Emission Rates
An updated version of the GREET model, developed by Argonne National
Laboratory, has been integrated into MOVES2004 to provide energy and emission rates
for the well-to-pump process. A database table named GREETWellToPump in the
MOVES Default database contains default well-to-pump rates generated by the integrated
version of GREET; the user only needs to run GREET if a change from default well-to-
pump assumptions is desired. These rates are expressed in terms of well-to-pump energy
use (or emissions for CH4 and N2O) per unit pump-to-wheel energy consumption - i.e.
well-to-pump KJ per pump-to-wheel KJ, or well-to-pump grams per pump-to-wheel KJ.
A main feature of the updated GREET model is the ability to generate these rates
by calendar year, taking into account changes in fuel production efficiencies or pathway
mixes over time. The rates used in MOVES therefore vary by fuel subtype and calendar
year. Rates by fuel subtype for two calendar years, 1999 and 2020, are shown in Tables
10-1 and 10-2. The full set of calendar year rates can be accessed by querying the
GREETWellToPump table in the MOVES Default database. Because the energy rates in
Table 10-1 are expressed as KJ upstream per KJ downstream, the numbers are simply
well-to-pump energy consumption expresses as the fraction of pump-to-wheel energy
consumption. Results for the hydrogen pathways are under development and will be
added to the model when they are ready.
65
-------
Table 10-1:
Default Well-To-Pump Energy Rates for Calendar Year 1999 & 2020
(Well-To-Pump KJ / Pump-To-Wheel KJ)
Fuel Subtype
Conventional Gasoline
Reformulated Gasoline
E10
Conventional Diesel
Biodiesel (BD20)
Fisher-Tropsch (FT 100)
CNG
LPG
Ethanol (E85)
Methanol (M85)
Electricity
Total Energy
1999
0.229
0.266
0.264
0.179
0.275
0.725
0.151
0.114
0.631
0.520
1.654
2020
0.236
0.270
0.268
0.213
0.291
0.643
0.146
0.111
0.613
0.452
1.499
Petroleum Energy
1999
0.107
0.105
0105
0.082
0.101
0.020
0.008
0.030
0.088
0.055
0.095
2020
0.110
0.108
0.109
0.099
0.111
0.020
0.006
0.025
0.088
0.055
0.067
Fossil Energy
1999
0.225
0.262
0.259
0.176
0.269
0.724
0.140
0.113
0.620
0.519
1.624
2020
0.232
0.266
0.265
0.210
0.286
0.643
0.137
0.110
0.604
0.451
1.476
gas and diesel subtypes account for phase-in of low sulfur fuel in required years
Table 10-2:
Default Well-To-Pump Emission Rates for Calendar Year 1999 & 2020
(Well-To-Pump Grams / Pump-To-Wheel KJ)
Fuel Subtype
Conventional Gasoline
Reformulated Gasoline
E10
Conventional Diesel
Biodiesel (BD20)
Fisher-Tropsch (FT100)
CNG
LPG
Ethanol (E85)
Methanol (M85)
Electricity
C02
1999
0.017
0.018
0.015
0.014
0.004
0.026
0.011
0.008
-0.006
0.022
0.203
2020
0.017
0.018
0.015
0.016
0.006
0.022
0.010
0.008
-0.009
0.018
0.189
CH4
1999
l.OOE-4
1.16E-4
1.01E-4
9.71E-5
9.51E-5
1.16E-4
2.35E-4
1.08E-4
1.08E-4
1.37E-4
2.81E-4
2020
1.01E-4
1.16E-4
1.02E-4
9.99E-5
9.61E-5
1.16E-4
2.35E-4
1.11E-4
1.07E-4
1.35E-4
2.55E-4
N2O
1999
2.77E-7
2.95E-7
3.68E-6
2.24E-7
1.73E-6
9.51E-8
1.59E-7
1.44E-7
3.91E-5
3.76E-7
2.69E-6
2020
2.91E-7
3.08E-7
3.48E-6
2.66E-7
1.60E-6
9.56E-8
1.72E-7
1.41E-7
3.67E-5
3.68E-7
3.08E-6
These rates are a product of the default GREET inputs for each fuel pathway,
including variables such as gasoline oxygenate content and natural gas export share -
there are in total several hundred inputs that are too numerous to list here, but can be
found in separate documentation. A report to be included with MOVES2004 technical
documentation entitled "GREET Users Manual and Technical Issues for MOVES
Integration", prepared by Argonne Labs, documents many of the "top-level" default
inputs. More extensive documentation of underlying GREET methodologies can be
found on the GREET website (available online at
http ://www.transportation. anl .gov/software/GREET/index.html
66
-------
11. Data Source Identification
From the preceding sections it should be clear that emission rates were generated
from several different approaches: direct binning of test data, a variety of hole-filling
methods, and estimation of rates based on ratios of advanced technology to conventional
technology performance. To provide a means of keeping track of how a specific
emission rate in the default database was generated, we added the field "data source ID"
to each record of the emission rate table. DataSourcelD is a meta-data field which tells
the user which method was used to generate the rate found in that record, distinguishing
among the many methods discussed in this report. The full description of data source IDs
are contained in the DataSourcelD table in the MOVES default database, and are also
shown in Tables 11-1 and 11-2.
67
-------
Table 11-1: Data Source ID descriptions
ID
1001
1002
2001
3001
4011-4242
5001
5002
5003
6000
Description
Data: Binned from second-by-second data
Data: Calculated from bag data
Hole Filling: Manual
Hole Filling: Physical Emission Rate Estimator
Hole Filling: Interpolator / Copier Program (see Table 11-2)
Hole filling: Pre-2001 Alt Fuel & Adv Tech script
Hole filling: 2001-2010 Diesel Conv 1C script
Hole filling: Extended Idle Calculation script
Future Emission Rate Creator
Table 11-2: Data Source ID descriptions for Interpolator / Copier Program
ID
4011
4012
4021
4022
4031
4032
4041
4042
4051
4052
4061
4062
4111
4112
4121
4122
4131
4132
4141
4142
4151
4152
4161
4162
4241
4242
Description
all opmodes except idle interpolate by weight class in first pass
all opmodes except idle interpolate by weight class in second pass
all opmodes except idle interpolate by displacement class in first pass
all opmodes except idle interpolate by displacement class in second pass
all opmodes except idle interpolate by model-year group in first pass>
all opmodes except idle interpolate by model-year group in second pass>
all opmodes except idle adopt meanbaserate from neighboring sourcebin by model-year
group in first pass
all opmodes except idle adopt meanbaserate from neighboring sourcebin by model-year
group in second pass
all opmodes except idle adopt meanbaserate from neighboring sourcebin by weight class in
first pass
all opmodes except idle adopt meanbaserate from neighboring sourcebin by weight class in
second pass
all opmodes except idle adopt meanbaserate from neighboring sourcebin by displacement
class in first pass
all opmodes except idle adopt meanbaserate from neighboring sourcebin by displacement
class in second pass
idle opmode interpolate by displacement class in first pass
idle opmode interpolate by displacement class in second pass
idle opmode interpolate by weight class in first pass
idle opmode interpolate by weight class in second pass
idle opmode interpolate by model-year group in first pass>
idle opmode interpolate by model-year group in second pass>
idle opmode adopt meanbaserate from neighboring sourcebin by model-year group in first
pass
idle opmode adopt meanbaserate from neighboring sourcebin by displacement class in first
pass
idle opmode, adopt meanbaserate from neighboring sourcebin, by displacement class, in first
pass
idle opmode adopt meanbaserate from neighboring sourcebin by displacement class in
second pass
idle opmode adopt meanbaserate from neighboring sourcebin by weight class in first pass
idle opmode adopt meanbaserate from neighboring sourcebin by weightt class in second
pass
idle opmode adopt meanbaserate from neighboring sourcebin by model-year group in second
pass
idle opmode adopt meanbaserate from neighboring sourcebin by displacement class in
second pass
68
-------
12. Acknowledgments
The authors gratefully acknowledge the following individuals for their ideas and
support towards the work presented in this report: Chad Bailey, Connie Hart and David
Brzezinski of EPA's Office of Transportation and Air Quality; Michael Wang, Yu We,
Dan Santini, Feng An, Anant Vias and Amgad Elgowainy of Argonne National
Laboratory; Matt Earth, Ted Younglove and George Scora of UC Riverside; Chris Frey
of North Carolina State University; and Nigel Clark and Ralph Nines of West Virginia
University.
69
-------
Appendix A: Binning Methodology Proof-Of-Concept
A.I Background
The section lays out the methodology for determining the 17 operating mode bins
used for running total energy consumption. This analysis builds on previous studies
conducted in the preliminary stages of MOVES, including 1) the on-board emission
"shootout" involving multiple participants,23 2) the proof-of-concept modal binning
analysis conducted by North Carolina State University (NCSU),24 and 3) follow-on
investigation of the binning approach by EPA.25 The reader should consult the
referenced reports for a full explanation of these findings. The conclusions drawn from
these studies which form the basis of the work presented in this section are:
• The process of binning - i.e. calculating average mass per time emission rates
from second-by-second data within a pre-defined range of vehicle operation - is
an effective means of characterizing emission changes from vehicle operational
changes.
• Vehicle Specific Power (VSP) is an effective parameter for binning, but does
result in some bias at lower speeds if not supplemented with additional variables.
• Average cycle speed was initially proposed as a supplemental binning variable,
with some success. However, a drawback of this approach is not having the
ability to define bins based on "real-time" vehicle operation.
A.2 Investigation of Alternatives: Engine Friction and RPM
Additional analysis focused on the role of engine friction in generating the bias
which was appearing with the VSP-only approach. The effect of engine friction on fuel
consumption is relatively large at low loads. To account for this we investigated
modifying the VSP equation to explicitly include engine friction. While showing some
promise, this approach was ultimately scrapped because it wasn't feasible to determine an
appropriate value (or algorithm) for the additional engine friction "gamma" term which
could apply across vehicles.26
Engine friction is directly proportional to RPM, so the idea of binning by VSP
and RPM was considered. It was rejected because RPM is not directly available in many
datasets, and would need to be derived from vehicle speed and estimates of transmission
logic and gear ratio. To do this on a vehicle-to-vehicle basis would not be feasible.
A.3 VSP/Speed Binning Proof-Of-Concept
Instantaneous vehicle speed was chosen as the second binning parameter. Speed
serves as a reasonable surrogate for engine friction and is readily available across the
range of datasets which would be considered for MOVES. NCSU's analysis of modal
binning approaches showed that speed was an important binning variable, and that model
performance would be improved by using it in conjunction with VSP.
70
-------
The primary objective of this analysis was to develop bin definitions using VSP
and instantaneous speed, which could be applied across all source use types and
pollutants, and which improved on VSP-only approaches. The VSP-only approach used
as the basis for comparison was a 14 VSP bin approach developed by NCSU and
documented in their modal binning analysis report. Another objective of the analysis, in
response to comment received on the proposed methodology (Appendix G), was to show
the feasibility of the binning approach on a vehicle sample which included heavy-duty
vehicles and high-emitting light-duty vehicles, as the preliminary binning analyses
focused on later model light-duty vehicles.
A.3.1 Data Sources
The datasets used for the light-duty analysis was compiled from 26 LDVs and
LDTs tested by ARE over their Unified Cycle Correction (UCC) cycles, and 11 LDVs
tested on-road by EPA as part of the "shootout" dataset. For the latter, only the segments
of testing run under warmed-up conditions with the A/C off were used. The model year
range for the ARB dataset was from 1983 - 1998, and included high emitting vehicles in
the sample. High emitters were purposely included to address the question of whether
the modal binning approach would work on high as well as low emitting vehicles. The
EPA dataset consisted of 1996 and newer vehicle. The combined dataset consisted of
253 unique "trips".
The dataset used for the heavy-duty analysis was compiled from 11 Class 8 trucks
tested by CE-CERT on-road using their heavy-duty trailer configuration, over a variety of
cycles, and EPA shootout data on 15 Transit Buses. These data were separated into 64
"trips".
A.3.2 Bin Options Evaluated
Five binning approaches were developed for the analysis, termed Bin Option 1
through 5 (BO1 through BOS). The approaches are shown in Table A-l. Each of the
options defined idle and "braking" as distinct modes; this approach was initially proposed
as part of NCSU's proof-of-concept work. We did change "decel" to "braking" in order
to segregated actual braking operation, which is useful for modeling hybrid vehicles with
regenerative braking. Whereas the NCSU work defined one mode (bin) each for cruise
and accel, we expanded cruise and accel operation into subdivisions of VSP and speed.
71
-------
Table A-l: Bin Approaches Analyzed
Braking (common)
Idle (common)
Cruise/Accel
Bin Option 1
Bin Option 2
Bin Option 3
Bin Option 4
Bin Option 5
Accel
Speed = 0
< -2 mph/s, or <-l mph/s for 3 consecutive seconds
(used between -1 and 1 mph to account for signal noise)
VSP Bin Ranges (kw/ton)
<
2
2-6
6-10
10-14
>
<
14
1
1-4
4-7
7-10
10-14
14-20
>20
<
1
1-4
4-7
7-10
10-13
13-16
16-19
>
<
19
1
1-4
4-7
7-10
10-13
>
< 50 mph
<0
0-3
3-6
6-9
9-12
13
> 50 mph
<6
6-12
> 12
Instantaneous Speed Ranges (mph)
0-15
15-45
45-60
>60
0-40
40-55
>55
0-40
>40
0-30
>30
0-25
25-50
>50
The determination of bin options began with performing Hierarchical Tree-Based
Regression (HTBR) on the analysis dataset, using the open-source statistical software
package R. Separate HTBR runs were performed on the light-duty and heavy-duty sets,
and by pollutant. Since the goal of the analysis was to come up with a common set of bin
definitions across vehicle class and pollutant, the bin approaches were composites of the
HTBR runs. In general speed cutpoints were above 40 mph. VSP cutpoints were not as
cleanly defined, so the VSP points used were picked arbitrarily. BO1 was an initial
attempt at a composite of the HTBR results. BO2 refined BO1 to more evenly distribute
time/emissions for both light-duty and heavy-duty, add higher VSP bin to improve CO
prediction. BOS, 4 and 5 were based on bins which could be filled by EVI240 alone: BOS
was the initial cut, BO4 a refinement to improve distribution of time (i.e. making sure
bins would be adequately filled by the IM240), and BOS adds higher speed bins to
72
-------
improve fuel prediction and reshuffles VSP and speed outpoints to maintain distribution
of time in each bin.
The latter three bin approaches focused on the IM240 because (as detailed in
Section 4.3.1) much of the second-by-second data for light-duty vehicles used in
MOVES2004 was from the New York Instrumentation Protocol Assessment (NYIPA)
program, which included over 10,000 lab-grade EVI240 tests. Since this program has the
most vehicle coverage by far of any of the MOVES2004 data sources, allowing the bins
to be completely filled with IM240 would improve coverage across the source bins. This
approach also leaves the door open for using EVI240 program data directly, which could
make emitter distributions for criteria pollutants more straightforward.
A.3.3 VSP Calculation
VSP was calculated according to Equation A-l. Road load coefficients for light-
duty were derived from the 50 mph road load horsepower values reported for each
vehicle. The coefficients for heavy-duty were derived from Petrushov.27 The weight of
the CE-CERT trailer test rig, as estimated by CE-CERT staff (44,000 Ibs), was added to
the reported cab weight.
Equation A-l: VSP = (A* Speed + B * Speed2 + C * Speed3 + Mass * Speed * Accel) / Mass
Where:
VSP is in KW/Metric Ton
Speed is in meters/second (mps)
Accel is in meters/second2
A is rolling resistance term in KW / mps
B is friction term in KW / mps2
C is aerodynamic drag term in KW / mps3
Mass is in metric tons (1000 kg)
73
-------
A.3.4 Source Bin Assignment
The analysis followed the same steps MOVES would in modeling an aggregate
estimate of fuel consumption and emissions. The first step was to assign the vehicles in
the analysis set to source bins. For fuel consumption, the MOVES2004 source bin
categories for energy consumption (Table 4-1) were used to assign each vehicle to a bin.
For HC, CO, and NOx source bins were defined based on categories of pollutant/CO2
ratio by trip. The criteria for defining a bin were that the range of emission points within
any bin should not be more than 25 percent of the range of the entire sample. We
originally looked at defining categories based on vehicle emissions (e.g. emitter
categories). Large trip-to-trip variability reduced the effectiveness of this approach.
A.3.5 Analysis Methodology and Results
A random sample of trips selected out of from light-duty and heavy-duty analysis
sets (validation sample). For light-duty, 20 percent of the trips were chosen (64 trips);
for heavy-duty, 30 percent of trips (23 trips). Binned fuel consumption and emission
rates were produced by averaging all one-second values in each source bin and operating
mode bin for the remaining trips (prediction sample). Fuel consumption and emission
rates predicted for each second in the validation sample using the fuel consumption and
emission rates from predicted sample, for the source bin and operating mode bin in that
second. Fuel consumption and emissions aggregated over each trip in the validation
sample to generate predicted and observed trip-average gram per second fuel
consumption and emission rates.
Since MOVES will predict aggregate emissions produced by several vehicles over
a period of time (e.g. one hour), the primary evaluation metric for this analysis is the
magnitude of difference between the average of the observed and predicted
fuel/emissions over all trips in the validation sample (sample average). Table A-2 shows
the results of this comparison for each of the alternatives evaluated, in terms of percent
difference from observed for the average per-trip emissions. These comparisons are
shown for a) all trips in the sample, b) all trips with an average speed < 30 mph, and c) all
trips with an average speed > 30 mph. The purpose of the latter two comparisons is to
identify biases for lower speed or higher speed cycles; in some cases very good
agreement in the "all trip" case can mask large differences in each speed range. The VSP
row is for the 14 VSP bin approach developed by NCSU.
74
-------
Table A-2: Sample Average Results (percent difference observed vs. predicted)
All Trips
Light-Duty
VSP
BO1
BO2
BOS
BO4
BOS
Fuel
9%
4%
4%
5%
6%
4%
HC
1%
2%
2%
1%
0%
1%
CO
6%
4%
5%
6%
3%
3%
NOx
5%
-2%
-3%
-1%
1%
-3%
Heavy-Duty
Fuel
1%
-3%
-2%
-1%
HC
15%
9%
10%
10%
CO
13%
15%
16%
14%
NOx
-3%
-5%
-5%
-4%
Trips w/ Average Speed < 30
Light-Duty
VSP
BO1
BO2
BOS
BO4
BO5
Fuel
22%
8%
7%
10%
14%
8%
HC
14%
6%
6%
9%
9%
6%
CO
8%
5%
5%
5%
4%
4%
NOx
14%
-8%
-8%
-5%
0%
-7%
Heavy-Duty
Fuel
10%
1%
1%
0%
HC
36%
25%
25%
23%
CO
25%
20%
18%
21%
NOx
19%
5%
5%
7%
Trips w/ Average Speed > 30
Light-Duty
VSP
BO1
BO2
BOS
BO4
BOS
Fuel
-1%
2%
2%
2%
0%
1%
HC
-6%
-1%
0%
-3%
-6%
-2%
CO
6%
4%
6%
6%
3%
3%
NOx
1%
0%
0%
0%
1%
-1%
Heavy-Duty
Fuel
-6%
-6%
-4%
-2%
HC
-15%
-12%
-11%
-9%
CO
-4%
7%
13%
3%
NOx
-16%
-8%
-11%
-11%
In general adding speed bins improves prediction relative to the VSP-only
approach, based on the results by speed range. This is particular the case for fuel
consumption. For light-duty, BO1, BO2 and BOS predict average trip fuel consumption
within 10 percent over all trips and for both speed ranges. For heavy-duty, BO1, BO2
and BOS generally predicts average trip fuel consumption and emission within 15 percent
over all trips and for high-speed trips; HC and CO predict high on lower speed trips.
Overall, BO2 and BOS performed the best.
We also looked at the distribution of time and emissions in each bin. Tables A-3
and 4 show the distribution of time and emissions in the analysis datasets for two bin
options, BO2 and BOS (plus the distribution of time on the IM240 cycle for BOS). This
comparison shows that the distribution of time and emissions is more balanced BO2.
Basing bin definitions on the IM240 results in unbalanced bin distributions for heavy-
duty, where the majority of operation occurs above 50 mph. However, BOS prediction
results were not hurt by this in this analysis.
75
-------
Table A-3: Distribution of Time, Fuel Consumption and Emissions by Bin for BO2
Bin
(Spd/VSP)
Brake
Idle
<40/<1
40-55/55/55/l-4
<40/4-7
40-55/4-7
>5 5/4-7
<40/7-10
40-55/7-10
>55/7-10
<40/10-14
40-55/10-14
>55/10-14
<40/14-20
40-55/14-20
>55/14-20
<40/>20
40-55/>20
>55/>20
Light-Duty
Time
12.3%
11.5%
12.0%
3.4%
2.4%
7.8%
2.7%
2.6%
6.1%
2.9%
4.2%
4.6%
2.2%
5.0%
3.9%
1.9%
5.2%
2.4%
1.0%
3.3%
0.7%
0.4%
1.7%
Fuel
6.4%
4.6%
6.4%
2.5%
3.0%
5.7%
2.4%
3.5%
5.7%
2.9%
6.2%
5.2%
2.6%
8.2%
5.2%
2.5%
9.4%
3.9%
1.6%
6.3%
1.3%
0.9%
3.6%
HC
8.3%
5.8%
7.7%
2.8%
2.9%
5.9%
2.1%
2.8%
5.5%
2.5%
4.6%
5.5%
2.4%
6.2%
5.6%
2.5%
7.7%
4.9%
2.0%
5.8%
2.2%
1.2%
3.1%
CO
5.8%
2.6%
5.7%
2.1%
3.5%
4.7%
1.6%
3.0%
4.7%
2.0%
4.7%
4.9%
2.3%
6.9%
5.5%
2.7%
10.1%
5.4%
2.4%
8.5%
2.8%
2.4%
5.9%
NOx
6.6%
2.5%
5.3%
2.5%
4.1%
4.3%
2.2%
4.6%
4.0%
2.6%
8.1%
3.6%
2.3%
11.6%
3.5%
2.1%
13.4%
2.6%
1.4%
7.3%
0.8%
0.9%
3.8%
Heavy-Duty
Time
5.3%
18.8%
17.3%
2.4%
5.8%
6.7%
4.0%
4.8%
4.6%
4.6%
6.5%
3.3%
2.6%
4.2%
1.9%
1.3%
2.6%
0.5%
0.7%
1.4%
0.0%
0.2%
0.5%
Fuel
0.5%
2.9%
4.1%
1.7%
3.7%
5.4%
5.8%
6.7%
6.0%
8.6%
11.1%
5.8%
6.0%
8.6%
3.7%
3.3%
6.1%
1.2%
2.3%
4.1%
0.1%
0.8%
1.6%
HC
1.1%
3.4%
6.8%
3.7%
6.6%
3.1%
9.6%
9.1%
2.0%
12.3%
12.4%
1.4%
7.2%
10.0%
0.8%
2.8%
4.3%
0.2%
0.8%
1.6%
0.0%
0.3%
0.6%
CO
1.4%
5.9%
9.7%
2.2%
3.5%
10.5%
4.3%
4.2%
10.9%
6.3%
6.4%
10.4%
4.6%
4.9%
4.4%
2.4%
3.1%
0.9%
1.0%
1.8%
0.1%
0.4%
0.7%
NOx
0.9%
3.5%
4.4%
1.9%
3.8%
4.9%
6.3%
7.3%
4.8%
9.4%
12.4%
4.1%
6.1%
8.9%
2.6%
3.3%
6.0%
0.8%
2.3%
3.9%
0.1%
0.8%
1.5%
76
-------
Table A-4: Distribution of Time, Fuel Consumption and Emissions by Bin for BOS
Bin
(Spd/VSP)
Brake
Idle
<25/<0
25-50/<0
<25/0-3
25-50/0-3
<25/3-6
25-50/3-6
<25/6-9
25-50/6-9
<25/9-12
25-50/9-12
<25/>12
25-50/>12
>50/<6
>50/6-12
>50/>12
Light-Duty
Time
12.3%
11.5%
3.9%
6.2%
5.5%
5.7%
2.8%
5.8%
2.3%
4.5%
1.7%
3.1%
1.8%
4.3%
7.9%
11.0%
9.9%
Time
(IM240)
13.3%
4.6%
6.7%
3.8%
9.2%
4.2%
5.4%
13.8%
3.3%
2.1%
4.6%
1.7%
2.5%
6.3%
5.8%
7.5%
5.4%
Fuel
6.4%
4.6%
1.8%
3.6%
3.5%
4.0%
2.5%
5.1%
2.5%
4.7%
2.1%
3.8%
2.8%
7.2%
15.3%
17.8%
12.5%
HC
8.3%
5.8%
2.1%
4.7%
3.7%
4.2%
2.3%
5.0%
2.4%
4.9%
2.1%
4.1%
3.1%
9.4%
13.9%
13.7%
10.4%
CO
5.8%
2.6%
1.5%
3.4%
2.7%
3.3%
1.9%
4.0%
1.9%
4.4%
1.9%
4.0%
3.2%
11.4%
21.3%
15.8%
10.9%
NOX
6.6%
2.5%
1.2%
3.8%
2.2%
3.7%
1.4%
4.4%
1.3%
4.1%
1.1%
3.1%
1.4%
5.8%
18.0%
24.0%
15.7%
Heavy-Duty
Time
5.3%
18.8%
6.6%
4.3%
9.7%
3.8%
2.6%
4.0%
2.0%
2.8%
1.0%
1.6%
0.4%
1.5%
3.6%
11.0%
20.9%
Fuel
0.5%
2.9%
1.0%
1.0%
4.6%
2.9%
3.5%
4.5%
3.7%
4.5%
2.1%
3.2%
0.9%
3.7%
10.3%
23.7%
27.0%
HC
1.1%
3.4%
2.8%
1.5%
4.4%
2.5%
1.3%
3.3%
0.8%
2.4%
0.4%
1.4%
0.1%
1.0%
4.7%
27.1%
41.6%
CO
1.4%
5.9%
2.9%
2.3%
10.0%
3.3%
8.1%
4.2%
8.8%
4.0%
4.2%
2.7%
1.1%
2.2%
5.0%
14.8%
19.1%
NOX
0.9%
3.5%
1.2%
1.3%
4.1%
3.2%
2.4%
4.6%
2.3%
4.1%
1.2%
2.8%
0.5%
3.2%
10.1%
24.8%
29.8%
Although BO2 has a more even distribution of time and emissions for the analysis
dataset, we chose to use BOS because it met the objective of the analysis (good
performance across vehicle class and pollutant) and represents the best balance between
performance and source/operating mode bin coverage with available data sources,
particularly for light-duty (BO2 has bins which the EVI240 couldn't fill). Using this
approach will enable the New York IPA dataset to populate bins across the spectrum of
the light-duty fleet without requiring PERE to fill more extreme operating mode bins.
Possible downsides of BOS are the unbalanced bin distributions for heavy-duty, and the
heavier reliance on the relatively lax IM240 cycle to populate even the aggressive driving
bins. We will revisit bin definitions for future implementations, as more data
(particularly PEMS data) becomes available. The MOVES design is flexible in that bin
definitions can be changed with only minor programming, database, and binner program
changes.
To understand patterns across the spread of fuel consumption and emissions,
predicted vs. observed per-trip results are shown in the following charts (Figures A-l
through A-8) for Bin Option 5, with 1:1 line superimposed. Per-trip predictions for
BOS show good performance across the range of fuel consumption and emissions,
indicating the robustness of the approach across emitter class and driving behavior.
77
-------
2.5
i/r 2.0 -
0
-i—*
CD
— 1.5 -
0
I- 1.0 -
T3
0
0
W
s -^
0.0
Light Duty Fuel Consumption
0.0
.5
1.0
1.5
2.0
2.5
12
10-
a>
ro
o:
"ai
T3
CD
£
CD
CO
.a
O
6-I
4-
2-
Predicted Trip Fuel Rate (g/s)
Heavy Duty Fuel Consumption
10
12
Predicted Trip Fuel Rate (g/s)
78
-------
"co
Q:
o
T3
CD
CD
C/>
.Q
O
Light Duty HC
.010
.008-
.006-
.004-
.002-
0.000
0.000
.002
.004
.006
.008
.010
O
X
Q.
T3
CD
CD
W
.Q
O
.012
.010-
3 .008 H
.006-
.004-
.002-
0.000
Predicted Trip HC Rate (g/s)
Heavy Duty HC
0.000
.002
.004
.006
.008
.010
.012
Predicted Trip HC Rate
79
-------
Light Duty CO
Predicted Trip CO Rate (g/s)
Heavy Duty CO
Predicted Trip CO Rate (g/s)
80
-------
Light Duty NOx
0.00
0.00
Predicted Trip NOx Rate (g/s)
.6-
co .4
o:
x
O
.3-
T3
0> .2
CD
C/>
O .1-
0.0
Heavy Duty NOx
0.0
.2
.3
.4
.5
.6
Predicted Trip NOx Rate (g/s)
81
-------
Appendix B: Calculation of Running Energy
Consumption Rates Using "The Binner Program"
B.I The EmissionRate Table
This section describes the development of emission rates for runningEnergy
consumption rates from motor vehicles in the MOVES emissionRate table. The
emissionRate Table includes five fields, as shown in Table B-l. Consistent with the
MOVES modal approach, the table contains mean base emission rates (meanBaseRate)
and associated estimates of uncertainty in these means for motor vehicles classified as
"emissions sources" (sourceBinID), and by "operating mode" (opModelD). The
uncertainty estimates are expressed as coefficients of variation for the mean
(meanBaseRateCV). In this section, we will describe the processes of data classification
by source bin and operating mode, calculation of energy consumption, and statistical
evaluation of the results.
Table B-l. Fields in the MOVES Emission Rate Table
Label
SourceBinID
OpModelD
MeanBaseRate
MeanBaseRateCV
PolProcessID
Symbol
—
Jcell
CV?
—
Description
Source Bin
Operating Mode (defined
specifically for a given pollutant
process)
Cell mean (for Energy
Consumption)
Coefficient of variation of the cell
mean
Pollutant Process = 9101 (running
energy)
B.2 Data
The data used to populate the emissionRate Table were compiled from selected
test programs and studies. Following quality-assurance, the data was loaded into the
Mobile Source Observation Database (MSOD).28 The identity and scope of specific
programs and studies is described in Section 3.
B.2.1 Fields Extracted from the Mobile-Source Observation Database
Consistent with the approach adopted to estimate modal emission rates, this
analysis relied on data collected on a continuous "second-by-second" basis, i.e., on a
measurement frequency of approximately 1.0 hertz. No data representing integrated
"bag" samples was included. In addition, all data used was collected under laboratory
conditions using chassis dynamometers following a variety of prescribed test procedures;
no data collected "in-use" using portable instrumentation was included. Prior to data
82
-------
classification and analysis, we built two temporary data tables by extracting specific data
fields from MSOD. One table, VEHICLE, contains descriptive information for vehicles
and tests, and the second table, DYNO, contains emissions measurements. The fields
used and their sources in MSOD are listed in Table B-2. The data aggregation,
classification and analysis was performed by a program written in structured query
language. During this process, the test identifier field RESULTID served as a primary
key linking corresponding information in the two tables.
Table B-2 Data Fields Extracted from the Mobile Source Observation Database (MSOD)
for Use in Development of the MOVES Emission Rate Table
Program Table
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
VEHICLE
DYNO
DYNO
DYNO
DYNO
DYNO
DYNO
Description
Data source: study or program
Unique vehicle identifier
Unique test identifier
Test cycle
Test schedule
Nominal temperature for cycle (°F)
Initial temperature for test (°F)
Vehicle class
Preconditioning for test cycle
Fuel type
Vehicle make
Vehicle model
Vehicle model year
Test weight
Engine displacement (cubic in.)
Engine displacement (L)
System Disablements
Engine Technology
Cumulative time counter within cycle (sec)
Vehicle speed (continuous at 1.0 Hz)
THC emission rate (continuous at 1.0 Hz)
CO emission rate (continuous at 1.0 Hz)
NOx emission rate (continuous at 1.0 Hz)
CO2 emission rate (continuous at 1.0 Hz)
MSOD Table
RESULT
VEHICLE
RESULT
RESULT
RESULT
RESULT
DYNOTEST
VEHICLE
DYNOTEST
M_SOURCE
VEHICLE
VEHICLE
VEHICLE
DYNOTEST
M_SOURCE
M_SOURCE
RESULT
M_SOURCE
DYNOTIME
DYNOTIME
DYNOTIME
DYNOTIME
DYNOTIME
DYNOTIME
MSOD Field
WA_ID
MS_ID
RESULTID
TEST_PROC
SCHEDJD
NOM_TEMP
INIT_TEMP
VEHCLASS
PRECOND
FUELTYPE
MAKE
MODEL_NAME
MODEL_YR
TESTWGHT
DISP_CID
DISP_LITER
DISABLE
FIJTYPE
DYNOSECS
SPEED
R_THC
R_CO
R_NOX
R_CO2
83
-------
B.2.2 Scope of Analysis
The goal of the analysis was to estimate base energy-consumption rates for
running operation. To subset for running operation, two conditions applied. First, if the
test cycle were preceded by a warm-up cycle or another test cycle, all data were retained.
Second, for test cycles not preceded by a warm-up cycle, the first 100 seconds of data
from each test were removed. Cycles assumed to include a cold start included the EPA74,
FLA4, FTP, FTPSS, ST01, and LA92. Finally, if the DISABLE field indicated that air
conditioning was on during the test, corresponding measurements were removed.
B.3 Assignment of Data to MOVES SourceBins
B.3.1 SourceBin Definition for RunningEnergy
Within the MOVES modal framework, motor vehicles will be characterized as
emissions sources on the basis of fuel, engine and vehicle characteristics. To estimate
energy consumption, emissions data is classified on the basis by six parameters: (1) fuel
type, (2) engine technology, (3) regulatory class (4) model-year, (5) engine displacement
and (6) vehicle weight. A specific six-way combination of these characteristics is defined
as a "Source Bin." Within the MOVES database, each source bin is identified by a
specific 20-character numeric code. For purposes of the current analysis, i.e., estimation
of energy consumption, the regulatory-group attribute was not applied. Thus, the source
bins as defined for this analysis effectively represent a five-way classification by the
remaining attributes.
Each of the sourcebin characteristics listed above is defined as a MOVES
attribute. The attributes corresponding to each characteristic are (1) fuelTypelD, (2)
engTechID, (4) modelYearGroupID (shortMdYrGrpID), (5) engSizelD and (6)
weightClassID. Each of the attributes is further defined below.
B.3.1.1 Fuel Type (fuelTypelD)
As the name implies, this characteristic is defined simply in terms of the type of
fuel used in the engine. For example, most common fuel types currently in the database
are "gasoline" or "diesel." The attribute will also account for the introduction of future
fuel types, such as "hydrogen" or "electric" vehicles. Assignment of fuel Type based on
values of the MSOD FUELTYPE field are presented in Table B-3.
B.3.1.2 Engine Technology (engTechID)
With respect to currently available emissions measurements, this attribute is
defined in terms of fuel delivery technology. All currently available fuel delivery
technologies are designated as "conventional." The attribute is also defined to contain
values for anticipated future technologies, as shown in Table B-4. Assignment of values
for current technologies based on the MSOD field "FI_TYPE" is also shown.
84
-------
B.3.1.3
Model Year Group (modelYearGroupID, shortModYrGroupID)
This attribute assigns emissions data to classes based on the model years of
vehicles represented in the emissions database. We assigned model-year group
designations based on predefined model-year ranges, as shown in Table B-5. Note that
the "short" model-year group labels are used in the sourceBin attribute labels, rather than
the "long" values, in order to save space.
B.3.1.4
Engine Size (engSizelD)
We define engine size in terms of displacement, expressed in liters (L). On this
basis, we classified the data into predefined displacement classes, as shown in Table B-6.
B.3.1.5
Vehicle Loaded Weight (weightClassID)
This attribute represents the weight of the vehicle bearing a load during operation.
It is represented by the test weight of the vehicle as recorded for each vehicle test. We
classified the data into a predefined set of weight classes, as shown in Table B-7.
Table B-3. Equivalence between MSOD and MOVES databases for the SourceBin
Attribute "Fuel Type" (fuelTypelD)
MSOD Field "FUELTYPE"
Value
GAS
DIES
CNG
LPG
E85
EOO
M85
MOO
Any other non-null value
NULL
Definition
GASOLINE POWERED
DIESEL POWERED
COMPRESSED NATURAL
GAS
LIQUID PETROLEUM GAS
85%ETHANOL, 15%
GASOLINE
100% ETHANOL
85%METHANOL, 15%
GASOLINE
MOVES Attribute
fuelTypelD
1
2
3
4
5
5
6
6
7
8
9
NULL
NULL
Definition
Gasoline
Diesel
Compressed Natural Gas
Liquid Petroleum Gas
Ethanol(E85orE95)
Ethanol(E85orE95)
Methanol(M85orM95)
Methanol(M85orM95)
Gaseous Hydrogen
Liquid Hydrogen
Electricity
85
-------
Table B-4 Equivalence between MSOD and MOVES databases for the SourceBin
Attribute "Engine Technology" (engTechID)
MSOD Field "FI_TYPE"
Value
PFI
TBI
NOTFI
FICARB
DIRECT
Any other non-null value
NULL
Definition
PORT FUEL INJECTION
THROTTLE-BODY FUEL
INJECTION
NOT FUEL INJECTED
CARBURETED (IS THIS IN
MSOD?)
INTO CYLINDER
INJECTION
MOVES Attribute
EngTechID
1
1
1
1
1
1
NULL
10
11
12
15
16
17
20
25
26
27
Definition
Conventional
Conventional
Conventional
Conventional
Conventional
Conventional
Hybrid Electric IC-A
Hybrid Electric IC-B
Hybrid Electric IC-C
Hybrid Electric DI-A
Hybrid Electric DI-B
Hybrid Electric DI-C
Fuel Cell
Hybrid Electric - Fuel Cell
A
Hybrid Electric - Fuel Cell
B
Hybrid Electric - Fuel Cell
C
86
-------
Table B-5 Definition of the MOVES Model-Year-Group
Attribute (shortModYrGrpID) for Light-duty and Heavy-duty
Motor Vehicles
Model Year
1980 and earlier
1981 - 1985
1986 - 1990
1991-2000
2001-2010
MOVES Attribute1
modelYearGroupID
19601980
19811985
19861990
19912000
20012010
shortModYrGrpID
01
02
03
04
05
1 The short attribute values were used in actual assignment of model-year groups.
The long attribute values (modelYearGroupID) are included here for
completeness.
Table B-6 Definition of the MOVES Engine-Size
Attribute (engSizelD) in terms of Engine
Displacement
Engine Displacement (L)
displacement < 2.0
2.0 < displacement < 2.5
2.5 < displacement < 3.0
3.0 < displacement < 3.5
3.5 < displacement < 4.0
4.0 < displacement < 5.0
5.0 < displacement
NULL
engSizelD
0020
2025
2530
3035
3540
4050
5099
NULL
87
-------
Table B-7 Definition of the MOVES Vehicle Weight-
Class Attribute (weightClassID) for Light-duty and
Heavy-duty Motor Vehicles
Vehicle Weight (Ib)1
weight < 2,000
2,000 < weight < 2,500
2,500 < weight < 3, 000
3, 000 < weight < 3, 500
3, 500 < weight < 4,000
4,000 < weight < 4,500
4,500 < weight < 5,000
5,000 < weight < 6,000
6,000 < weight < 7,000
7,000 < weight < 8,000
8,000 < weight < 9,000
9,000 < weight < 10,000
10,000 < weight < 14,000
14,000 < weight < 16,000
16,000 < weight < 19,500
19,500 < weight < 26,000
26,000 < weight < 33,000
33,000 < weight < 40,000
40,000 < weight < 50,000
50,000 < weight < 60,000
60,000 < weight < 80,000
80,000 < weight < 100,000
100,000 < weight < 130,000
130,000 < weight
NULL
weightClassID
20
25
30
35
40
45
50
60
70
80
90
100
140
160
195
260
330
400
500
600
800
1000
1300
9999
NULL
1 Defined as the equivalent test weight for a given vehicle test, as obtained
from the MSOD field DYNOTEST.TESTWGHT.
88
-------
B.3.1.6 Source Bin Identifier (sourceBinID).
Assignment of the attributes just described allowed assignment of the source-bin
identifier. The identifier is a 19-digit numeric label, of the form
IffiteeyysssswwwwQQ, where each component is defined as follows:
1 is the literal value "1," which serves as a leading value to set the magnitude of the
entire label,
^represents the fueltypelD,
tt represents the engTechID,
ee represents the regClassID (This attribute is not defined for running energy
consumption in MOVES2004. The value 0 is inserted as a place holder),
yy represents the shortModYrGrpID,
ssss represents the engSizelD,
wwww represents the weightClassID, and
00 is the literal value "00," which serves to provide two trailing zeroes at the end of
the label.
The individual attributes are assembled in the proper sequence by constructing the
sourceBinID as a pattern variable, where
sourceBinID = IxlO18 +
fuelTypeIDx!016 +
engTechID xlO14 +
emisTechlDxlO12 +
shortModYrGrpID x 1010 +
engSizelDxlO6 +
weightClassID xlO2 (B-l)
As an example, Table B-8 shows the construction of the sourceBin label
(1010100040020002500) for light-duty gasoline vehicles, with displacement less than 2.0
L, weighing between 2,000 and 2,500 Ib, and manufactured in 1991 and later.
89
-------
Table B-8 Definitions of Source Bin Attributes for sourceBinID =
1010100032025003000
Description
Leading integer
Fuel
Engine Technology
Regulatory Class
Model year Group
Displacement Class
Weight Class
Trailing zeroes
Attribute
—
fuelTypelD
engTechID
regClassID
shortMdYrGrpID
engSizelD
weightClassID
—
Form
—
ff
tt
ee
yy
ssss
wwww
...
Value
1
01
01
00
03
2025
0030
00
Definition
-
Gasoline
Conventional
—
1986-1990
2.0- 2.5 L
2,500 < weight < 3, 000
—
90
-------
B.4 Operating Modes
Within source bins, we further sub-classified data on the basis of "operating
mode," designated as the MOVES attribute "opModelD." For motor vehicles, operating
mode is defined in terms of seventeen classes defined in terms of vehicle-specific power
(VSP), vehicle speed and vehicle acceleration. The derivation of the operating mode
classes is discussed in Appendix A.
B.4.1 Calculation of Vehicle-Specific Power (VSP)
The first step in assigning operating mode is to calculate vehicle-specific power
(VSP) for each emissions measurement. At a given time t, the instantaneous VSPf
(kW/tonne, at a frequency of 1.0 Hz) represents the vehicle's tractive power /Vac,f
normalized to its weight miorme. The VSP parameter is expressed as a third-order
polynomial in speed, with additional terms describing acceleration and road-grade
effects. The coefficients for this expression are defined in terms of rolling-resistance and
aerodynamic drag coefficients, as
V+ atvt + gvt sin 0t
mt,
tonne
(B-2)
where
_ zero_or(jer tire rolling-resistance coefficient (unitless),
= first_or(jer tjre rolling-resistance coefficient (sec/m),
_ secon(j.or(jer tire rolling-resistance coefficient (sec2/m2),
Cd = aerodynamic drag coefficient of the vehicle (unitless),
R = cross-sectional frontal area of the vehicle (m2),
pair =density of air ( 1.202 kg/m3),
vf = vehicle speed at time t (m/sec),
at = vehicle acceleration at time t (m/sec2),
Wtonne = vehicle weight (metric tonne),
g = acceleration due to gravity (9.8 m/sec2),
9t = road grade (radians)
In practice, a simplified form of the equation is substituted for the theoretical form
presented above. In the simplified expression, the polynomial coefficients for vehicle
speed are expressed as track road-load coefficients (A, B, C). In addition, because vehicle
speed and weight are provided to the calculation in english units (mph and Ib),
appropriate conversion factors are inserted in each term. Finally, because we relied on
91
-------
laboratory data to calculate emission rates, road grade = 0 and the corresponding term
drops out. The revised expression is
VSp,=
(B-3)
where
A = the rolling resistance coefficient (rollingTermA, kW-sec/m),
B = the rotational resistance coefficient (rotatingTermB, kW-sec2/m2),
C = the aerodynamic drag coefficient (dragTermC, kW-sec3/m3),
m = vehicle weight (Ib)
vf = instantaneous vehicle velocity at time t (1.0 Hz, mi/hr),
at = instantaneous vehicle acceleration (1.0 Hz, mi/hr-sec),
c\ = a conversion factor for speed (0.44704 m-hr/mi-sec),
GI = conversion factor for vehicle weight (0.4536 kg/lb)(0.001 tonne/kg),
B.4.2 Track Road-Load Coefficients
We estimated values for the track road-load coefficients separately for light-duty
and heavy-duty vehicles, as described below. As a first step, we assigned data to "light-
duty" and "heavy-duty" duty classes, on the basis of vehicle-class and vehicle weight.
We defined vehicle weight as the vehicle's test weight, as recorded for a given vehicle or
set of measurements (See Table B-9).
Table B-9 Definition of Vehicle Duty Class in terms of Vehicle Class and
Weight
MSOD Field "VEHCLASS"
Value
CAR
LDV
LOT
LDT1
LDT2
TRUCK
TRUCK
BUS
HDDV7
HDDV8
Definition
Car
Light-duty Vehicle
Light-duty Truck
Class 1 Light-duty Truck
Class 2 Light-duty Truck
Truck
Truck
Bus
Class 7 Heavy-duty Vehicle
Class 8 Heavy-duty Vehicle
Weight (Ib)1
All
All
All
All
All
< 8,500
> 8,500
All
All
All
Duty Class
LIGHT
LIGHT
LIGHT
LIGHT
LIGHT
LIGHT
HEAVY
HEAVY
HEAVY
HEAVY
1 Defined as test weight for a given vehicle test.
92
-------
B.4.2.1 Track Road-Load Coefficients: Light-Duty Vehicles
For light-duty vehicles, we calculated the track load coefficients from an
additional parameter, the "track road load horsepower at 50 mph" (TRLHP), based on
equations B-4a - c).
A = pF /TRLHP.^
TRLHP.c,
c=pp , TRLHP.
(B-4a, b, c)
Where:
PF.4 = default power fraction for coefficient^ at 50 mi/hr (0.35),
PF# = default power fraction for coefficient B at 50 mi/hr (0.10),
PFc = default power fraction for coefficient C at 50 mi/hr (0.55),
ci = a constant, converting TRLHP from hp to kW (0.74570 kW/hp),
v50 = a constant vehicle velocity (50 mi/hr),
C2 = a constant, converting mi/hr to m/sec (0.447 m-hr/mi-sec)).
In the process of performing these calculations, we converted from english to metric
units, in order to obtain values of the track road-load coefficients in SI units, as listed
above.
Values of TRLHP for most domestic and imported vehicles manufactured
between 1973 and 2000 can be obtained from the "I/M Lookup Table," a dataset
developed for USEPA for use in Inspection and Maintenance programs.29 Parameters
contained in the file include TRLHP, test-weight, displacement, and body style. To
estimate values of TRLHP for specific vehicles represented in the emissions database, we
used a simple regression model to relate TRLHP to testweight and vehicle class.
Graphical presentation of the data shows the expected strong relationship between
TRLHP and test weight (See Figure B-l). The plot also suggests the value of vehicle
class as a categorical predictor. However, The I/M Lookup table does not contain vehicle
class as classified by EPA, however, the "body style" can be readily translated to vehicle
class, as shown in Table B-10, along with frequencies of measurements in the dataset,
cross-tabulated by body style and vehicle class. In the Lookup Table, each record
represents a specific combination of division, model and model year (e.g., Toyota,
Corolla, 1973).
93
-------
Figure B-l. Track road-load horsepower (TRLHP) for light-duty vehicle models
over the period 1973-1999, versus test weight. Source: I/M Lookup Table, version
1.8.5 (Sierra Research 2000).
40 -J
30-
S 2CH
I
i
10:
H-
4-
++ I!
4
4
-i-
1000 2000 3000 4300 5000 6000 7000 8000
Equivalent Test Weight
Vehicle Class + + + Idt + + + Idv 444
non
Table B-10. Frequencies of Records in the I/M Lookup Table, by Body
Style and Vehicle Class
Body Style
Sedan
Wagon
Pickup
Sport-utility Vehicle
Mini-Van
Full-size Van
Total
Light-duty Vehicles (LDV)
8,209
1,699
9,908
Light-duty Trucks (LDT)
3,204
1,954
463
2,087
7,708
To estimate TRLHP in terms of test weight and vehicle class, we developed a
general linear model designed to fit different intercepts and slopes for LDVs and LDTs.
The model is in the form
94
-------
(B-5)
where
y
/?o
/?i
02
TRLHP for a given combination of division, vehicle model and model year,
= vehicle test weight (Ib),
= an indicator variable for vehicle class:
= 1 for light-duty vehicles (LDV),
= 0 for light-duty trucks (LOT),
= a y-intercept term,
= a coefficient for the vehicle-class indicator, representing the difference in the
intercept between LDV and LDT,
= a slope coefficient for vehicle test weight,
= slope coefficient for an interaction term between test weight and vehicle class,
representing the difference between slopes between LDV and LDT, and
residual error between the mean estimated value of TRLHP, and the specific
value for a given division/model combination, for a given value of test weight.
Use of the indicator variable for vehicle class and the interaction term allowed us to
fit separate intercepts and slopes for LDV and LDT, within a single model. Table B-l 1
presents coefficients and goodness-of-fit statistics.
Table B-ll Coefficients and Goodness-of-Fit Statistics for the Regression Model
Relating Track Road-Load Horsepower (TRLHP) for Light-duty Vehicles to Test
Weight and Vehicle Class (R2= 0.8064)
Coefficient
Intercept (/?0)
Test weightC/?,)
Vehicle Class ($,)
Test weight x Vehicle Class (/%)
Mean
5.978016174
0.003165941
-1.617898959
-0.000390014
Standard Error
0.11881559
0.00002766
0.14343255
0.00003574
^-statistic
50.31
114.47
-11.28
-10.91
/7-value
(Pr>U|)
< 0.0001
< 0.0001
< 0.0001
< 0.0001
B.4.2.2
Track Road-Load Coefficients: Heavy-Duty Vehicles
No public data source is available that provides track road-load parameters for
specific truck models over a series of model years. However, a source for these
coefficients has been published which gives estimates of the road load coefficients for
general vehicle classes, e.g., cars, short-haul delivery trucks, tractor-trailer vehicles.30
Drawing on this work, we have elected to estimate general values for these coefficients in
terms of the vehicles' rolling resistance and aerodynamic drag coefficients. From the
physical expression of VSP (Equation B-2), the relationships defining the road-load
coefficients A, B and C for heavy-duty vehicles in terms of vehicle weight (m), and the
rolling resistance and aerodynamic terms are
95
-------
A = -
= 0.0
c=
(B-6a,b,c)
where
c\ = conversion factor for vehicle weight, from Ib to metric tonnes (2,204.6 Ib/tonne)
Because the vehicle weight (mass) is input to the calculation in pounds, an
additional conversion factor c\ from pounds to metric tonnes is needed. The final form of
the above relations between the A, B, and C road load coefficients are as displayed in
Table B-12 below.
Table B-12. Road-Load Coefficients for Heavy-Duty Trucks and Buses
Coefficient1
A
B
C
Heavy-duty Vehicle Weight Class, Trucks2
8,500-14,000 Ib
(3. 855-6. 350 tonne)
0.0996/wcj
0.0
mr 1 5 ^ "10~5
\ 1,000 wq )
14,000-33,000 Ib
(6.350-14. 968 tonne)
0.0875/wcj
0.0
— , 1 i c n,,in-5 1
(iflOOm^ )
> 33,000 Ib
(> 14. 968 tonne)
0.066 Imci
0.0
m- \ 2'89 1 1 ^"lO"5 1
(iflOOm^ )
Buses
0.0643/wcj
0.0
,„„„ ' i c n^.- 1 n~5
\l,000mcl )
1 As in Eq. B-3 above, units for coefficients /4, B and C are kW-sec/m, kW-sec2/m2 and kW-sec3/m3, respectively.
2 As in Eq. B-3 above, m is vehicle weight (Ib) and c\ is a conversion factor between Ib and metric tons (0.0004536 tonne/lb). For C, 1,000^
is a conversion factor between Ib and kg (0.4536 kg/lb).
The heavy-duty truck coefficients derived using this approach were compared to
an analysis of speed-time traces of coast downs for eight heavy-duty tractor-trailer
vehicles.31 In this analysis, vehicle speed during the coast down (external forces/vehicle
mass) was parameterized as a function of time and then fit to corresponding measured
values of vehicle speed determined from the vehicle speed-time traces. More explicitly,
the vehicle speed was assumed to have a second order dependence on time
(B-7)
96
-------
Relating speed to time also allows forces on the vehicle during coast-down to be
parametrized in time, as
F = m— = m(b1+2b2t)
dt (B-8)
This equation can then be equated to the road-load equation, by normalizing force to
vehicle weight
m v
which allows development of relationships between the fit parameters, bo, b\, and b2, and
the road load coefficents, A, B, and C. Finally, the physical road load parameters can be
determined using the relationships in Equations B-6a-c.
To address the fact that these general relationships are under-constrained, we
applied additional physical constraints. For example, we ensured that acceleration values
were less than zero, loss of vehicle kinetic energy was within 5% of the actual vehicle
kinetic energy loss, and that vehicle aerodynamic drag and rolling resistance coefficients
were within reasonable ranges.
These preliminary analyses gave confidence that the values derived from
Petrushov (1997) are reasonable initial estimates of road-load coefficients for a large
variety of heavy-duty truck shapes. However, at least two considerations may merit
further consideration. First, vehicle sizes may differ between the Russian and U.S fleets.
European vehicles may tend to have smaller frontal areas than those in the U.S., resulting
in smaller aerodynamic drag terms. Secondly, Petrushov's tire rolling resistance factor
does not include a second-order term (in speed) which leads to the road load coefficient B
being equated to zero. Although some studies express the rolling resistance as the sum of
a constant and a quadratic term in speed, most literature sources report non-zero values
for B32 In addition, the method used to derive track road-load coefficients for light-duty
vehicles in this analysis generates non-zero values for B.
B.4.3 Calculation of Vehicle Acceleration
In addition to vehicle velocity at the 1.0 Hz measurement frequency, the VSP
calculation requires estimation of acceleration at the same frequency. Thus, for each
measurement at time t, we calculated acceleration at as the first differential of vehicle
velocity, or
at = vt —v
vt-i
(B-10)
97
-------
where vt and vt.\ are the vehicle velocity for the current and previous measurements,
respectively. In addition, we calculated the acceleration for the two seconds prior to the
current measurement (at.\ and at.^) as
at-\ = Vt-\ ~ Vt-2
These latter two acceleration values are not used in calculation of VSP, but rather in
defining the "deceleration" operating mode, as described below.
B.4.4 Assignment of Operating Mode (opModelD)
After calculating VSPf for each measurement, and associated values of velocity
and acceleration, we assigned operating mode on the basis of these parameters. As the
name implies, opModelD is a modal variable, with 17 specific modes defined as shown
in Table B-13. Only the "deceleration" mode is defined in terms of acceleration, with all
others defined in terms of VSP and velocity ranges. In quantitative terms, "deceleration"
occurs when acceleration was less than or equal to -2.0 mph/sec during the current
second, or less than -1.0 mph/sec for three consecutive seconds, including the current
second and the two previous seconds.
98
-------
Table B-13 Definition of the MOVES Operating Mode Attribute for Motor
Vehicles (opModelD)
OpModelD
0
1
11
12
13
14
15
16
21
22
23
24
25
26
33
35
36
Operating Mode
Description
Deceleration/Braking
Idle
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Vehicle-Specific
Power
(VSPt, kW/tonne)
VSP,< 0
0
-------
MC02 = molecular weight of CO2 (44.0098 g/mol),
MCO = molecular weight of CO (28.0104 g/mol),
Me = molecular weight of carbon (12.0110 g/mol),
MH = molecular weight of hydrogen (1.008 g/mol),
y = H:C ratio for fuel,
HV = heating value for fuel (kJ/g)
Table B-14 lists assumed values of H:C ratio and heating value used for each fuel.
Table B-14. Hydrogen:Carbon Ratios and Heating Values for Selected Fuels
Fuel Type
Gasoline
Diesel
Natural Gas1
Liquid Petroleum Gas (LPG)
Ethanol Blends2
Methanol Blends3
H:C Ratio (y)
1.85
1.80
3.80
3.80
1.851
1.851
Heating Value (kJ/g)
44.00
43.20
45.02
46.40
26.90
20.00
1 Includes compressed and liquid natural gas.
Includes ethanol blends "E85" and "EOO."
3 Includes methanol blends "M85" and "MOO."
Note that the term Mc + yMK represents the mass of fuel equivalent to 1.0 mol of
emissions (CC>2 + CO), i.e., units for this expression are g fuel/mol CO2+CO. If the fuel is
assumed to contain only carbon and hydrogen, and if its nonstructural formula is
represented as C«Hfe, then the fuel's molecular weightMfuei can be expressed as
aMc+bMft. Because the fuel's H:C ratio is defined as b/a, the number of hydrogens b is
equal to ay, andMfuei can also be written as aMc+ayMH, or a(Mc+yMH). Further,
assuming complete combustion, approximately a mols of CO2+CO are produced by one
mol of fuel, represented as
1 mol fuel -» a mol (CO 2 + CO)
(B-13)
Dividing both sides by a, we see that
1
- mol fuel -> 1 mol (CO 2 + CO)
a
(B-14)
If we convert the fuel to a mass basis, we see that
i
- mol fuel a(Mc +yMR ) -» 1 mol (CO2 + CO)
a X mol fuel )
(B-15)
100
-------
and finally, that
Mc +yMH gfuel -> lmol(CO2 +CO) (
Thus, we find that the molecular weight of the fuel normalized to moles of CC>2 serves as
a conversion factor between moles of emissions and mass of fuel. Additionally, the
molecular weight of HC emissions is assumed to equal that for the fuel, which allows the
HC emission rate to be simply added to the term describing the fuel equivalent of CC>2
and CO.
B.6 Table Generation and Summary Statistics
After the steps described above, we were ready to calculate numbers to populate
the EMR Table. After classifying the emissions data by source bin, we further
subclassified it by operating Mode.
We calculated means and other summary statistics for each combination of
sourceBinID opModelD. For simplicity, we will refer to a specific combination of
sourceBinID and opModelD as a "cell." In estimating variances for cell means, we
treated the data within cells as effective cluster samples, rather than simple random
samples. This approach reflects the structure of the data, which is composed of sets of
multiple measurements collected on individual vehicles. Thus, measurements on a
specific vehicle are less independent of other measurements on the same vehicle than of
measurements on other vehicles. Accordingly, means and variances for individual vehicle
tests were calculated to allow derivation of between-test and within-test variance
components. These components were used in turn to calculate the variance of the mean
for each cell, using the appropriate degrees of freedom to reflect between-vehicle
variability.33 To enable estimation of variances under this approach, we calculated a set
of summary statistics, as listed below:
Cell mean (Jcell, meanBaseRate): the arithmetic mean of all measurements in the cell.
Test mean (yt): the arithmetic mean of all measurements in a given test on a specific
vehicle.
Cell sample size («Ceii, nCell), the number of individual measurements in a cell, where
each count represents a measurement collected at a frequency of 1.0 Hz, (i.e., "second-
by-second).
Test sample size (wtest, nTest), the number of individual vehicle tests represented in a cell.
Measurement sample size («meas, nMeas): the number of measurements in a cell
representing an individual test on an individual vehicle.
101
-------
Test variance (s2esti , varTest): the variance of measurements for each vehicle test
represented in a cell, calculated as the average squared deviation of measurements for a
test about the mean for that test. Thus, we calculated a separate value of varTest for each
test in each cell.
Between-Test variance component ( s^etw , varBetw): the component of total variance due
to variability among tests, or stated differently, the variance of the test means about the
overall cell mean, to be calculated as
s
betw
"test 1 (B.17)
Within-Test Variance Component (s^ith, varWith): the variance component due to
variability within tests, or the variance of measurements within individual tests about
their respective test means, which can be expressed in terms of the sums of squares for
each test:
"cell "test (B-18)
whereby is an individual measurement /' in testy. In practical terms, however, it is simpler
to calculate varWith in terms of the test variances, summed over all tests in the cell:
X 17 c<
/ j meas,z te^,z
"cell ""test (B-19)
Note that the sum of squares for test /' is calculated from the test variance by multiplying
by the sample size wmeas,;, rather than the degrees of freedom nmeas,rl. This form is
unusual with sample data, but applies in this case because the STDEV function in
mySQL calculates a population standard deviation, rather than a sample standard
deviation.
Variance of the cell mean (s-, varMeanCell): this parameter represents the uncertainty in
the cell mean, and is calculated as the sum of the between- vehicle and within-test
variance components, with each divided by the appropriate degrees of freedom.
2 2
C2 _ 'S'betw , ^with
~
"test "cell (B-20)
B.6.1 Cell Evaluation Criteria
Based on the summary statistics, we evaluated two criteria for each sourceBinID
x opModelD cell. These criteria determined whether the data in the cell would be
considered adequate to populate the cell.
102
-------
Minimum Number of Tests per Cell: To be considered as "populated," we required that a
minimum of three individual vehicle tests be represented in the cell («test ^ 3). This
number is the bare minimum to calculate variance components for a cell and the statistics
derived from them.
Coefficient of Variation of the Mean (CV-, CVmean): this parameter gives a relative
measure of the uncertainty in the cell mean, allowing comparisons among cells. It is
calculated as the ratio of the cell standard error to the associated cell mean
y°a (B-21)
For a cell to be considered "populated" we require that its standard error be no larger than
one-half of its mean, or CV- < 0.50 . Note that the term "CV of the mean" is synonymous
with the commonly used term "relative standard error" (RSE).
B.6.2 Results: Data-Populated Cells in the EmissionRate Table
The product of the data classification and calculations described is the
emissionRate Table in the MOVES database. The emissionRate Table contains five
fields, as described in Table B-l. Each record in the table represents a
sourceBinxopModelD cell, as previously described. For the present data compilation, the
table includes results describing 2,848 cells in 154 to 171 source bins, depending on
operating mode.
103
-------
Appendix C: Proof-Of-Concept Assessment of Hole
Filling Methods
In order to develop high-quality and dependable emissions inventory models, the
MOVES model is designed to be primarily data-driven. It is assumed that as more
emissions data from vehicles are collected, the emissions rates in the model can be
updated and thus the fidelity should correspondingly improve. However, due to the
limited existing data, there will always be data "holes" in MOVES. This appendix
presents three proof-of-concept methodologies for filling these data holes for energy rate
estimations in MOVES.
Preliminary analysis suggested that only about 50 percent of source bin and
operating mode combination (termed "cells") actually are filled with data (this is based
on the raw number cells, not weighted for prevalence in the fleet as presented in Table 4-
6). It should be noted that while all of the cells were selected because they had reflected
some portion of the existing US fleet, many of these cells make up only a very small
minority of the fleet mix (e.g. 1990's compact car with diesel engine). Many of the
emission rates in these empty cells may be filled by interpolating or extrapolating from
nearby cells, where a physical relationship can be drawn. Others cells may be filled
through disaggregating emissions bag data into their respective operating mode bins.
Often bag data is present when second-by-second data is lacking. A third method of
filling the operating mode bins (for a particular source bin) is through the use of a
physical model such as PERE (Section 4.3.3.1). This appendix describes a proof-of-
concept study that compares these three separate methods of filling empty bins. The study
was performed only for light-duty cars (not trucks) since this was the extent that the data
binner was prepared at the time of the study. As of the writing of this report, data quality
checks on the MSOD were still being performed, the results of which will affect some of
the emissions rates obtained. As a result this study does not present actual energy
consumption rates as MOVES would use. The results are all at the proof-of-concept
level.
Each of the methods listed above has its advantages and disadvantages. In some
cases, all of the methods may be employed. The first section of this paper compares the
methods for the cases when they are all applicable. In general PERE can be employed in
almost all cases, though some additional developmental work may be required. However,
only a fraction of the empty cells have neighboring cells populated, from which its own
emission rate can be interpolated or extrapolated. Therefore, in many cases, the
methodology will be guided by whether there is data available in nearby cells. This
document will first identify the cells, which require filling. Then it will provide
recommendations on which process is optimal for integration into MOVES by comparing
the three methodologies.
104
-------
C.I Identifying Empty Bins
The following series of figures show an example of empty bins pertaining to
light-duty vehicles, based on preliminary runs of the binner program described in
Appendix B. Group 1 (model year pre-1980) is shown in Figure C-l. Vehicle weight is
across the top, and can be determined by essentially multiplying the bin number by 100
Ibs (e.g. bin 40 is 3500-4000 Ibs). Engine displacement is down the left, where the bin ID
defines the range of engine sizes. It is read in the following way: e.g. 4050 = 4.0 to 5.0
Liter. Green cells are cells with second-by-second data, orange cells only have bag data,
and red cells have no data. Figures C-2 through C-5 shows the same chart as Figure C-l
for the other model years.
Figure C-l. Model year group 1 (pre 1980). The (x,y) = (35, 3540) bin is the interpolated
bin.
105
-------
model year
group 2, op
mode id = 0
0
20
2025
2530
3035
.3540
4050
5099
engine
displ id
0
20
25
30
35
40
45
5D
BO
70
SO
90
100
sbs data
bag data
vehicle
weight id
Figure C-2. Model year group 2 (1981-1985). The (x,y) = (40, 20) bin is the extrapolated
bin.
model year
group 3; op
mode id = 0
25
45
50
BO
70
vehicle
weight id
20
2025
sbs data
2530
bag data
4050
5099
engine
displ id
Figure C-3. Model year group 3 (1986-1990)
106
-------
model year
group 4; op
mode id = 0
0
20
25
35
40
45
50
70
90 100
weight id
TI25
1530
bag data
31335
-540
11350
51399
engine
displ id
Figure C-4. Model year group 4 (1991-2000)
C.2 Hole Filling Method 1: Interpolation/Extrapolation
This is the simplest of the three methods in principle and will be described first.
The process of interpolation and extrapolation involves statistically estimating the value
of a particular point based on its relationship to other known values of the same function.
If the unknown value has neighboring known values that are both larger and smaller, one
can 'interpolate' between the known points to estimate the unknown. If the unknown
value extends outside the range of known values (either smaller or larger) of the function,
then one can 'extrapolate' the known points to estimate the unknown. Statistically, the
tools required are identical for these estimations. However, interpolated values have
much more confidence than those that are extrapolated. This is due to various unknown
physical phenomena, which may occur outside the known range of values. For example,
some vehicles on the "edges" may have a higher proportion of turbo driven engines,
which could change the basic behavior of energy consumption in relation to the vehicles
in the neighboring bins.
All of the interpolations and extrapolations conducted in this section will be within
specific vehicle source bin "families". These families are identical in all respects except
for 1 "trait". This trait defines a physical characteristic of the family. Two examples of
traits are engine size and vehicle weight. It is logical to assume that there is a relationship
between fuel consumption (energy) and engine size, or vehicle weight. As a rule-of-
thumb, the bigger the engine, the more fuel is burned in order to follow the same driving
trace. Also the heavier the car, the more work is required to move it. It is upon these
simple physical principles that our justification for this methodology lies.
The previous section described all of the empty and filled cells for cars (it does not
include LDT). We select the following source bin families in this study: Model year
group 1 (pre-1980) for the interpolation exercise and group 2 (1981-1985) for the
107
-------
extrapolation case. The interpolated bin in question is the 3540 (3.5-4.0L) cross-
referenced with the 35 (3000-3500 Ib) weight bin (Figure C-l). The extrapolated bin is
the 20 (<2.0L) cross-referenced with the 40 (3500-4000 Ib) weight bin (Figure C-2).
C.2.2 Interpolation
The interpolated emissions rates for each operating mode bin are determined
using a linear regression using MS Excel. Each regression is determined from 4 known
values of engine displacement bin. Figure C-5 shows the results. The vertical red line
shows all the interpolated values determined from the regressions. It is interesting that
there is a dip in the emissions rates measured in the 3.0-3.5L bin for all of the operating
mode bins. This unphysical result, is likely due to a sampling bias. There was only a
single vehicle in this bin.
200 T
180 --
160 --
140 --
120 --
Linear Fit of Four Engine Size Bins to
Determine Base Rate for Engine Size Bin 3540
(Model Year Group 1; model years < 1980)
3 3.5
engine size id mid-point
4.5
Figure C-5. Energy rates as a function of engine size for each operating mode bin.
The red vertical line is the interpolated bin.
Figure C-6 shows the same results by operating mode bin. Note that there is little
discrepancy between engine size bins. This will be discussed more later. The next section
will address the extrapolation case.
108
-------
200
180
160
140
120
100
60
20
-braking, idle, and speed bin 1, engine size 2025
-speed bin 2, engine size 2025
-speed bin 3, engine size 2025
-braking, idle, and speed bin 1, engine size 2530
-speed bin 2, engine size 2530
-speed bin 3, engine size 2530
-braking, idle, and speed bin 1, engine size 3035
-speed bin 2, engine size 3035
-speed bin 3, engine size 3035
•braking, idle, and speed bin 1, engine size 3540
•speed bin 2, engine size 3540
•speed bin 3, engine size 3540
-braking, idle, and speed bin 1, engine size 4050
-speed bin 2, engine size 4050
-speed bin 3, engine size 4050
Linear Fit of Four Engine Size Bins to
Determine Base Rate for Engine Size Bin 3540
(Model Year Group 1; model years < 1980)
10
15 20 25
op mode id
30
35
40
Figure C-6. Energy rates as a function of operating mode bin. The red lines are the
interpolated bins.
C.2.3 Extrapolation
The cell to be extrapolated is engine size bin 20 (<2.0L) and weight bin 40 (3500-
4000 Ibs). The cell has 4 filled (known) weight bins that are smaller, and 6 filled engine
size bins that are larger. It is interesting to compare the extrapolations from each of these
two cases. Table C-l shows the extrapolation results for each of the cases. The error bars
are determined from the uncertainty in the linear regressions. The differences between the
two cases are mainly pronounced in the higher VSP valued bins (16 and 26). All others
are within 10%. In all cases however, they are consistent within the estimated uncertainty
bounds.
109
-------
Table C-l. Extrapolated energy rates in each operating mode bin from engine
displacement and vehicle weight.
op rrom
mode eng. from %
bin displ error weight error difference
0
1
11
12
13
14
15
16
21
22
23
24
25
26
33
35
36
13.63
12.16
21.43
29.83
42.04
52.00
66.19
84.72
30.86
33.95
45.19
60.54
78.72
117.21
59.47
73.25
93.85
2.97
2.39
0.77
1.72
3.13
4.66
4.30
5.66
3.35
1.24
1.46
3.50
2.72
9.51
4.44
5.28
28.61
14.79
13.83
22.63
30.84
41.03
51.58
63.90
71.61
31.24
35.22
43.76
59.62
81.02
102.08
65.62
78.14
87.28
7.38
4.36
7.82
14.19
18.34
21.41
18.53
25.61
18.15
19.89
25.23
23.77
25.58
29.76
40.57
36.83
73.07
7.86
12.11
5.29
3.26
-2.47
-0.81
-3.58
-18.31
1.22
3.58
-3.28
-1.55
2.84
-14.83
9.38
6.26
-7.53
Figure C-7 shows the extrapolation for engine size (as in the previous case).
Figure C-8 shows the extrapolation for vehicle weight. In each case, the red line represent
the extrapolated values. Note in comparing these two figures that the emission rates from
the various engine displacement bins are very similar, with the possible exception of the
high speed bins (33, 35, and 36). This is not the case in Figure C-8, when viewed by
vehicle weight. This may lead us to the conclusion that the separation of the source bins
by engine size may not be necessary, or that fewer engine size bins can capture the
variability. "Collapsing" the engine size bins has the advantage that there will necessarily
be fewer bins, thus allowing the model to run more efficiently. As a consequence, there
will also be fewer empty bins to fill. This also implies that extrapolation by vehicle
weight is more reliable than that by engine displacement.
Figure C-9, demonstrates that the base rates for engine size and engine weight
follow a similar behavior, as we would expect.
110
-------
130
120
110
100
90
80
70
60
50
40
30
20
10
0
l. 1.25L, idle, braking, bin
. 25L.bin 2
^^disp!1.25L, bin3
O displ 2.25L, idle, braking, bin 1
-e-displ2.25L.bin 2
-e-disp!2.25L, bin3
-X-displ 2.75L, idle, braking, bin 1
-H-displ 2. 75L.bin 2
-H-displ2.75L, binS
A displ 3.25L, idle, braking, bin 1
-A-displ3.25L, bin 2
-A-displ3.25L, bin 3
Model Years 1981
Weight Class
hrough 1985,
id = 40
10
15
30
20 25
op mode id
Figure C-7. Extrapolated values from engine displacement.
35
40
weight id 20,
weight id 20,
weight id 20,
weight id 25,
weight id 25,
weight id 25,
weight id 30,
weight id 30,
weight id 30,
weight id 35,
weight id 35,
weight id 35,
weight id 40,
weight id 40,
weight id 40,
idle, braking, bin 1
bin 2
bin3
idle, braling, bin 1
bin 2
bin3
idle, braking, bin 1
bin 2
bin3
idle, braking, bin 1
bin 2
binS
idle, braking, bin 1
bin 2
bin3
15 20 25
op mode id
Figure C-8. Extrapolated values from vehicle weight.
Ill
-------
130 -,
120 -
110 E
|100-
i 90 -
1 8°-
.1 70 -
3
V 60 -
"ro
CD 50 -
(0
ro
^ 40-
ro
| 30-
20 -
10 -
n
i Comparison of Linear Fits of Engine Size Bins and
i Linear Fits of Vehicle Weight Groups to
1 Determine Base Rate for
\r\gme Size Bin 20, Model Year Group 2, Ve
&s
$r
is*
H
^
^
^
^
s
\ — 1
•?
XX
>XH
licle Weic
h
"54
*5P
V<
\—
*
^
— -H
^
^
^
htid
i— <
> — ^
s^
=40
>>
i
/
s
'/,
^ •
<
sl^
^
X^9559x
2 = 0.9634
X
••
0
10
20
30
40 50 60 70 80 90 100 110 120 130
mean base rate (kJ) - engine size
Figure C-9. A comparison of base rates for engine size and vehicle weight.
C.3 Hole Filling Method 2: Deriving Modal Rates from Bag Data
In this section, our goal is to estimate the energy consumption rate (henceforth
called energy rate) within a particular operating mode bin given only the bag
measurements and the VSP distribution within the driving cycle. Due to the
simplifications required, we assume linear relationships only. This analysis draws on
previous work performed by NC State University as part of the proof-of-concept modal
binning analysis referenced in Appendix A.
In symbolic terms, we want to estimate the average energy rate in a bin: Et, where
/' is the operating mode bin number (/' = 0, 1,21, 22, ... 36). We know the total (or bag)
energy consumption orEbag, which is determined from CO2 bag emissions measurement.
For each driving cycle and vehicle, the VSP distributions are also known. Dimensionally,
the product of VSP and mass is energy:
Ebin =
where t is summed over the N individual second by second VSP values in a particular bin
/'. This is a total road-load energy, and not energy at the engine. Dimensionally, the total
road-load energy over the driving cycle would then be
112
-------
Ecycle = ^ Ebini
Assume that the energy at the engine is proportional to the road-load energy. This energy
would be the energy consumed in the bag measurement, thus
Ebag = k» Ecycle
where k is a scaling constant. The total bag energy can then be proportioned amongst the
VSP determined representative energy bins:
Et = k»(Ebin1)
where
N
Therefore, the average value of representative energy in a bin is
Ecycle
Ebag
1
Ecycle N
In this analysis, the bin being filled contains 5 FTP and 14 EVI240 tests, for which only
bag data is available. For the FTP tests, only bags 2 and 3 were used in order to minimize
any potential cold start effects. In all, this source bin includes 17 distinct vehicles. The
mass and TRLHP are determined for each vehicle from I/M lookup tables. 34 And then
combined with the driving trace to determine VSP or energy distributions, i.e., the
These energy distributions for both the IM240 and the FTP cycles were then binned
according to the operating mode numbering scheme from which total cycle energies,
Ecycle = / Ebint
and average bin energies
113
-------
N
for the two individual cycles were determined. (Note: Because there are energy losses
such as engine friction and braking that are not included in the VSP equation, many VSP
values are negative. All negative values of energy are set equal to zero in the averages.
This will be discussed below.)
Bag data for these cycles included total NOx, total hydrocabon, CO, and CC>2
measurements in grams/mile and total mileage. These were then used to determine the
total fuel consumed and the total energy used via carbon balance and an assumed lower
heating value (44MJ/kg). Finally using the bag energy, Ebag, and the cycle energy,
Ecycle, a scaling factor could be determined which was used to scale the average bin
values to average bag values for a particular bin.
Figure C-10 shows the energy rates determined from FTP and IM bag results. The
values are considerably different from the rates determined from vehicle weight
extrapolation (previous section). However, this may be an unfair comparison because the
vehicles in the extrapolated bin are typical of vehicles in the bin, whereas the bag data are
generated from a number of specific vehicles. Thus the average engine size of the bag
sample may be quite different from the typical vehicle extrapolated from its neighboring
bins. Moreover, the discrepancy at the low VSP valued bins (11, 12, 21, 22, 33, etc) is
likely due to the fact that the relationship between road-load and engine load (equation #)
is assumed to be linear and constrained to go through the origin. It has been demonstrated
from (gamma study reference) that (mechanical) engine friction offsets this relationship,
especially at low loads. Accounting for this low speed effect could compensate for this
discrepancy. Thus a new method could be proposed, one that fills bins based on bag data,
but using PERE. By contrast, the following section uses PERE to fill second-by-second
data, which is subsequently binned.
Figure C-l 1 compares the scaled energy rates, with that determined from
extrapolation. Again, note that all of the scaled rates are low at lower energies, but
correspond on average at higher energies.
114
-------
3
.2
ro
cu
ro
.0
130 -,
120 -
110 -
100 -
90 -
80 -
70 -
60
50
40
-^—displ. 1.25L, idle, braking,
binl
•^—displ 1.25L.bin 2
•^—displ 1.25L, bin3
-X—scaled ftp, idle, braking, bin
1
-X—scaled ftp, bin 2
-X—scaled ftp, bin 3
-® -scaled im240, idle, braking,
binl
-O -scaled im240, bin 2
-O- -scaled im240, bin 3
/tede
Ylears 198
through 19
ight Cla^s id = 40
I
I I
10
15
20
op mode id
30
40
Figure
black
130 T
120
110
| 100
I 9°
Q.
* 80
.3!
§ 70
C-10. The red lines are energy rates extrapolated from vehicle weight. The
and green lines are determined from ftp and IM bag results respectively.
Comparison of Bag Data and T (h
Linear Fits of Vehicle Weight Groups to Tv = 1.1382x /
Determine Base Rate for
Engine Size Bin 20, Model Year Group 2, Vehicle Wei
O scaled ftp
D scaled im240
X y=x
- —Linear (scaled ftp)
" " Linear (scaled im240)
Linear (y=x)
20
30
40 50 60 70 80
mean base rate (kJ) - engine weight
90
100
110
120
Figure C-ll. A comparison of the emissions rates determined from FTP and IM bag
data in comparison to the rates from vehicle weight extrapolation.
115
-------
C.4 Hole Filling Method 3: PERE
This methodology calculates energy rates by employing constraints that are
strictly defined by a physical model, PERE. The second-by-second output is binned to
obtain emissions rates. There are two methods of employing PERE: a modeled average
vehicle (within the source bin) may be calibrated to bag data, or it can predict fuel
consumption as a stand-alone (pre-calibrated) model. The latter is possible since PERE is
a scalable model.
The model is run for each driving cycle (for which bag data exists). Average
masses and TRLHP are determined. Using typical coefficients for the engine efficiency,
friction, transmission shift points, transmissions efficiency, etc., second-by-second
energy rates can be determined. These energy rates are then binned by operating mode
and summed to give a total bag estimate. This comparison (for CO?) is shown on Figure
12. The PERE estimates are within 4 percent of the bag measurements. The model could
in turn be calibrated to the bag data by adjusting some of the parameters in PERE. This
would ensure proper correlation to the bag.
Alternately, instead of modeling an average vehicle in the source bin, one could
model each vehicle (represented by bag data). This is more cumbersome, but doable,
especially if automated within MOVES.
CO2 bag (PERE vs meas) for ID
380 -i
370
DCO2PERE
• CO2 meas
bag 2
Figure C-12. Comparison of measured COi with that calculated from PERE.
116
-------
Figure C-13 shows the energy rates by operating mode bin. Figure C-14,
compares the rates determined from PERE with those determined from the previous
section. The PERE results averaged the FTP and IM rates. The PERE derived values
(solid blue diamonds) more closely follow the extrapolated values at low VSP. This is
due to the fact that PERE models engine friction explicitly. However, at higher speeds
(>50mph), the model predicts significantly lower levels at all 3 VSP levels. As stated in
the previous section, a direct comparison between method 3 and method 1 is not
appropriate. The "average" vehicles in these two cases are not identical. Therefore, lower
values at high speed may be real and correct. However, if there is valid reason to presume
that the mode is underpredicting, this could be improved by calibrating PERE more
accurately. In this study, PERE was not calibrated, the default values of coefficients were
used. Moreover, it is conceivable that these "older" (carbuereted) vehicles ran richer,
which would increase energy consumption. This too could be calibrated explicitly in
PERE.
The uncertainty bars in Figure C-14, derive from the natural variability of PERE
within each bin. A better estimate of uncertainty could be achieved through the use of an
application which, when linked to Excel, produces Monte Carlo uncertainty estimates.
This was not done for this study.
CD
c
LJJ
O
10
O)
100
80
60
40
20
J1L
1.
ILL
JlL
N= 197 263 107 161 111 78 45 26 75 160 153 46 25 39 37 45 39
0 1 11 12 13 14 15 16 21 22 23 24 25 26 33 35 36
VSPBIN2
Figure C-13. Energy rates determined from PERE binned by operating mode.
117
-------
130
fears 1981 through 19!
120 -h
Weight Cla
ss id = 40
0 5 10 15 20_, ._, 25 30 35
op mode id
Figure C-14. PERE rates (method 3) compared with methods 1 & 2.
40
C.5 Conclusion
This study compared three methods for filling data holes in MOVES. The following
conclusions can be drawn from the study
• Interpolation and extrapolation from the existing data of surrounding cells (method 1)
is possible, however, due to data limitations (sampling issues), there may be
unphysical trends visible in the data (see next point).
• A source bin, which contains only a single vehicle, may not be much better than an
empty bin.
• Interpolation and extrapolation by vehicle weight is more dependable than that by
engine size.
• It may be advantageous to "collapse" the number of engine size categories. This will
decrease the overall number of bins and should decrease the number of "holes".
• Estimating energy rates by scaling to bag data (method 2) gives skewed results. It
tends to underpredict at low VSP and overpredict at high VSP compared to method 1.
This may be due to the fact that engine load is not necessarily directly proportional to
road load.
• Calculating energy rates using PERE (method 3) predicts bag accurately (within 4
percent) even without a calibration step. VSP trends are similar to those obtained
from method 1, but is lower at higher speeds. This may be due to older cars running
fuel-rich, or due to the different vehicles represented in the two methods.
Any recommended course of action must meet the criteria for integration into MOVES.
These are:
118
-------
• Accuracy
• Ease of use
• Ease of coding and integration
• Ease of updates
According to this study, interpolation/extrapolation by vehicle weight technique of
filling holes is probably the simplest to integrate into MOVES. Scaling bag data (method
2) is not recommended due to the systematic errors involved. Of the three methods
compared, using bag data to calibrate PERE is probably the most accurate method of
filling holes. It has the advantage of employing bag data, thus taking advantage of as
much data as possible. It can also estimate rates where bag data is not present. Moreover,
PERE is relatively easy to implement and code at this stage (without criteria pollutants).
Certainly in data poor regions, where neighboring points are sparse, there is little
alternative but to use PERE. An example of data poor regions are in the source bins for
light duty diesel vehicles, heavy duty diesel trucks, and motorcycles.
In the case of the unphysical dip in one of the source bins in Figure 5, arguably,
PERE would provide a more reliable estimate than the data from that single vehicle. It is
possible to use method 1 in situations where neighboring cells are plentiful and method 3,
where they are not. However, it is recommended to stay consistent throughout the model
and use a single approach, which would simplify the coding process and overall design of
MOVES.
Looking ahead to criteria pollutants, the same questions will arise: How will data
holes be filled? In that situation, weight will be less important for extrapolation, and bag
data may be a more important guide. In such a case, PERE would likely be the most
accurate option as it provides another independent check on the results.
119
-------
Appendix D: Algorithm for Running Energy Hole
Filling Using Interpolation / Copying
D.I Purpose and Scope
The goal of this process was to generate values for the meanBaseRate for
sourceBinID x opModelD cells not populated by data or by the Physical Emission Rate
Simulator (PERE). Throughout this discussion, we will refer to these cells as "empty
cells." We generated values for the empty cells by imputing them from neighboring cells
having values.
The process of imputation was applied to runningEnergy representing gasoline
and diesel vehicles produced in model years to 2010 (shortModYrGrpID = 01 to 05).
Values for natural-gas vehicles were generated using the process described in Section 4.
The scope of the imputation analyses covered 689 sourceBins, with 17 opmodes
in each sourceBin, giving a total of 11,713 cells. Of these, 2,848 cells contain values
estimated from data and 1,624 cells have values estimated by PERE. Thus, following the
use of data and PERE, 7,241 cells remained "empty," i.e., lacked values for the
meanBaseRate.
To maintain correspondence between the fleet activity and emissionRate
components of MOVES, it is important to provide values for the empty cells. Because
MOVES represents 100% of fleet activity, it is preferable to provide meanBaseRate
values representing all activity. If cells in the emissionRate table were left empty, the
model calculations would effectively assume that activity corresponding to empty cells
requires no energy, resulting in underestimation of energy consumption.
D.2 Approach
The approach selected was to impute values from "neighboring sourcebins,"
where neighboring sourcebins are defined as set of sourcebins for which the values of all
attributes are equal, except for those in the attribute used for imputation. This analysis
used three attributes for imputation: weight class (weightClassID), displacement class
(engSizelD) and model-year group (shortMdYrGrpID). Considering the sourcebin
attributes for runningEnergy as defining a five-way matrix, these three attributes will be
referred to as the "dimensions" of imputation, denoted by the labels and symbols weight
(wt, m), disp (V) and myg (7), respectively. Table D-l shows examples of neighboring
sourcebins in each of the three imputation dimensions.
120
-------
Table D-l Examples of Neighboring Cells By Three Attributes Used for
Imputation
SourceBin Description
Source-Bin label1
MeanBaseRate
in opMode 15
(kJ/SHO)2
Data Source
Neighbors by Weight Class (Gasoline, Conventional, MY 1991-2000, 3.0-3.5 L)
< 2,000 Ib
2,000 - 2,500 Ib
2,500 - 3,000 Ib
3,000 - 3,500 Ib
3,500- 4,000 Ib
4,000 - 4,500 Ib
4,500 - 5,000 Ib
5,000 - 6,000 Ib
01-01-00-04-3035-0020
01-01-00-04-3035-0025
01-01-00-04-3035-0030
01-01-00-04-3035-0035
01-01-00-04-3035-0040
01-01-00-04-3035-0045
01-01-00-04-3035-0050
01-01-00-04-3035-0060
—
196,416
230,131
230,904
242,122
262,610
279,675
—
empty
PERE
data
data
data
data
data
empty
Neighbors by Displacement Class (Gasoline, Conventional, MY 1991-2000, 5, 000-6, 000 Ib)
2.0- 2.5 L
2.5- 3.0 L
3.0- 3.5 L
3.5- 4.0 L
4.0- 5.0 L
5.0 L and up
01-01-00-04-2025-0050
01-01-00-04-2530-0050
01-01-00-04-3035-0050
01-01-00-04-3540-0050
01-01-00-04-4050-0050
01-01-00-04-5099-0050
—
—
279,675
239,560
310,731
315,892
empty
empty
data
data
data
data
Neighbors by Model-Year Group (Gasoline, Conventional, 3.0-3.5 L, 5,000-6,000 Ib)
1986-1990
1991-2000
2001-2010
01-01-00-03-3035-0050
01-01-00-04-3035-0050
01-01-00-05-3035-0050
—
279,675
278,776
empty
data
data
1 The sourceBin label includes hyphens between the attributes to improve readability. For
example, a label of 01-01-00-04-3035-0050 corresponds to a sourceBinID of
1010100043035005000.
2 Operating mode 15 represents a VSP range of 9-12 kW/tonne, and a vehicle speed range
ofO-25mph.
D.3 Methods
We imputed values for empty cells using one of two methods, "interpolation" or
"substitution." The preferred method was to estimate a value by linear interpolation
between two neighboring sourcebins, designated as the "high-side" and the "low-side"
neighbors to the empty cell. With respect the the weight dimension, "high" means
"heavier" and "low" means "lighter." Similarly, with respect to the disp dimension,
"high" means "larger" and "low" means "smaller," and with respect to the myg
dimension, "high"means "later" and "low" means "earlier." The second option, adopted
when two eligible neighbors were not available, was to directly substitute the value of the
nearest neighbor into the empty cell.
121
-------
To allow interpolation, it was necessary to translate the coded values into numeric
values in the appropriate units, e.g., Ibs for weight, L for disp or years for myg. Thus, for
all records, we assigned numeric values to each coded value, to represent a "typical"
value of the attribute. For example, for the weight dimension, a weightClass value of
' 140' was assigned a midpoint value of 12,000 Ibs, calculated as the average of the upper
and lower bounds of the class, or (10,000+14,000)72, as shown in Table D-2. Midpoints
for disp and myg dimensions were calculated similarly, as shown in Tables D-3 and D-4.
The performance of imputation involved three major steps:
1) Identify and store values of the nearest non-empty neighbor(s) to each cell, in
each imputation dimension,
2) Assess the "eligibility" of neighbors so identified for use in imputation,
3) Perform imputation, following pre-specified orders of precedence for selection of
the method and dimension to be used.
D.3.1 Identify Nearest Non-empty Neighbors
The first step was to identify the nearest non-empty neighboring cells to each
record in the table, by each of the imputation dimensions, weight, disp and myg. We
identified neighbors in three sub-steps: (1) sorting the table to arrange neighboring cells
in sequence, (2) storing the meanBaseRate and associated fields for 15 "lags" to each
record in the table (for a given sort order, defined below) and (3) storing the nearest non-
empty lag.
D. 3.1.1 Sorting the Table
To arrange neighboring cells in sequence, we sorted the file using the operating
mode and sourceBin attributes as sort keys. The sort order of the attributes for each
imputation dimension differed (as in table D-l above). The sort orders always placed the
opMode in first position, and maintained the ordering of the attributes used in the
sourceBinID label, with the exception that the attribute for the dimension under
consideration was placed in the final position. The sort orders used for each dimension
are portrayed in Table D-5.
D. 3.1.2 Storing Lags
After each sort of the file, the process stored the meanBaseRate and identifying
information for a maximum of 15 "lags" to each record in the file. The lags are defined
as the records preceding each record in the table, in a given sort order. All lags were
stored initially, whether or not they contained values for the meanBaseRate. For each lag,
the values of the meanBaseRate and meanBaseRateCV were stored, along with attributes
needed to identify the lag and perform imputation, including:
122
-------
operating mode (opmodelD),
sourceBin attribute values (fueltypelD, engTechID, regClassID,
shortMdYrGrpID, engSizelD, weightClassID),
weight class midpoint,
displacement class midpoint,
model-year group midpoint,
dataSourcelD
D. 3.1.3 Storing Non-empty Lags
After storing up to 15 lags, the process identified the lag "nearest to" each record
in the table in the designated imputation dimension for which the meanBaseRate was not
empty. In this context, nearest means that the difference (in the imputation dimension)
between the cell under consideration and the non-empty neighbor is as small as possible.
For example, if the cell under consideration represents engSizelD 2025, and the cell
representing engSizelD 2530 is empty, then the cell representing 3035 is the nearest
neighbor, if it is not empty. Note that it would be a "high-side" neighbor, as 3035
represents larger engines than 2025 (Table D-l).
The process repeated the steps to identify non-empty neighbors six times, to
identify low-side and high-side neighbors in each of the three dimensions. For each
dimension, the table was sorted twice. In the first sort, the attribute representing the
imputation dimension was sorted in ascending order, to identify low-side neighbors; in
the second, the attribute for the imputation dimension was sorted in descending order, to
identify high-side neighbors. All attributes except that for the imputation dimension were
sorted in ascending order. Any given cell can have a maximum of six non-empty
neighbors; corresponding to 0, 1 or 2 non-null neighbors in each dimension.
123
-------
Table D-2 Midpoints for Values of the weightClassID Attribute
WeightClassID
0020
0025
0030
0035
0040
0045
0050
0060
0070
0080
0090
0100
0140
0160
0195
0260
0330
0400
0500
0600
0800
1000
1300
9999
Class Midpoint (Ib)
Formula
...
(2,000 + 2,500)72
(2,500 + 3,000)72
(3,000 + 3,500)72
(3,500 + 4,000)72
(4,000 + 4,500)72
(4,500 + 5,000)72
(5,000 + 6,000)72
(6,000 + 7,000)72
(7,000 + 8,000)72
(8,000 + 9,000)72
(9,000 + 10,000)72
(10,000 + 14,000)72
(14,000 + 16,000)72
(16,000 + 19,500)72
(19,500 + 26,000)72
(26,000 + 33,000)72
(33,000 + 40,000)72
(40,000 + 50,000)72
(50,000 + 60,000)72
(60,000 + 80,000)72
(80,000 + 100,000)72
(100,000 + 130,000)72
—
Value
1,800
2,250
2,750
3,250
3,750
4,250
4,750
5,500
6,500
7,500
8,500
9,500
12,000
15,000
17,750
22,750
29,500
36,500
45,000
55,000
70,000
90,000
115,000
148,000
Table D-3 Midpoints for Values of the engSizelD Attribute
(Engine Displacement)
engSizelD
0020
2025
2530
3035
3540
4050
5099
5099
FuelTypelD
all
all
all
all
all
all
01
02
Class Midpoint (L)
Formula
...
(2.0+2.5)72
(2.5+3.0)72
(3.0 + 3.5)72
(3.5+4.0)72
(4.0 + 5.0)72
—
—
Value
1.80
2.25
2.75
3.25
3.75
4.50
6.50
13.0
124
-------
Table D-4 Midpoints for Values of the engSizelD Attribute
(Engine Displacement)
shortModYrGrpID
01
02
03
04
05
Class Midpoint (L)
Formula
...
(1981 + 1985)72
(1986 + 1990)72
(1991+2000)72
(2001+2010)72
Value
1975.0
1983.0
1988.0
1995.5
2005.5
Table D-5 Sort Orders for sourceBin Attributes Used to Identify Neighboring
sourceBins by Imputation Dimension
Attribute
opModelD
FuelTypelD
EngTechID
RegClassID
ShortMdYrGrpID
EngSizelD
WeightClassID
Dimension
"Weight"
(weight)
1st
2nd
3rd
4th
5th
6th
-7th 1
"Displacement"
(disp)
1st
2nd
3rd
4th
5th
-7th 1
6th
"Model-year
Group"
(myg)
1st
2nd
3rd
4th
-7th 1
5th
6th
1 To identify low-side neighbors, the attribute used as the seventh sort key was sorted in
ascending order; to identify high-side neighbors, the same attribute was sorted in descending
order. All other keys were sorted in ascending order.
125
-------
D.3.2 Assess Eligibility of Identified Neighbors for Imputation
Following identification and storage of the nearest non-null neighbor or
neighbors, the process assessed whether the neighbor(s) would be "eligible" for use in
imputation. "Eligibility" was assessed on basis of several criteria, defined in terms of the
operating modes and sourceBinID attributes of the neighbors, in relation to those for the
cell under consideration, i.e., the "current cell." In all cases, the process required that
values of the operating mode, fuelTypelD, engTechID, and regClassID for the neighbor
and the current cell be equal. With respect to the dimensions themselves, additional
criteria applied, as follows:
When assessing the eligibility of neighbors by weight, the process required that
values for the other two dimensions (myg and disp) also be equal. In the weight
dimension itself, the weightClassID for a low-side neighbor had to be lower than that for
the current cell. Similarly, the weight class for a high-side neighbor had to be greater than
that for the current cell. This criterion ensured that "low-side" and "high-side" neighbors
were indeed "low" and "high," as expected. Finally, to preclude the possibility of
imputing values from "light-duty' vehicles to "heavy-duty" vehicles and vice versa, we
defined an additional criterion related to vehicle weight itself. We defined an additional
variable, "duty class" and stipulated that its values for the neighbor and the current cell
must be equal. Duty class takes two values: "light-duty" and "heavy-duty." We defined
"light-duty" vehicles as those weighing 8,000 pounds or less (weightClassID < '0080'),
and "heavy-duty" as all vehicles weighing more than 8,000 Ib. Criteria specific to the
other two dimensions were similar.
When assessing the eligibility of neighbors by disp, the process required that
values for the other two dimensions (weight and myg) be equal. In the disp dimension
itself, the engSizelD for a low-side neighbor had to be lower than that for the current cell.
Similarly, the engSizelD class for a high-side neighbor had to be greater than that for the
current cell.
When assessing the eligibility of neighbors by myg, the process required that
values for the other two dimensions (weight and disp) be equal. In the myg dimension
itself, the shortModYrGrpID for a low-side neighbor had to be lower than that for the
current cell. Similarly, the shortModYrGrpID class for a high-side neighbor had to be
greater than that for the current cell. Table D-6 summarizes the eligibility criteria.
After assessing the eligibility of low-side and high-side neighbors, we assessed
the possibility that interpolation would be an option for imputation. If the current cell had
eligible low-side and high-side neighbors, as defined above, interpolation was possible.
However, if only one eligible neighbor was available (either low-side or high-side), its
value was substituted into the current cell. If no eligible neighbors were available, no
imputation was attempted.
126
-------
Table D-6 Criteria for Selection of Neighboring SourceBins Containing Values
Eligible for Interpolation or Substitution
SourceBin Attribute by Dimension
Low-side neighor
High-side neighbor
All Dimensions
opMode
FuelTypelD
engTechID
regClassID
EQUAL
EQUAL
EQUAL
EQUAL
EQUAL
EQUAL
EQUAL
EQUAL
Dimension = " Weigh t"
shortMdYrGrpID
engSizelD
Duty class
weightClassID
EQUAL
EQUAL
EQUAL
<
EQUAL
EQUAL
EQUAL
>
Dimension = "Displacement"
shortMdYrGrpID
weightClassID
engSizelD
EQUAL
EQUAL
<
EQUAL
EQUAL
>
Dimension=" Model-year Group"
engSizelD
weightClassID
shortMdYrGrpID
EQUAL
EQUAL
<
EQUAL
EQUAL
>
D.3.3 Precedence of Imputation Dimensions and Methods
Zero to two neighbors were potentially available in each of the three imputation
dimensions, for a theoretical maximum of six neighbors. To govern the selection of the
method and dimension to be used for inputation, we applied one of three selection
sequences on a cell-by-cell basis.
The selection sequence differed for each of three groups of cells defined on the
basis of fuel type, model-year group and operating mode. The definition of each
"precedence group" is presented in Table D-7a. Table D-7b presents the options and
precedence for each group. Each option for imputation is identified with a numeric code.
The code is used to develop a dataSourcelD, which identifies the option used to fill any
given cell and on which of two rounds of imputation a cell was filled.
For each cell, the process would evaluate each option in order of precedence, until
it identified the option with the highest precedence that was also feasible. It would assign
an option ID to each cell, and perform the imputation accordingly.
127
-------
Table D-7a Precedence Groups for Selection of Imputation Options
Group
Precedence Group 1
Precedence Group 2
Precedence Group 3
fueltypelD
01 (gas)
02 (diesel)
01 (gas)
02 (diesel)
02 (diesel)
shortModelYrGrpID
all
01-04(pre2000)
all
01-04(pre2000)
05(2000-2010)
OpModelD
All except idle
All except idle
Idle (1)
Idle(l)
All opmodes
Table D-7b Options and dataSourcelD Values for Imputation of meanBaseRate
for Empty Cells
Imputation
method
Dimension
Precedence
Option
ID
dataSourcelD
1st round
2nd round
Precedence Group 1
Interpolation
Substitution
None eligible
Weight
Displacement
Model-Year Group
Model-Year Group
Weight
Displacement
1
2
3
4
5
6
7
401
402
403
404
405
406
400
4011
4021
4031
4041
4051
4061
4001
4012
4022
4032
4042
4052
4062
4002
Precedence Group 2
Interpolation
Substitution
None eligible
Displacement
Weight
Model-year Group
Model-year Group
Displacement
Weight
1
2
3
4
5
6
7
411
412
413
414
415
416
410
4111
4121
4131
4141
4151
4161
4101
4112
4122
4132
4142
4152
4162
4102
Precedence Group 3
Substitution
None eligible
Model-year Group
1
2
424
420
4241
4201
4242
4202
128
-------
D.3.4 Linear Interpolation
The process performed interpolation in two steps. In the first step, it calculated a
slope term between the low-side and high-side neighbors, in the dimension previously
assigned for a given cell. The slope terms for the three dimensions are shown in equations
D-l, D-2 and D-3, in which the superscripts on all terms represent the imputation
dimension. The denominator terms, m, Fand /represent midpoint estimates for high-side
and low-side neighboring sourcebins for the weight, disp and myg dimensions,
respectively (Tables D-2 - D-4).
In the weight dimension (superscript = wf), the slope term AE/Am represents the
change in the meanBaseRate per unit change in vehicle weight between the high and low
neighbors, as shown in Figure D-l. The additional slope terms AE/AFand AE7A7
represent corresponding changes in the meanBaseRate per unit change in displacement
volume or model year for the disp and myg dimensions, respectively.
_ .Dhigh
Wt Wt
Wfeigh OTlow CD-I")
\disp
AF I 77 p — 77
^^ -Dhigh -Did
high
fallow * ™° - (D-2)
AV I i/disp — i/disp
^v ),.,, Khigh Kiow
\"iyg
T^myg j-ttnyg
t-j high tL low
AY vmyg — vmyg
AY l^ Y^ Yl- (D.3)
In the second step, the process applies the slope term to calculate an estimate for the
empty cell by linear interpolation., as shown in equation D-4.
(D-4)
The interpolation is in simple linear form (y=bx+c). Using the example of
interpolation in the weight dimension, the terms are defined as follows:
y = the meanBaseRate for the empty cell (£1^, kJ/SHO),
x = the difference between the weight midpoints for the empty cell and its low-side
neighbor ( m^ - mL ),
b = the slope term between the high-side and low-side neighbors (AE/Am, kJ/SHO-lb),
129
-------
c = the meanBaseRate for the low-side neighbor (E\™, kJ/SHO), which acts as a y-
intercept.
Equations D-5 and D-6 show corresponding interpolations in the disp and myg
dimensions.
j-,disp _ j-,disp _,_
.fc empty ~~ .Glow ~"~
_
empty V low
AE
AV
AE_
j empty d low ~*~ I . T, I \i empty i low
(D-5)
(D-6)
If only one eligible neighbor were available, imputation was performed by
substitution rather than interpolation. However, if no eligible neighbor were available, no
imputation was attempted.
Figure D-l. Conceptual illustration of imputation by linear interpolation, using the
example of interpolation in the vehicle-weight dimension (wt).
meanBaseRate (£)
Am
weightClass
Midpoint (m)
D.3.5 Rounds of Imputation
To maximize the number of empty cells filled, we performed two rounds of
imputation. During first round, only cells filled by data or PERE were available as
neighbors. During the second round, the revised file, including initial imputation results,
130
-------
was resubmitted to the process, during which cells filled during the first round were also
available as neighbors.
Specific values of the dataSourcelD were generated to identify cells filled by
imputation. The first three digits indicate the method and dimension of imputation, as
indicated by the option ID, and the fourth digit indicates the round of imputation. The
value of the fourth digit is either T or '2,' as shown in Table D-7 above.
D.4 Results and Examples
D.4.1 Results
Table D-8 shows examples of imputation in the three dimensions. The examples
shown include those presented in Table D-l, and show how values were imputed to the
empty cells and the identity of the neighbor(s) used for imputation. The table shows
several examples of substitution in different dimensions, and one example of
interpolation by displacement class. The set of cells differing only by weightClass
(Example 1) shows that in some cases, a single cell may serve as a neighbor to several
empty cells, e.g., the '0050' cell acts as a low-side neighbor to the '0060,' '0070' and
'0080' cells. These same cells illustrate how the method avoided imputation by weight
between light-duty and heavy-duty vehicles. In imputing a value for the '0090' cell, the
method avoided substitution from the '0050' cell, defaulting instead to the next
dimension in precedence, and substituting a value by displacement. In addition, Example
3 shows that the same cell used for substitution by weight in Example 1 was also used for
substitution by model-year group (myg).
Table D-9 summarizes the results of the entire process, showing counts of cells
cross-tabulated by data source and operating mode, for all cells included in the scope of
this analysis, as defined in D.I above. Within each operating-mode column, the count
represents the number of sourceBins represented in that mode. Also within each
operating mode, the sum of all rows, representing the total of cells populated by all
datasources, is identical. This summation represents the disposition of all cells as
populated by differing dataSources.
131
-------
Table D-8 Examples of Imputation of meanBaseRate Values to Empty Cells, for Three Groups
of Related SourceBins, in Operating Mode 151
Low-side Neighbor2
Source Bin
"Current Cell"
High-side Neighbor3
meanBaseRate
(kJ/SHO)
dimension
method
DataSourcelD
Example 1: Gasoline, Conventional, MY 1991-2000, 3.0-3.5 L
01-01-00-04-3035-0050
01-01-00-04-3035-0050
01-01-00-04-3035-0050
01-01-00-03-3035-0140
01-01-00-04-3035-0020
01-01-00-04-3035-0025
01-01-00-04-3035-0030
01-01-00-04-3035-0035
01-01-00-04-3035-0040
01-01-00-04-3035-0045
01-01-00-04-3035-0050
01-01-00-04-3035-0060
01-01-00-04-3035-0070
01-01-00-04-3035-0080
01-01-00-04-3035-0090
01-01-00-04-3035-0100
01-01-00-04-3035-0140
01-01-00-04-3035-0160
01-01-00-04-3035-0025
01-01-00-04-4050-0090
01-01-00-04-4050-0100
01-01-00-04-5099-0160
196,416
196,416
230,131
230,904
242,122
262,210
279,675
279, 675
279,675
279,675
529,056
611,056
685,872
896,544
•weight
weight
•weight
•weight
disp
disp
myg
disp
subst
subst
subst
subst
subst
subst
subst
subst
4051
394
398
398
398
398
398
4051
4051
4051
4061
4061
4041
4061
Example 2: Gasoline, Conventional, MY 1991-2000, 5,000 - 6,000 Ib
01-01-00-04-2025-0045
01-01-00-03-2530-0050
01-01-00-04-2025-0050
01-01-00-04-2530-0050
01-01-00-04-3035-0050
01-01-00-04-3540-0050
01-01-00-04-4050-0050
01-01-00-04-5099-0050
260,034
237,455
279,675
239,560
310,731
315,892
weight
myg
subst
subst
4051
4041
398
398
398
398
ExampleS: Gasoline, Conventional, 3.0-3.5 L, 5.000-6,000 Ib
01-01-00-03-2530-0050
01-01-00-03-3035-0050
01-01-00-04-3035-0050
01-01-00-05-3035-0050
01-01-00-03-4050-0050
253,087
279,675
278,776
disp
interp
4021
398
398
1 Each group, as indicated by the Sourcebin Label, differs in only one sourceBin attribute.
The "low-side" neighbor used for interpolation or substitution. The attribute that differs between the "low-side" neighbor and the "current
cell" is underlined.
The "high-side" neighbor used for interpolation or substitution. The attribute that differs between the "high-side" neighbor and the "current
cell" is underlined.
132
-------
Table D-9 Distribution of Cells Representing Vehicles Manufactured Prior to 2010, for which Values were Derived from Data, Generated
by PERE, or Imputed by Interpolation or Substitution, by Operating Mode
DataSourcelD1
1001 (data)
2001 (manual)2
3001 (PERE)
4011
4111
4012
4112
4021
4121
4022
4122
4041
4141
4042
4142
4051
4151
4052
4152
4061
4161
4062
4162
4241
4242
Total
Operating Mode
0
in
i
95
10
4
16
11
105
26
115
1
83
1
22
22
689
1
170
7
95
19
13
7
2
106
26
152
2
46
0
22
22
689
11
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
12
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
13
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
14
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
75
166
7
95
15
6
16
11
105
28
115
1
79
1
22
22
689
16
156
7
95
25
9
16
11
105
29
115
1
75
1
21
23
689
21
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
22
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
23
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
24
111
1
95
10
4
16
11
105
26
115
1
83
1
22
22
689
25
168
7
95
13
5
16
11
105
28
115
1
80
1
22
22
689
26
159
7
95
22
8
16
11
105
29
115
1
76
1
22
22
689
33
168
7
98
10
4
16
11
105
26
115
1
83
1
22
22
689
35
168
7
98
10
4
16
11
105
26
115
1
83
1
22
22
689
36
154
7
98
24
8
15
11
105
29
115
1
77
1
20
24
689
Total
2,848
119
1,624
209
19
80
13
255
7
176
2
1,680
106
429
26
1,840
152
16
2
300
46
16
0
371
377
11,713
1 Values for dataSourcelD other than data or PERE are defined in table D-l
2 Cells for which values were not imputed following the second round of imputation. These cells were subsequently filled "manually."
133
-------
D.4.2 Example of Interpolation
Figure D-2 shows an example of interpolation by displacement for six operating modes.
In this case, the empty cells (squares) represent gasoline vehicles with displacement in the range
of 3.0-3.5 L, weighing from 5,000-6,000 Ibs and produced during the period 1986-1990 (01-01-
00-03-3035-0060). Estimates for these vehicles were imputed by interpolating between the low-
side neighbor (2.0-2.5L, engSizeID=2025), and the high-side neighbor (4.0-5.0 L,
engSizeID=4050), both denoted by diamonds. Note that in this case the next lower size class to
the empty cells (2.5-3.0 L, engSizeID=2530) is not available to serve as a low-side neighbor.
Thus, if the cells for size classes 0020 and 2025 were empty, interpolation would not have been
possible in the disp dimension, rather, the value for the high-side neighbor would have been
substituted. Note that although the imputed rates are not shown in the graph, the two intervening
source bins with engSizelD classes 2530 (2.5-3.0 L) and 3540 (3.5-4.0 L) were also interpolated
between size classes 2025 and 4050.
Figure D-2. Examples of interpolation by displacement class (disp) in six operating modes.
The empty cells (squares) represent sourceBin label 01-01-00-03-3035-0060, which are
interpolated between the low-side neighbor 01-01-00-03-2025-0060 and the high-side
neighbor 01-01-00-03-4050-0060 (diamonds).
800,000
700,000 -
600,000 -
500,000 -
400,000 -
300,000 -
200,000 -
100,000 -
0
opModelD = 16
opModelD = 15
opModelD = 14
opModelD = 13
opModelD = 12
opModelD = 11
0.0 1.0 2.0 3.0 4.0
Displacement Class midpoint (L)
5.0
134
-------
Appendix E: Start Energy Rates
MOVES2004 start energy rates were generated by multiplying the energy content values of
122,893 KJ/gallon for gasoline and 138,451 KJ/gallon for diesel by the fuel consumption values
presented in this appendix. Bins with a designation of "N.A." denote that the bins do not exist,
according to fleet data used for bins.
Table E-l: Fuel Consumed During Engine Starts
For Gasoline-Fueled Light-Duty Cars and Trucks (gallons per start)
Bins for Light-Duty Gasoline-Fueled
Mdl_Yr
Range
Pre-1981
1981-85
1986-90
1991+
Displ.
Range
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
No. of
Tests
444
212
73
89
225
368
232
1,993
1,614
634
251
743
1,029
115
1,888
1,470
1,031
170
443
645
55
224
322
131
133
92
57
19
Mean
(gal/start)
0.0252
0.0315
0.0309
0.0372
0.0372
0.0441
0.0512
0.0222
0.0257
0.0310
0.0389
0.0351
0.0409
0.0510
0.0177
0.0210
0.0259
0.0224
0.0248
0.0308
0.0359
0.0173
0.0189
0.0250
0.0237
0.0246
0.0295
0.0346
Std
Dev.
0.0094
0.0084
0.0120
0.0080
0.0103
0.0166
0.0230
0.0086
0.0102
0.0099
0.0076
0.0155
0.0180
0.0202
0.0078
0.0096
0.0088
0.0046
0.0062
0.0100
0.0107
0.0066
0.0048
0.0073
0.0071
0.0058
0.0084
0.0066
S.E. of
Mean
0.0004
0.0006
0.0014
0.0008
0.0007
0.0009
0.0015
0.0002
0.0003
0.0004
0.0005
0.0006
0.0006
0.0019
0.0002
0.0003
0.0003
0.0004
0.0003
0.0004
0.0014
0.0004
0.0003
0.0006
0.0006
0.0006
0.0011
0.0015
90% Confidence
Interval
0.0245
0.0306
0.0286
0.0358
0.0360
0.0427
0.0487
0.0219
0.0253
0.0304
0.0381
0.0341
0.0399
0.0479
0.0174
0.0206
0.0255
0.0219
0.0243
0.0302
0.0335
0.0166
0.0185
0.0239
0.0227
0.0236
0.0277
0.0321
0.0259
0.0325
0.0332
0.0386
0.0383
0.0455
0.0537
0.0225
0.0261
0.0317
0.0397
0.0360
0.0418
0.0541
0.0180
0.0214
0.0264
0.0230
0.0253
0.0315
0.0383
0.0180
0.0194
0.0260
0.0247
0.0256
0.0313
0.0371
135
-------
Table E-2: Fuel Consumed During Engine Starts
For Gasoline-Fueled Heavy-Duty Vehicles (gallons per start)
Bins for Heavy-Duty Gasoline-Fueled
Mdl_Yr
Range
Pre-1981
1981-85
1986-90
1991+
Displ.
Range
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
No. of
Tests
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
68
0
0
0
0
0
6
173
Mean
(gal/start)
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0515
0.0464
N.A.
N.A.
N.A.
N.A.
N.A.
0.0449
0.0460
Std
Dev.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0086
0.0241
N.A.
N.A.
N.A.
N.A.
N.A.
0.0068
0.0185
S.E. of
Mean
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0050
0.0029
N.A.
N.A.
N.A.
N.A.
N.A.
0.0028
0.0014
90% Confidence
Interval
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0434
0.0415
N.A.
N.A.
N.A.
N.A.
N.A.
0.0403
0.0437
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0597
0.0512
N.A.
N.A.
N.A.
N.A.
N.A.
0.0494
0.0483
136
-------
Table E-3: Fuel Consumed During Engine Starts
For Diesel-Fueled Light-Duty Cars and Trucks (gallons per start)
Bins for Light-Duty Diesel-Fueled
Mdl_Yr
Range
Pre-1981
1981-85
1986-90
1991+
Displ.
Range
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
No. of
Tests
3
0
4
0
0
0
2
6
0
9
0
0
0
1
0
0
0
0
0
0
0
1
0
1
0
0
0
0
Mean
(gal/start)
0.0163
N.A.
0.0387
N.A.
N.A.
N.A.
0.0358
0.0165
N.A.
0.0344
N.A.
N.A.
N.A.
0.0223
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0075
N.A.
0.0221
N.A.
N.A.
N.A.
N.A.
Std
Dev.
0.0039
N.A.
0.0063
N.A.
N.A.
N.A.
0.0002
0.0075
N.A.
0.0081
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
S.E. of
Mean
0.0023
N.A.
0.0032
N.A.
N.A.
N.A.
0.0001
0.0031
N.A.
0.0027
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
90% Confidence
Interval
0.0126
N.A.
0.0335
N.A.
N.A.
N.A.
0.0356
0.0114
N.A.
0.0300
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0201
N.A.
0.0439
N.A.
N.A.
N.A.
0.0360
0.0215
N.A.
0.0389
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
137
-------
Table E-4: Fuel Consumed During Engine Starts
For Diesel-Fueled Heavy-Duty Vehicles (gallons per start)
Bins for Heavy-Duty Diesel-Fueled
Mdl_Yr
Range
Pre-1981
1981-85
1986-90
1991+
Displ.
Range
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
<2.0
2.1-2.5
2.6-3.0
3.1-3.5
3.6-4.0
4.1-5.0
>5.0
No. of
Tests
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
0
0
0
0
0
0
29
Mean
(gal/start)
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0397
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0269
Std
Dev.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0037
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0255
S.E. of
Mean
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0026
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0047
90% Confidence
Interval
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0354
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0191
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0439
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
0.0347
138
-------
Appendix F: CH4 & N2O Rates by Model Year
Table F-l: NiO and CH4 Emission Rates for Gasoline-Fueled Motorcycles
Fuel Type
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Reg
Class
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
MC
Mdl_Yr
Group
Pre-73
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001-
2010
2011-
2020
2021-
2050
N2O Running
(g/hr)
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.131734
0.103147
0.103147
0.103147
0.103147
0.103147
0.103147
0.103147
0.103147
N2O Start
(g/start)
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.014912
0.011676
0.011676
0.011676
0.011676
0.011676
0.011676
0.011676
0.011676
CH4 Running
(g/hr)
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.667984
1.243390
1.243390
1.243390
1.243390
1.243390
1.243390
1.243390
1.243390
CH4 Start
(g/start)
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.032618
0.024315
0.024315
0.024315
0.024315
0.024315
0.024315
0.024315
0.024315
139
-------
Table F-2: N2O and CH4 Emission Rates for Gasoline-Fueled Passenger Cars
Fuel Type
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Reg
Class
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
Mdl_Yr
Group
Pre-73
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001-
2010
2011-
2020
2021-
2050
N2O Running
(g/hr)
0.250542
0.250542
0.250542
0.555818
0.574897
0.574897
0.593977
0.593977
0.625594
0.784376
0.786167
0.789749
0.811242
0.811242
0.811242
0.811242
0.811242
0.811242
0.811242
0.811242
0.811242
0.811242
0.600749
0.390256
0.284740
0.279343
0.249104
0.193783
0.130111
0.008406
0.008406
0.008406
N2O Start
(g/start)
0.028361
0.028361
0.028361
0.062917
0.065077
0.065077
0.067236
0.067236
0.070815
0.088789
0.088992
0.089397
0.091830
0.091830
0.091830
0.091830
0.091830
0.091830
0.091830
0.091830
0.091830
0.091830
0.100218
0.108606
0.112137
0.112016
0.109853
0.105323
0.100116
0.090150
0.090150
0.090150
CH4 Running
(g/hr)
3.172287
3.020166
3.020166
2.666425
2.644317
2.644317
2.622208
2.622208
2.503794
1.408570
1.394812
1.367297
1.202202
1.202202
1.202202
1.202202
1.202202
1.202202
1.202202
1.202202
1.202202
1.202202
0.817028
0.431854
0.247517
0.242012
0.230393
0.216592
0.200625
0.170263
0.170263
0.170263
CH4 Start
(g/start)
0.062035
0.059060
0.059060
0.018868
0.016356
0.016356
0.013844
0.013844
0.013112
0.030436
0.030690
0.031198
0.034250
0.034250
0.034250
0.034250
0.034250
0.034250
0.034250
0.034250
0.034250
0.034250
0.042634
0.051018
0.054534
0.054406
0.052176
0.047512
0.042151
0.031890
0.031890
0.031890
140
-------
Table F-3: NiO and CH4 Emission Rates for Gasoline-Fueled Light-Duty Trucks
Fuel Type
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Reg
Class
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
Mdl_Yr
Group
Pre-73
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001-
2010
2011-
2020
2021-
2050
N2O Running
(g/hr)
0.286861
0.283361
0.283361
0.660396
0.714259
0.687327
0.687327
0.714259
0.714259
0.848572
0.875161
0.928338
0.981516
1.034694
1.087871
1.327171
1.327171
1.327171
1.327171
1.327171
1.327171
1.327171
1.129403
0.905046
0.792867
0.792867
0.638985
0.462021
0.523574
0.023458
0.023458
0.023458
N2O Start
(g/start)
0.032472
0.032076
0.032076
0.074755
0.080852
0.077803
0.077803
0.080852
0.080852
0.096056
0.099065
0.105085
0.111104
0.117124
0.123144
0.150232
0.150232
0.150232
0.150232
0.150232
0.150232
0.150232
0.172129
0.191016
0.200460
0.200460
0.172150
0.139594
0.150918
0.058910
0.058910
0.058910
CH4 Running
(g/hr)
3.632152
3.415790
3.415790
2.804289
2.716932
2.760611
2.760611
2.716932
2.716932
2.475471
2.408724
2.275231
2.141738
2.008244
1.874751
1.274031
1.274031
1.274031
1.274031
1.274031
1.274031
1.274031
0.906793
0.606301
0.456055
0.456055
0.406286
0.349051
0.368959
0.207209
0.207209
0.207209
CH4 Start
(g/start)
0.071028
0.066797
0.066797
0.089675
0.092943
0.091309
0.091309
0.092943
0.092943
0.098129
0.096777
0.094074
0.091371
0.088668
0.085965
0.073802
0.073802
0.073802
0.073802
0.073802
0.073802
0.073802
0.076410
0.080370
0.082350
0.082350
0.075078
0.066715
0.069624
0.045990
0.045990
0.045990
141
-------
Table F-4: N2O and CH4 Emission Rates for
Gasoline-Fueled Buses and Heavy-Duty Trucks
Fuel Type
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Gasoline
Reg
Class
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
Mdl_Yr
Group
Pre-73
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001-
2010
2011-
2020
2021-
2050
N2O Running
(g/hr)
0.653748
0.653748
0.653748
0.653748
0.653748
0.653748
0.653748
0.653748
0.653748
0.653748
0.706594
0.706594
0.706594
0.676812
0.676812
1.107280
1.216107
1.216107
1.484948
1.484948
1.484948
1.484948
1.484948
1.484948
1.756906
1.685840
1.556483
1.273626
0.896482
0.047909
0.047909
0.047909
N2O Start
(g/start)
0.074002
0.074002
0.074002
0.074002
0.074002
0.074002
0.074002
0.074002
0.074002
0.074002
0.079984
0.079984
0.079984
0.076613
0.076613
0.125341
0.137659
0.137659
0.168091
0.168091
0.168091
0.168091
0.168091
0.168091
0.345849
0.383027
0.397853
0.345815
0.276431
0.120317
0.120317
0.120317
CH4 Running
(g/hr)
8.277566
8.277566
8.277566
8.277566
8.277566
8.277566
8.277566
8.277566
8.277566
8.277566
8.038779
8.038779
8.038779
7.302683
7.302683
6.345122
5.945032
5.945032
5.373339
5.373339
5.373339
5.373339
5.373339
5.373339
1.564622
0.945586
0.475571
0.465752
0.452659
0.423200
0.423200
0.423200
CH4 Start
(g/start)
0.161870
0.161870
0.161870
0.161870
0.161870
0.161870
0.161870
0.161870
0.161870
0.161870
0.164551
0.164551
0.164551
0.150156
0.150156
0.162482
0.169359
0.169359
0.176430
0.176430
0.176430
0.176430
0.176430
0.176430
0.178185
0.169328
0.160324
0.147875
0.131276
0.093929
0.093929
0.093929
142
-------
Table F-5: N2O and CH4 Emission Rates for
Diesel-Fueled Passenger Cars
Fuel Type
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Reg
Class
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
Mdl_Yr
Group
Pre-73
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001-
2010
2011-
2020
2021-
2050
N2O Running
(g/hr)
0.027897
0.027897
0.027897
0.027897
0.027897
0.027897
0.027897
0.027897
0.027897
0.027897
0.027897
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023538
0.023028
0.023028
0.023028
0.023028
0.023028
0.023028
0.023028
0.023028
N2O Start
(g/start)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
CH4 Running
(g/hr)
0.023579
0.023579
0.023579
0.023579
0.023579
0.023579
0.023579
0.023579
0.023579
0.023579
0.023579
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019895
0.019464
0.019464
0.019464
0.019464
0.019464
0.019464
0.019464
0.019464
CH4 Start
(g/start)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
143
-------
Table F-6: Proposed N2O and CH4 Emission Rates
For Diesel-Fueled Light-Duty Trucks
Fuel Type
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Reg Class
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
LOT
Mdl_Yr
Group
Pre-73
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001-2010
2011-2020
2021-2050
N2O
Running
(g/hr)
0.036326
0.036326
0.036326
0.036326
0.036326
0.036326
0.036326
0.036326
0.036326
0.036326
0.036326
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.031443
0.032094
0.032094
0.032094
0.032094
0.032094
0.032094
0.032094
0.032094
N2O Start
(g/start)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
CH4
Running
(g/hr)
0.030703
0.030703
0.030703
0.030703
0.030703
0.030703
0.030703
0.030703
0.030703
0.030703
0.030703
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.026576
0.027127
0.027127
0.027127
0.027127
0.027127
0.027127
0.027127
0.027127
CH4 Start
(g/start)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
144
-------
Table F-7: Proposed N2O and CH4 Emission Rates
For Diesel-Fueled Buses and Heavy-Duty Trucks
Fuel Type
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Diesel
Reg
Class
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
HOT
Mdl_Yr
Group
Pre-73
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001-
2010
2011-
2020
2021-
2050
N2O Running
(g/hr)
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
0.095981
N2O Start
(g/start)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
CH4 Running
(g/hr)
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
0.081124
CH4 Start
(g/start)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
145
-------
Table F-8: N2O and CH4 Emission Rates for Alternative-Fueled Vehicles
Fuel Type
CNG
CNG
CNG
CNG
CNG
CNG
CNG
LPG
LPG
LPG
LPG
LPG
LPG
Ethanol
Ethanol
Methanol
Methanol
Electric
Reg
Class
LDV
LDV
LOT
LOT
Buses
(HDV
)
HDVs
(ALL)
HDVs
(ALL)
LDV
LDV
LDV
LOT
HDVs
HDVs
HDVs
HDVs
HDVs
HDVs
ALL
Mdl_Yr
Group
1996-2000
2001-2010
1996-2000
2001-2010
1985-1988
1989-2000
2001-2010
1982-1984
2001-2010
1977-1989
2001-2010
1979-2000
2001-2010
1989-1993
2001-2010
1992
2001-2010
ALL
N2O Running
(g/hr)
1.083132
1.083132
1.083132
1.083132
1.552808
2.846815
2.846815
1.456956
1.456956
1.456956
1.456956
1.437785
1.437785
3.489026
3.489026
2.079996
2.079996
0
N2O Start
(g/start)
0.427926
0.427926
0.427926
0.427926
0.613488
1.124727
1.124727
0.575618
0.575618
0.575618
0.575618
0.568044
0.568044
1.378453
1.378453
0.821770
0.821770
0
CH4 Running
(g/hr)
11.116180
11.116180
11.116180
11.116180
151.004912
117.109077
117.109077
7.406733
7.406733
7.406733
7.406733
1.313509
1.313509
25.285053
25.285053
7.856731
7.856731
0
CH4 Start
(g/start)
2.563660
2.563660
2.563660
2.563660
34.825390
27.008190
27.008190
1.708172
1.708172
1.708172
1.708172
0.302927
0.302927
5.831346
5.831346
1.811952
1.811952
0
146
-------
Appendix G: Comments on Emission Analysis Plan
In the fall of 2002 EPA published planning documentation for MOVES in the form of
two reports: "Draft Design and Implementation Plan for EPA's Multi-Scale Motor Vehicle &
Equipment Emission System (MOVES)", and "Emission Analysis Plan for MOVES GHG".
These reports subsequently underwent two paths of review: public review, in which comments
were solicited from model stakeholders, and independent formal peer review, in which
comments were solicited from peer reviewers paid by the Agency according to Agency peer
review guidelines.
This appendix provides a summary of written comments pertaining to energy and
emission analysis issues received as a result of both review paths (comments pertaining to the
design plan are responded to in a separate document, the MOVES2004 Software Design
Reference Manual). The commenter of a specific comment is identified in parentheses,
identified according to the bolded designation in the list below. Each comment number captures
a unique sentiment, although variation on the general idea may have been submitted by several
commenters as reflected in commenter identifications.
Comments were received from the following parties:
Public Review:
• Alliance of Automobile Manufacturers (AAM)
o Review prepared by Tom Darlington, AIR
• Engine Manufacturer's Association (EMA)
o Review prepared by Tom Darlington, AIR
• David Roden, AECOM Consult
o Review prepared for U. S. DOT with respect to MOVES design
applicability to TRANSEVIS
• U.S. DOT - FHWA
• Peter McClintock, Applied Analysis
o written comments in response to 11/2002 workshop
• Wayne El son, EPA Region 10
• Donald Stedman, Professor, University of Denver
• Natural Propane Gas Association / Propane Vehicle Council
• Phyllis Jones, North Carolina DAQ
Formal Peer Review:
• Administered by Southwest Research Institute
• Reviewers:
o Marc Ross, Professor Emeritus, University of Michigan
o Ted Russell, Professor, Georgia Tech
o Michael Replogle, Transportation Director, Environmental Defense
o Janet Buckingham, Principle Analyst, Southwest Research Institute
(Emission Analysis Plan only)
147
-------
Comments are subdivided into subareas as appropriate.
Fuel-based emission rates and inventories:
1. Comment: Producing fuel-based emission estimates will be useful (Russell). Fuel-based
emission estimates will reduce variability (Stedman).
EPA Response: These comments are more relevant to the yet-to-be-developed criteria
pollutant version of MOVES, but asMOVES2004 estimates energy consumption (easily
converted to fuel consumption) it does provide a foundation for investigating this
approach.
2. Comment: An alternative method for generating emission inventories would be to use
fuel sales rather than VMT, and remote sensing device (RSD) data used for direct
emission factors. The proposed "bottom-up" approach should be checked against top-
down fuel sales data (Stedman, Russell, U.S. DOT)
EPA Response: The primary purpose for developing MOVES2004 as an energy
consumption model is to make the comparison of bottom-up fuel consumption to top-
down fuel sales, as suggested. The results of this comparison will be published in a
separate document, "MOVES2004 Validation Results ". As mentioned under comment 1,
MOVES2004 could provide the foundation for the inventory methodology provided, but
the efficacy of this approach would need to be investigated further. One drawback of the
suggested approach is it does not allow any breakdown of emission inventory to sub-
regional levels, e.g. roadway type.
Binning/analysis approach:
3. Comment: The binning approach is more problematic for some variables than others.
VSP binning (and the use of VSP as the driving variable that determines emissions) is a
good approach, but vehicle odometer would be better served through a linear function
(Ross, Russell). Linear functions are preferable to binning, particularly for uncertainty
estimation (Russell)
EPA Response: The main benefit of the binning approach is that it allows easier
processing of raw test data into emission rates, which serves the use case of having a
model which can be easily updated based on new data. While we think this approach has
proven feasible, we can see the merit of using functions instead of bins from the
standpoint of model runtime performance, so it is something we would consider as model
development progresses. We agree some variables can't be binned, hence the inclusion
of adjustment factors for temperature and air conditioning.
4. Comment: Additional binning by speed, engine displacement and odometer is
recommended, as is investigation of other variables (Replogle). Where other operating
variables also examined besides cycle speed (Buckingham)?
148
-------
EPA Response: As detailed in this report, we did end up binning by speed and engine
displacement; odometer doesn 't turn up as very important with regard to fuel
consumption, but clearly some mechanism (or vehicle age) will need to be added for
criteria pollutants. We did investigate other variables than cycle speed (and ended up
choosing instantaneous speed), as discussed in Appendix A.
5. Comment: Allowing source bins to vary by pollutant is supported, to allow MOVES to
be credible across a wide range of applications (Replogle)
EPA Response: none required- this supports our current design.
6. Comment: Direct measurement of acceleration should be used in VSP calculation, rather
than indirect methods (Ross). Using an aggregate statistic for acceleration/deceleration
rates (e.g. "jerk") appears warranted, particularly for HC (Replogle)
EPA Response: Most test programs don't include a direct measure of acceleration,
other than the difference in speed measurement from one second to the next. We 're not
sure how much this would improve the precision of the model. We do agree that the
binning approach for HC might be improved, but in general HC emissions are so varied
(especially second-by-second) there may always be more uncertainty than with other
pollutants.
1. Comment: Evaluations presented in emission analysis plan do not consider sensitivity to
changes in control systems over the years. The approach should be analyzed by
comparing two or more datasets with vehicles that use control technologies of different
vintages (Ross). The proposal should be evaluated on second-by-second data from for
non-Tier 1 light duty gasoline vehicles, light-duty gasoline trucks, light-duty diesel
trucks, heavy-duty gasoline and heavy-duty diesel vehicles (AAM, EMA).
EPA Response: The proof-of-concept binning evaluation presented in Appendix, A was
designed to address this comment - we purposely included trucks, and a range of model
years and control technologies, in the light-duty evaluation, and extended to analysis to
include heavy-duty diesel. We feel the binning approach proved effective across the
range of these vehicles.
8. Comment: EPA should not rely on CC>2 performance in the selection of an appropriate
emissions modeling approach for any group of vehicles (AAM, EMA)
EPA Response: To address this comment, the proof-of-concept binning evaluation
focused not only on CO2 (fuel), but HC, CO andNOx as well.
9. Comment: The proposed binning approach based on VSP was based on Tier 1 vehicles
and may not be appropriate for LEV or Tier 2 vehicles. Tier 2 vehicles no longer
differentiate based on weight class, so the MOVES design should ensure that emissions
are equivalent for all passenger vehicle weight categories. (AAM, EMA)
149
-------
EPA Response: We don't think anything about newer technology vehicles will affect the
appropriateness of the binning approach, but can investigate it as data of these vehicles
becomes available. For energy consumption, the weight classes are for loaded weight,
and the importance of this on fuel consumption won't change for Tier 2. Accounting for
the provisions of Tier 2 will be handled in the criteria model.
10. Comment: Concern that large number of VSP bins brought on by criteria that no more
than 10 percent of total emissions be allowed in a given bin may have sparse emission
data; was analysis performed with looser cutpoint (e.g. 15 percent)? (Buckingham)
EPA Response: The original analysis didn 't consider a looser cutpoint, but the final
binning analysis presented in Appendix A did relax this criteria, and some of the bins
account for more than 10 percent of the time.
11. Comment: EPA has not adequately characterized which variables are most important in
explaining variability in HC, CO, and NOx emissions. These analyses should be
conducted on test data from a variety of vehicle types and technologies before selecting a
binning strategy for each of these pollutants.
EPA Response: We still consider everything we 've published on HC, CO and NOx as
preliminary, although the binning analysis did consider a broader range of vehicles and
vehicle operation, with good result.
12. Comment: The "Modeling Dataset" used by NCSU uses an uneven mixture of data from
onboard vehicles, twin-roll dynamometers, and single-roll dynamometers. Due to the
limitations of twin-roll dynamometers and onboard instrumentation, and differences in
time-alignment methods, the quality of the second-by-second data is suspect (AAM,
EMA)
EPA Response: We did substantial quality checking of second-by-second data, including
time alignment were necessary, as discussed in Sections 3 and 4 of this report.
13. Comment: NCSU should have done an apples-to-apples comparison by comparing the
same vehicles driven on different cycles rather than validating on an independent
sampling of vehicles and cycles. In addition, NCSU used the same estimates for frontal
area and rolling resistance for all vehicles in the "Modeling Dataset". The emissions
correlations could have been improved by using vehicle-specific estimates for these
inputs (AAM, EMA).
EPA Response: The binning analysis from Appendix A was designed to address this
comment - we did the "apples-to-apples " comparison suggested, to remove the added
uncertainty of vehicle-to-vehicle variability. We also used vehicle specific estimates for
road load coefficient as suggested.
150
-------
14. Comment: How will EPA estimate VSP coefficients by vehicle use type, and how will
uncertainty in these values affect uncertainty of the emission results (AAM, EMA).
EPA Response: The method for doing this in the development of emission rates is
discussed in Appendix B. An algorithm relating road load coefficients to vehicle weight
was required, since reliable data generally isn 't available from test programs.
Use of Physical Model (PERE)
15. Comment: The use of both physically and empirically-based analysis is supported. Both
have advantages and limitations which can be traded off (Replogle, Ross)
EPA Response: no response required
16. Comment: Engine-out emissions have become relatively stable in recent years, so
modeling engine-out distinctly from tailpipe could be useful for modeling emission
variability and modeling new technologies (Ross)
EPA Response: PERE employs this approach.
17. Comment: Calibration of physical model to empirical data should include inventory-
weighted comparison (Russell). The method for calibrating the physical model has not
been adequately described and requires more detail (Buckingham). Physical model
provides more opportunity for "gaming", which should be managed by calibration
process (Replogle).
EPA Response: Any physical model developed will be calibrated to known data, whether
that data is from chassis dynamometer, on-road, remote sensing, or some combination
has yet to be determined.
18. Comment: Given the number of source and operating mode bins it would be helpful to
estimate how many might be flagged as not having enough data (i.e. less than 40 seconds
based on the criteria presented in the emission analysis plan) (Buckingham). Not enough
analysis presented on the 40 second criteria (AAM, EMA)
EPA Response: This quality check has been done and is described in some of the
MOVES documentation. The statistical criteria for bin adequacy has been expanded.
19. Comment: What method for estimating emission rates will be used for bins with little
data, but no bag data (Buckingham)? EPA currently does not plan to use the MSOD bag
data (other than to "calibrate" the PERE), a huge source of EPA data which has been
relied upon in developing the MOBILE series of models. EPA should explore more
options for using these data to "anchor" the MOVES model, especially for older vehicles
(AAM, EMA).
151
-------
EPA Response: Hole-filling methods are addressed in Section 4. It is true that for the
binning approach, bag data is not used directly, except to calibrate PERE; although bag
data is used for generated adjustment factors, start rates, and CH4 /N2O emission rates.
We will investigate whether we need to employ bag data for criteria pollutants.
20. Comment: EPA plans to use the PERE model to fill "data-gaps." The development
work on PERE has focused on warmed-up operation from Tier 1 light duty vehicles only;
therefore, this work has not progressed to the point where a decision can be made to rely
on this model for this purpose (AAM, EM A).
EPA Response: The handling of cold start in PERE is discussed further in the PERE
documentation.
Time Weighting of Emissions:
21. Comment: Weighting by time rather than be vehicle will reduce the influence of high
emitters with small amounts of data (i.e. RSD) (Stedman, Ross). The one-second
weighting approach proposed for MOVES is accepted as valid (Buckingham, Replogle,
Russell)
EPA Response: MOVES takes the time-weighted approach, so these comments support
the chosen approach
Age and Odometer:
22. Comment: Age is a better indicator of vehicle deterioration than odometer and should be
considered in MOVES (McClintock). The correlation between odometer and age is not
static, and to assume so is a considerable source of uncertainty (Stedman)
EPA Response: This is more relevant to criteria pollutants, and we will be looking into
the question of age versus odometer.
23. Comment: More emphasis should be placed on older vehicles. It would be challenging
to do an analysis of policies addressing older vehicles in MOVES unless the user
provided data directly. Information on older vehicles will tend to be inaccurate and
location-dependent (Ross)
EPA Response: This is more relevant to criteria pollutants, and we will be looking into
the question of age versus odometer.
Uncertainty and Sensitivity:
24. Comment: In general the inclusion of uncertainty is supported (Russell, EPA Region
10). How will uncertainty be used in a policy context, i.e. the application of uncertainty
tolerances in determining conformity? (U.S. DOT)
152
-------
EPA Response: We are still working out how to include uncertainty in MOVES - it is
NOT included in MOVES2004. As discussed in the MOVES2004 Software Design
Reference Manual, we have investigated both a propagation of error method and Monte
Carlo simulation. We believe the latter is most feasible for the MOVES design, but need
to consider it would be implemented in terms of model runtime performance. The
MOVES database does have placeholders for coefficient ofvaration (CV), and these were
calculated by the binner program and used to populate the EmissionRate table. The
policy context of uncertainty would need to be addressed with official guidance.
25. Comment: The equation given for calculating uncertainty is an approximation; a more
pure form would explicitly account for uncertainty and sensitivity of each term (Russell)
EPA Response: We would attempt to do this with a Monte Carlo simulation approach.
26. Comment: Monte Carlo methods for generating uncertainty are time consuming;
analytical methods as proposed in the emission analysis plan are faster and can identify
sensitivities (Buckingham, Russell). Use of such methods requires an assumption of
normality, but this is appropriate (Buckingham).
EPA Response: We investigated propagation of error, but found it impractical due to
certain aspects of the MOVES design. Model runtime performance is the biggest concern
with Monte Carlo.
27. Comment: MOVES does not address location dependence of activity or emission
information well; specifically, operating mode distributions, information on older
vehicles, and vehicle age distribution (Ross). Uncertainty of vehicle activity and
characteristic components are not considered, and could be considerable, resulting in
serious underestimation of overall uncertainty (AAM, Replogle). Will MOVES consider
uncertainty of the model computation process and the input data supplied by the user?
(U.S. DOT)
EPA Response: While local areas can certainty customize any input if data is available,
we will not have the resources to develop location-specific defaults - we focus on
national average defaults. Although uncertainty is not employed in MOVES2004 we
have designed the database to include Coefficient of Variation (CV) for all inputs,
including activity. Thus uncertainty of input data could be accounted for, but not the
model computation process.
28. Comment: The uncertainty correction for different averaging times proposed in the
emission analysis plan appears justified and reasonable (Buckingham)
EPA Response: no response necessary
29. Comment: EPA should perform a comprehensive sensitivity analysis for the model and
publish the results so this task is not left to the users (U.S. DOT)
153
-------
EPA Response: We do plan to perform sensitivity analysis, but it will be after the model
is made public.
High Emitters and Tampered Vehicles:
30. Comment: Representative vehicle recruitment will continue to be an issue
(Buckingham). The success of using a continuous distribution to define variability in
fleet will depend largely on the ability to recruit high emitters (Ross). EPA voluntary
programs and IM programs tend to leave out the grossest emitters (Stedman). How will
high emitter data be selected, collected and included in bins (U.S. DOT)?
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES; the issue of representative sampling is long-standing issue for
MOBILE as well as MOVES, and we plan to look at IM program andRSD data to help
determine how to best represent the in-use fleet.
31. Comment: The single distribution method is not recommended for characterizing high
emitters, because the shape can easily be skewed, masking the true variation of normal
and high emitting groups. Malfunction categories may not be easy to implement. The
discrete emitter category approach is recommended (Buckingham).
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES.
32. Comment: EPA's proposed methods of calibrating emitter distributions in MOVES to
local data contradicts its preliminary decisions on the use of IM240 and remote sensing
data in developing emission rates for VSP bins (AAM, EMA).
EPA Response: We didn 't use IM240 (program) or RSD data for MOVES2004, but will
be looking at it as we develop the criteria pollutant version of MOVES.
33. Comment: Treatment of tampering needs to be more explicitly addressed; new surveys
are needed (EPA Region 10)
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES.
34. Comment: The "unrepresented" distribution method is preferable because it captures I/M
and other strategies as well as local data (NC DAQ)
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES.
154
-------
On-Board Measurement (PEMS):
35. Comment: EPA should not use PEMS until data accuracy issues are resolved (AAM,
EMA)
EPA Response: We didn 't end up using PEMS data directly in MOVES2004 (it was
used in proof-of-concept evaluations, e.g. Appendix A, but not the model itself); however,
we believe the shootout program and PEMS programs conducted since have
demonstrated the accuracy of the approach.
36. Comment: On-Board measurement may be costly and inefficient; flexibility should be
retained in data collection (Ross)
EPA Response: As we move into more on-boarddata collection, we would likely
supplement data analysis with other sources of data to ensure representativeness.
Ultimately we believe PEMS will be much more cost effective and efficient than
laboratory testing, especially for heavy duty vehicles.
Data Collection
37. Comment: Data proposed for MOVES GHG appears to be a good cross section of labs,
companies and studies with no bias toward a particular study. The EPA MSOD format
and a definitive set of units is suggested as a requirement for future studies to reduce the
merging effort. Is there a recommended method for cross-checking data quality
(Buckingham)?
EPA Response: We did do substantial QA on the data, as described in Section 3 and 4.
We do envision compiling our data requirements for the benefit of other test
organizations, so that their data might be more easily incorporated into MOVES.
38. Comment: IM240 data should not be used to determine VSP bin emission rates in
MOVES until the uncertainties in fuels, test temperatures, and vehicle preconditioning
are resolved or addressed. (AAM, EMA). IM240 data has fuel inconsistencies but would
be sufficient as a validation tool for CO2 data. (NC DAQ)
EPA Response: We did not use IM240 program data for MOVES2004, and agree that
these issues would need to be addressed in order to include it in MOVES in the future.
39. Comment: Use of the US06 cycle for model and inventory development, including its
use for the calibration of the PERE, is inappropriate. The US06 cycle is a severe cycle
and is not representative of real-world driving. We believe that the use of the US06
driving cycle in the development of MOVES would affect inventory development
assessments performed using the model. (AAM, EMA)
155
-------
EPA Response: While we did exclude certain cycles in the development of MOVES
(Section 4), we decided to keep the US06 data because it contains real-world aggressive
driving not found on other cycles.
40. Comment: Some vehicle classes/ages are poorly represented in the Mobile Source
Observational Database (MSOD). For example, the entire Tier 1 national vehicle
population over 50,000 miles is represented by only 50 vehicles, although they were
tested multiple times. In addition, the MSOD apparently does not include any second-by-
second data for NLEV or Tier 2 vehicles. When will this data be incorporated into the
MSOD? (U.S. DOT)
EPA Response: We added to the Tier 1 and NLEV dataset with the data sources in
Section 4, primarily the New York IP A dataset. Tier 2 vehicles are just starting to come
on the market, so it will likely be the next iteration of MOVES before we have in-use data
to analyze.
41. Comment: Concern with using limited dataset of diesel buses to represent heavy-duty
vehicles (EMA)
EPA Response: We added to the heavy-duty dataset with the data sources in Section 4,
primarily the WVU, CRC E-55 and CE-CERT datasets.
Remote Sensing Data
42. Comment: RSD can be helpful in "filling the gaps" for determining emitter distributions
and deterioration with age and odometer (McClintock). Only RSD captures the gross
emitters (Stedman). Remote sensing should be used if not directly, at least as a
verification step (Russell).
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES.
43. Comment: RSD sampling is not akin to 1 second on on-board data, because the vehicle
is moving relatively rapidly in the course of a 1 second RSD measurement (Ross)
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES.
Fuel Effects
44. Comment: Oxygenate is important for fuel economy (Stedman)
EPA Response: On an energy consumption basis there is no differentiation between
oxygenate use; it can be accounted for in post-processing conversion to energy or fuel.
156
-------
45. Comment: Binning fuel parameters is not a good approach. It would result in a very
high number of bins; and emission response should be continuous. A better approach is
to use continuous corrections within each bin (Russell)
EPA Response: We agree; fuel effects will be treated as a continuous function in future
versions of MOVES.
46. Comment: MOVES should add the functionality to deal with fuel properties and
reformulation issues, i.e. as the EPA Complex Model does (Russell)
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES.
47. Comment: MOVES GHG should include LPG as a fuel type (Natural Propane Gas
Association / Propane Vehicle Council)
EPA Response: LPG is included as a fuel type inMOVES2004
Methane
48. Comment: High emitters drive methane, and should be accounted for in MOVES
(Stedman). Proposed methane approach likely too simplistic, should add complexity of
CO2 or criteria pollutants (Russell).
EPA Response: We decided to take the simplistic approach for now, since there is
relatively little CH4 data available for stand-alone analysis. A more complex approach is
merited when we addHC to MOVES, to take advantage of the correlation between total
HCandCH4.
Temperature Effects
49. Comment: Low and high temperature effects should be included in the analysis (Russell)
EPA Response: We did include temperature effects, as discussed in Section 9
Model Accuracy
50. Comment: For binning analysis, CO2 should be able to be modeled within 10 percent in
average driving; the possibility of achieving anything near this level of accuracy for the
criteria pollutants is unclear (Ross). Validation results of VSP and average speed bins
presented in emission analysis plan (e.g. prediction of Bag 3 emissions within 10 percent,
prediction of most UCC cycles within 10-20 percent) were quite good (Buckingham).
157
-------
EPA Response: We met this criteria in the revised binning analysis presented in
Appendix A
51. Comment: The validation results performed by EPA for HC, CO, and NOx of the UCC
dataset (light duty vehicles only)—the best validation test of the VSP approach provided
by the EPA—indicate that there are serious concerns with using VSP and speed to
characterize HC, CO, and NOx (AAM, EMA).
EPA Response: We think the binning analysis (Appendix A) demonstrations that the
binning approach can perform well for HC, CO and NOx as well as fuel consumption.
General Emission-Related Comments
52. Comment: EPA's efforts in improving emission for MOVES should be as robust for
nonroad sources as for on-road sources (AAM, EMA)
EPA Response: This is the plan, based on large-scale data collection of nonroad data;
but the specific implementation details will need to wait until off-road is implemented in
MOVES, currently planned for 2007
53. Comment: Pre-Tier 0 vehicles should be split up to account for non-catalyst controls
implemented in the 1970's (AAM, EMA)
EPA Response: We think the binning analysis (Appendix A) demonstrations that the
binning approach can perform well for HC, CO and NOx as well as fuel consumption.
54. Comment: Will EPA consider incorporating fugitive dust in the PM emission factor
estimation process? (U.S. DOT) It would be worth considering (Replogle)
EPA Response: We will be investigating this issue as we develop the criteria pollutant
version of MOVES.
158
-------
Appendix H: Pre-Publication Peer Review Comments
Professor Lawrence Caretto of California State University Northridge was contracted to
provide formal peer review on a pre-publication version of this document. His comments are
included verbatim in this Appendix. Responses to substantive (i.e. non-editorial) comments have
been added following each comment, in italics to differentiate it from the original comments.
Editorial changes suggested by Dr. Caretto have generally been made in the final version of this
report.
159
-------
REPORT REVIEW
SEPTEMBER 13, 2004 DRAFT EPA REPORT: "MOVES2004
ENERGY AND EMISSION INPUTS"
Review Prepared for U. S. Environmental Protection Agency
Office of Transportation and Air Quality
Assessment and Standards Division
OCTOBER 17, 2004
L. S. CARETTO
7805 Cowper Avenue
West Hills, CA91304
160
-------
Introduction
The U. S. Environmental Protection Agency (EPA) is developing a new computer
model, known as MOVES, to model emissions from mobile sources. They have
prepared a series of reports describing the development of this model. This review
discusses the report "MOVES2004 Energy and Emission Inputs" that describes the data
that are inputs for the model, including the procedures used to fill missing data items.
In requesting this review,c EPA stated its primary interests as (1) report clarity, (2)
overall methodology, (3) appropriateness of the data sets selected, (4) data analyses
conducted, including the statistical approaches used, the models selected, and
appropriateness of the resulting conclusions. EPA also asked the following specific
questions:
• Are variables important to energy consumption or GHG emission production
captured, given available data?
• Has the effectiveness of the modal binning structure chosen for characterizing
emission rates been adequately demonstrated with regard to greenhouse gas
pollutants and HC, CO and NOx?
• Are methods for "hole filling" using PERE and interpolation/copying sound?
• Is the method for generating advanced technology rates using PERE sound?
The following discussion section of this review first considers the five items listed as
"primary interests" by EPA. This is followed by answers to the questions in the bulleted
list above. The final part of the discussion section contains miscellaneous comments on
the report, mainly technical questions.
For the most part, the comments in this review deal with changes to the report that can
readily be done. The comments that refer to improvements that are more exploratory in
nature or depend on data not available to EPA are identified as such when they are
made.
Letter from Chester J. France to Larry Caretto, September 16, 2004.
-------
Discussion
Scope of the report
The report that is the subject of this review describes the techniques used to provide the
data the MOVES model requires for energy consumption rates and emission rates of
C02, CH4 and N20 for a range of vehicles and operating conditions that encompass all
activities of the onroad vehicle fleet. Emission rates for C02, the main greenhouse gas,
are determined from energy consumption rates. Thus, most of the focus of the report is
on obtaining energy-rate data and making suitable approximations where such data are
missing.
The approach of the MOVES model is to assign data to bins. There are two kinds of
bins. Source bins describe the kinds of vehicles (cars, trucks, etc.) and basic vehicle
parameters used to classify their emissions performance (weight, engine displacement,
fuel type, etc.). Operating condition bins are defined in terms of vehicle specific power
(VSP) and vehicle speed. The set of operating condition bins describes all possible
emission rates (and energy use) for a vehicle in a source bin. The set of source bins,
appropriately weighted, will describe the entire vehicle fleet.
The report really deals with two separate but related topics. The first is the
development of specific input data for greenhouse gases. The second is a more
general analysis of the binning approach for all modeled quantities, and an extension of
previous work to use operating condition bins that are defined by a combination of
vehicle specific power (VSP) and vehicle speed. This second topic, which is more
fundamental to the overall development of MOVES, deals not only with greenhouse gas
elements, but also with criteria pollutants.
Where existing data do not provide information for one or more bins, the necessary bin
information is found by interpolation or by modeling using a model called PERE
(physical emission rate simulator). The approaches described here for greenhouse
gases can also be applied to some degree to other pollutants. However, the use of
PERE for modeling criteria pollutant rates is not discussed in the report being reviewed.
Comments on areas of primary interest to EPA
Report clarity
As noted in the Scope section above, the report deals with both general and specific
topics. The general topics of how to bin the existing data sets and what to do when
there are bins that have no data have implications beyond the present report. The
specific topics of obtaining the data for energy rates and greenhouse gas emissions
provide examples of how these general topics might be addressed in extending MOVES
to criteria pollutants. The introduction to the report should point out the significance of
general topics that might have applications in MOVES beyond the establishment of data
for greenhouse gas emissions.
-------
This report, like other reports on the details of emission model development, is aimed at
a limited audience with the technical background necessary to understand the
information presented. Although it potentially has reasonable clarity, the draft provided
for review appears to have been hastily assembled to meet a deadline. There are some
instances of internal authors' notes for items to be added or revised in a final draft.
There are also incomplete cross references to other parts of the text (such as "see
section 3-?") and simple errors that could be found by using the spelling and grammar
checker. Some basic editorial changes have been made to the soft copy sent to this
reviewer. This edited version has been sent to EPA as an email attachment.
The differences in colors used in some charts in the report, particularly those in
Appendix C, did not show up well in the printed version. Uses of color should be
augmented by different fill effects or other features to clarify charts for readers of a
version that is not printed in color. Figure C-9 is a good example of this practice. This
figure uses a dashed red line and a solid black line so that the differences between
these two lines clearly show up on the black-and-white copy.
Some additional comments on report clarity are included in the Miscellaneous
Comments section starting on page 169.
Overall methodology
There are two overall methodologies used in the report. The first is the modification of
earlier binning strategies to use a combination of vehicle speed and VSP to define
operating mode bins. The second is the development of the MOVES data set for
greenhouse gases. The latter requires methods for data analysis and selection, placing
data in bins, and filling missing bin data using the physical emission rate estimator
(PERE) or interpolation.
The report describes the quality control methods used for the data that are eventually
placed in bins. The procedures for data quality control appear sound, and the fact that
quality controls were used to ensure the accuracy of the data in the model is perhaps
more important than the specific details of the quality-control procedures.
The report describes procedures that were used to create data for bins that had no
available data. Appendix C of the report contains a discussion of three proposed
methods, of which two were selected. This "proof-of-concept" appendix shows the
potential errors in the processes ultimately used to provide approximate data for the
model in cases where actual data were not available. The proof-of-concept appendices
give appropriate comparisons between measured and modeled results, indicating the
level of accuracy to be expected from the model. The agreement between modeled and
actual results is usually good, showing that the methodology used here is sound.
Appropriateness of data sets selected
The data sets used in this report were appropriate. This reviewer is not aware of other
data sets, which were not examined by EPA, for consideration in this study. EPA used
appropriate quality control measures for the selection of data sets to be used and for the
selection of individual data points from those data sets.
-------
Data analyses conducted, including the statistical approaches used, the models
selected, and appropriateness of the resulting conclusions
The main data analyses conducted here are for the methods for "hole filling" obtaining
approximations for missing data bins in a model which requires data for every bin. The
basic approach of MOVES, using a combination of source bins and operating condition
bins, is based on other analyses published in previous reports and is not a point of this
review. The discussion of these analyses is covered in response to specific questions,
below.
The report also contains an analysis of fuel rates due to starts. This analysis is similar
to the one done in the development of start emission data for MOBILES.d Start
emissions for light-duty vehicles are defined as the difference between the initial 505
second cold-start portion of the federal test procedure (FTP) for light-duty vehicles and
a similar cycle run with no start, called the HR505. An initial regression analysis shows
that the HR505 results, which are not a conventional start of the FTP, are virtually the
same as the hot-start results, HS505, which are available for all FTP rests. The
regression equation for these two different types of tests is HR505 = 1.0095 HS505 -
0.002. Although this strongly suggests the near equality of the two results, it would be
helpful to add statistical tests for a zero intercept and a slope of 1. Although this
analysis was done for the light-duty FTP, there was no mention of a similar analysis for
the heavy-duty FTP. Presumably the same assumption - that start emissions were the
difference between the cold-start and hot-start portions of the heavy-duty FTP - was
made for those vehicles.
EPA Response: The statistical test suggested was performed and have been added to the report.
The lack of data for some source classes requires some significant approximations, as
described in the report. In addition, MOVES2004 does not account for different soak
times - the time between engine shut-off and restart. Presumably this will be added in
subsequent versions of MOVES.
EPA Response: soak time effects will be added to subsequent versions of MOVES
To a first-order approximation, the start energy is the energy required to increase the
average temperature of the engine and drive train from its initial value to its value at fully
warmed operation. Thus, the result of the analysis - which is to be added in the final
draft - that the start energy for hybrids depends on the engine displacement seems a
reasonable one. However, the assumption that the start energy for a fuel cell and an
electric vehicle is the same as that for a gasoline engine is not reasonable. Both of
these operate at a much lower temperature than a gasoline engine.
d Ed Glover and David Brzezinski, "Final Determination of Hot Running Emissions from FTP Bag
Emissions," Report EPA420-R-01-059, USEPA, Assessment and Standards Division, Office of
Transportation and Air Quality, November 2001.
-------
EPA Response: We will need to evaluate this when more data becomes available, particularly
on fuel cell vehicles.
The regression analyses for the effect of temperature on energy consumption for both
starts and running conditions are appropriate. However, no uncertainty analysis is
presented for the regression results for the temperature analyses. Table 9-1, which
shows the regression results for the effect of temperature on diesel start emissions, has
coefficients that do not match the results shown in equation 9-6 which is actually used
to compute the temperature effect.
EPA Response: The regression performed in Table 9-1 isn 't meant to produce the coefficients
used in the model, simply to assess the relative importance of different variables.
Specific questions
Are variables important to energy consumption or GHG emission production
captured, given available data?
Yes. The combined binning approach (source bins and operating condition bins) is
used for energy consumption which is directly related to C02 emissions. Here the
source bin variables of vehicle type, weight, fuel, and engine displacement will provide
the essential variables for determining differences from the different source types. The
operating bin boundaries should capture the variations due to vehicle operations.
Data for CH4 and N20 are limited to bag data so that operating mode bins were not
used for these species. (Presumably all bins were filled with the same average rate
shown in Appendix F for CH4 and N20.) It is possible to get second-by-second data for
these species using the method discussed in Appendix C for disaggregating bag data
into bins. This method was rejected in favor of other methods where data were
available. However, since only aggregate bag data are available for CH4 and N20, this
approach could be used if approximate binned data were required for these species.
Such a task should be reserved for future versions of the model and not considered for
immediate application.
The emission rates for CH4 and N20 in Appendix F are of the order of 1 g/hr. Emissions
of C02 are of the order of 6000 g/hr. The global warming potentials for N20 and
methane, given in Table 8-1, are, respectively, 320 and 21. This means that the error in
C02 equivalent emissions from ignoring these species completely is about 5%. The
error from assuming that the emission rate in all bins is the same should be much less
than this.
EPA Response: For future versions we are considering restructuring CH4 andN2O to be based
on ratios to HC andNOx, respectively; if this were the case then having binned emissions would
be more possible, although it isn't clear that having modal binned rates would be necessary for
these pollutant, given the relatively small contribution to total GHG emissions as noted by the
reviewer.
-------
As noted above, this report contains information on how the data for greenhouse gas
emissions were selected and a general discussion of how the overall approach for
modeling not only greenhouse gas emissions, but also criteria pollutants, would be
accomplished. The variables that affect energy consumption can be directly related to
the emissions of C02, the dominant greenhouse gas emitted from mobile source
operations. Although the variables that affect greenhouse gas emissions are captured
by the model, these emissions are relatively simple to model. The ultimate goal of
MOVES is to model not only greenhouse gases, but also criteria pollutants. Some of
the aspects of criteria pollutant modeling are discussed in this report. However, the fact
that the appropriate behavior of greenhouse gas emission is captured does not mean
that the extension to criteria pollutants can be done without some changes in the
approaches used here.
EPA Response: We plan to use the same general binning approach for criteria pollutants, but
the specific variables used to define source bins or for operating mode bins may vary depending
on the pollutant. Our goal would be to harmonize bin definitions across pollutants as much as
possible for simplicity, but will analyze each pollutant to assess how realistic this is.
Has the effectiveness of the modal binning structure chosen for characterizing
emission rates been adequately demonstrated with regard to greenhouse gas
pollutants and HC, CO and NOx?
Appendix A compares model predictions with observed results for both average data
over several trips and for individual trips. These comparisons show good agreement,
particularly for the average data. The comparison of average results for both high-
speed and low-speed conditions shows that good comparisons for the average results
are not due to cancellation of biased results at high and low speeds. Results for
individual trips show significant scatter, but this is to be expected from a model that is
intended to predict results for a vehicle fleet.
The method for obtaining trips in this analysis is not clear. It is easy to see how trips
can be determined from the onboard data sets, but how were trips determined from
cycle data?
EPA Response: For the light-duty dataset, trips were defined by each test cycle (for the lab
data) or each key on/ key off (for the on-boarddata). For the heavy-duty dataset, trips were
defined as key on/ key off for the trailer data and for buses, we simply defined all operation as a
single trip they were generally idling between excursions.
The use of one set of trips to derive the binned data for the model, and a separate set of
trips for comparison with the model results, is a good approach to give confidence in the
model results.
The HC emissions data in Appendix A apparently are for exhaust HC only. How will
evaporative emissions be handled in MOVES? If an approach to evaporative emissions
is currently under development, it would be interesting to add a brief statement to the
treatment of such emissions to this appendix.
-------
EPA Response: Evaporative emissions are beyond the scope of this report, but will be
considered for the criteria pollutant version of MOVES.
Are methods for "hole filling" using
PERE and interpolation/copying sound?
The approach to hole filling is discussed in section 4.3.3 which is supported by
appendices C and D. Appendix C provides a comparison of three possible methods.
Two of these - calculations from the physical emission rate simulator (PERE) and
interpolation - were actually used to obtain data for MOVES. Scaling of bag data was
rejected as being less accurate. Section 11.1 provides a set of codes that are used to
identify the sources of binned information in MOVES: actual data, PERE results, or a
variety of interpolation/copying processes. (Copying data from neighboring bins was
done in cases where PERE was not run and it was not possible to find appropriate data
for interpolation or extrapolation.)
Obviously, one would like to use data instead of approximations; however, the binning
approach selected for MOVES would require a large amount of data. Table D-9 shows
that there are 11,322 combinations of source bins and operating condition bins for
vehicles manufactured prior to 2010. Only 2,634 (23%) of these combinations were
filled with data. The remaining bins were determined by PERE (1,275 or 11 %) or by
interpolation/copying (7,413 or 65%). However, the number of bins that are filled by
PERE and interpolation/copying shows the need for a sound practice for filling these
empty bins.
EPA Response: it is worth noting that although the actual number of bins filled with data is
relatively low, the proportion of the fleet covered is much higher, as presented in Table 4-6.
The proof-of-concept discussion in Appendix C shows reasonable agreement between
procedures used for hole filling and actual data. It is likely that the data available for
bins comes from measurements on vehicles and operating conditions that comprise a
large fraction of actual driving. Thus, the hole filling may represent a fraction of overall
vehicle and driving conditions that is less than 65%.
As noted above, there are basically two kinds of bins used in MOVES: source bins and
operating condition bins. The discussion of hole-filling approaches is focused on
missing source categories. Apparently, there is no use of interpolation among operating
condition bins. This seems reasonable since the results shown for these bins are highly
nonlinear. It would be helpful to include a statement that no interpolation was done
across operating condition bins, if this is in fact the case.
Some improvements are obviously possible, but this would be a longer term effort that
could be done as MOVES is expanded to include criteria pollutants. The results from
Figure D-3 - showing the mean energy consumption rate as a function of vehicle weight
- are a good example of interpolated data that could be improved. The interpolated
data between points "0330" and "0800" would extrapolate to a negative energy rate for
weights below 18,000 pounds. The PERE result that the lower weight point "0160" has
a lower energy rate than the higher weight point "0330" seems unrealistic. There is no
-------
information on the number of data points that go into the "0330" data cell. A future hole-
filling approach could try to provide some combination of statistical and modeling results
to fill holes with data that are more realistic physically. (It is not apparent how such a
method might work, but it would be worth exploring for future MOVES hole filling.)
EPA Response: This example also illustrates uncertainty that can be introduced into the
imputation by the combination of data sources. In this case, the value for weightClasses 0160
and 0800 were generated by PERE, whereas that for 0330 was generated from data. We expect
the meanBaseRate to follow a positive linear trend with weight. In this case, however, apparent
scatter around the trend results in a decreasing local trend between 0160 and 0330, followed by
a steep increasing trend between 0330 and 0800.
Although Figure 14 on page 117 compares PERE rates with extrapolation and scaled
bag data, there did not appear to be any attempt to use a combination of PERE with
interpolation/extrapolation to fill a single hole. This combination could perhaps yield a
better result than either used separately. Such an approach is intended to be used in
regions that are relatively rich with data so that only a few holes have to be filled. This
should be contrasted with the process in data-poor areas where it appears that PERE
was used to fill a few holes and interpolation/extrapolation was then used based on
PERE results instead of actual data.
EPA Response: Indeed a combination of extrapolation/interpolation and PERE would probably
provide a smoother transition between the points. There are several reasons why this was not
done: 1) It is very labor intensive; 2) PERE is already calibrated to a number of vehicle types
and changing the calibration so that it matches results from these test programs, could throw it
off on others. A fixed calibration was maintained throughout for consistency. 3) It is possible
that the data is more limited than the model in that only a single driving cycle is included in the
test program(s) that generated the rates for that particular source bin. In this case, it is very
possible that the IM240 driving schedule is the only one defining the rates for several source
bins - or at least dominating the mean value. PERE runs a number of cycles so that the higher
power bins are filled, which the IM240 cycle may not.
There is a shift in the details of the "interpolation" between the proof of concept in
Appendix C and the actual application in Appendix D. Appendix D uses linear
interpolation of two data points for the data actually used in MOVES. The proof of
concept in Appendix C develops a linear regression equation from four data points.
This gives a small difference between the method used in the proof of concept and the
method used to actually get the data.6
EPA Response: It is true the methodologies shifted between the initial assessment in Appendix C
and actual application in Appendix D. As noted we would expect a small difference in the
results.
e In numerical analysis, interpolation usually refers to methods in which an n order polynomial is used to
fit n+1 data points exactly, as is done in Appendix D.
-------
The development of PERE is the subject of a previous report not under review here.
These comments refer simply to its use for the purpose of filling missing data in the bins
structure of MOVES.
Is the method for generating advanced
technology rates using PERE sound?
Regulatory mobile source emission models face the challenge of providing emission
results for vehicles that have not yet been built. These results are required for analyses
used to support state implementation plans and transportation conformity analyses. In
MOBILES and earlier models, the emissions of future vehicles, under actual operation,
were scaled based on the future emission standards. This approach assumes that the
technology for meeting future emission standards will be similar to the technology used
for current vehicles. This assumption seems more tenuous for advanced technology
vehicles and provides the justification for the use of a model such as PERE.
The background for using the PERE model for advanced technology vehicles is
provided in another report/ That report provides some comparisons of modeled fuel
consumption results with measured results for fuel economy over the federal test
procedure (FTP) cycle. These results show good agreement for conventional vehicles
and some early advanced technology vehicles including hybrids and one fuel cell
vehicle.
The approach for using PERE to model advanced technology vehicles appears sound.
Of course, when such vehicles do enter the fleet, their actual data can be used in
MOVES to replace the modeled results. An important data element, which is not part of
this report, is a good estimate of the fleet distribution of advanced technology vehicles in
future years.
Miscellaneous Comments
The comments in this section are classified as technical comments, report clarity
comments, and general comments.
Technical comments
The finite-difference equations for acceleration shown on pages 95 and 96, at = vt - VM,
use backward differences. A more accurate approach would be to use central
differences, computing at = vt+i - VM. The definition of acceleration used in the report
would be a central difference if it referred to the acceleration at the midpoint: at+i/2 = vt -
vt-i; this definition implies that the calculation of vehicle-specific power would use the
midpoint velocity, vt+i/2 = (at + at-i) for consistency.
The assumption of the same heating value for gaseous and liquid hydrogen in Table B-
14, pp 97-98, will lead to a small error in the fuel consumption. Less energy is available
for combustion because of the need to heat and evaporate the liquid hydrogen.
f Edward Nam, "Advanced Technology Vehicle Modeling in PERE," USEPA report 420-D-04-002,
Assessment and Standards Division, Office of Transportation and Air Quality, March 2004.
-------
Figure C-5, page 107, is called a linear fit of engine size bins and the text says that this
linear regression is done using Excel. However, the plot shows curves, not linear, fits to
the points. This appears to be due to a plotting error in which the Excel plot option of a
smoothed curve was selected rather than simply plotting the points and a linear trend
line.
The factors relating the relative global warming potential of different greenhouse gases
shown in Table 8-1 appear to be those for a 100-year time frame. Is this correct? Can
users substitute global warming potentials for a different time frame selected?
EPA Response: We did use 100 year time frame GWPs, in accordance with IPCC guidance.
Users can changes these values in the MOVESDefault database; the MOVES2004 Software
Design Reference Manual has more details on how to accomplish this.
Starting in Table D-1, page 120, the energy rate is stated in units of kJ/SHO. However,
the meaning of SHO is never defined. The energy rate data in this table are of the order
of 105 kJ/SHO, compared to the expected values (of the order of 10 kJ/s) shown in
other parts of the report. On page 15 the energy rate units are listed as kJ/hr. This is
less intuitive than kJ/s which is the same as kW. The energy input in kJ/s should be the
same as the engine power output in kW divided by the engine efficiency.
EPA Response: SHO stands for source hours operating, e.g. KJ per hour. Activity is expressed
in terms of hours, so it is easiest to express energy and emission rates per hour as well.
Comment on report clarity
Both simple classifications of bin boundaries and numeric program ID codes for bins are
used in the report. For the ease of the reader, data should be presented in conventional
terms. In tables 4-12 and 4-13, for example, a conventional description is used for the
fuel type and the model year. However, program codes are used for engine size and for
weight. When the code structure is first introduced in Table 4-2, there should be a
reference to Appendix B in which the individual parts of the bin ID code are specified.
General comment
During the development of MOBILES, EPA made data sets used in the generation of
data for that model available to users who could then check the parameters in the
model. Does EPA intend to do so with the data for MOVES? If so, a statement could
be added to the report giving readers directions for obtaining the data files.
EPA Response: All data used to generate the energy and emission rates in MOVES will be made
available, through request to mobile (q),epa. gov.
-------
References
1 Mobile Source Observation Data User Guide and Reference, U.S. EPA Office of
Transportation and Air Quality, EPA Report EPA-420-B-04-004, February 2004 (available
online at http://www.epa.gov/otaq/models/msod/420b04004.pdf)
2 Gerber, W. and P. Henson, Mobile Source Observation Data (MSOD) Database Update,
Report prepared for U.S. EPA Office of Transportation and Air Quality by Eastern Research
Group, Inc., December 2002 (available online at
http://www.epa.gov/otaq/models/ngm/r02033.pdf)
3 Gerber, W. and P. Henson, Mobile Source Observation Data (MSOD) Database Update, Report
prepared for U.S. EPA Office of Transportation and Air Quality by Eastern Research Group,
Inc., April 2003
4 Limsakul, B., Mobile Source Observation Data (MSOD) Database Update, Report prepared
for U.S. EPA Office of Transportation and Air Quality by Eastern Research Group, Inc.,
August 2003
5 User Manual for VALDATA Data Checking Software for MSOD Database ,EPA 2C-S026-
NTSA, Britton Information Systems, Inc., November 2002
6 User Manual for EFLOADData Loading Software for MSOD Database , EPA 2C-S026-
NTSA, Britton Information Systems, Inc., November 2002
7 DeFries, T., W. Gerber, and S. Kishan, Determination of Important Parameters for CO2 and
CH4 Emission Factor Modeling, Report prepared for U. S. EPA Office of Transportation and
Air Quality by Eastern Research Group, Inc., November 2001 (available online at
http://www.epa.gov/otaq/ngm.htm
8 Koupal, J. and J. Kremer, Air Conditioning Correction Factors inMOBILE6, U.S. EPA
Office of Transportation and Air Quality, EPA Report No. EPA420-R-01-055, November 2001
(available online at http://www.epa.gov/otaq/models/mobile6/r01055.pdf)
9 Carlson. T. and T. Austin, Development of Speed Correction Cycles, Report prepared for U.S.
EPA Office of Transportation and Air Quality by Sierra Research, Inc, MOBILE6 Report No.
M6.SPD.001, 1997 (available at http://www.epa.gov/otaq/models/mobile6/m6spd001.pdf)
10
Warila, I.E., E. Glover, J. Koupal and R. Giannelli. Running Energy Consumption Rates
within the MOVES Modal Framework. U.S. EPA Office of Transportation and Air Quality,
Proceedings of 14th CRC On-road Vehicle Emissions Workshop, San Diego, CA, March 29-
31, 2004. Coordinating Research Council, Alpharetta, GA.
-------
11 Nam, E. and R. Giannelli, Fuel Consumption Modeling of Conventional and Advanced
Technology Vehicles in the Physical Emission Rate Estimator (PERE) , U.S. EPA Office of
Transportation and Air Quality, EPA Report No. EPA420-P-05-001 February 2005
12 Chon, D. and J. Heywood, Performance Scaling of Spark-Ignition Engines: Correlation and
Historical Analysis of Production Engine Data, SAE 2000-01-0565, 2000
13 Wang, M., GREET 1.5 Transportation Fuel Cycle Model Volume 1: Methodology,
Development, Use and Results, Argonne National Laboratory Center for Transportation
Research, ANL/ESD-39, Vol 1. August 1999 (available online at
http://www.transportation.anl.gov/software/GREET/publications.html)
14 Lim, H. Study of Exhaust Emissions and Idling Heavy Duty Diesel Trucks and Commercially
Available Idle Reducing Devices, SAE 2003-01-0290, September 2002
15 Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990 - 2001, U.S. EPA Office of
Transportation and Air Quality, EPA Report Number 430-R 03-004, April 2003 (available
online at:
http://yosemite.epa.gov/oar/globalwarming.nsf/content/ResourceCenterPublicationsGHGEmiss
ionsUSEmissionsInventory2003.html)
16 Heywood, J. Internal Combustion Engine Fundamentals, McGraw-Hill, Inc., New York, New
York, 1988
17 Direct and Indirect Emissions from Mobile Combustion Sources, EPA Climate Leaders
Greenhouse Gas Inventory Protocol, December 2003. (available online at
http://www.epa.gov/climateleaders/pdf/mobilesourceguidance.pdf)
18 Browning, L. Update of Methane and Nitrous Oxide Emission Factors for On-Highway
Vehicles, Report prepared for U.S. EPA Office of Transportation and Air Quality by ICF
Consulting , EPA Report No. 420-P-04-016, February 2004 (available online at
http://www.epa.gov/otaq/models/ngm.htm)
19 See reference note 15
20 See reference note 15 (Annex N)
21 Koupal, J., Air Conditioning Activity Effects inMOBILE6, U.S. EPA Office of Transportation
and Air Quality, EPA Report No. EPA420-R-01-054, November 2001 (available online at
http://www.epa.gov/otaq/models/mobile6/r01054.pdf)
22 See reference note 8
23 Hart, C., J. Koupal and R. Giannelli, EPA 's Onboard Emissions Analysis Shootout:
-------
Overview and Results, U.S. EPA Office of Transportation and Air Quality, EPA Report No.
EPA420-R-02-026, October 2002 (available online at
http://www.epa.gov/otaq/models/ngm/r02026.pdf).
24Frey, C., Unal, A., Chen, J., Li, S., Xuan, C., Methodology for Developing Modal Emission
Rates for EPA 's Multi-Scale Motor Vehicle & Equipment Emission System, Report prepared
for U.S. EPA Office of Transportation and Air Quality by North Carolina State University
Computational Laboratory for Energy, Air and Risk, August 2002 (available online at
http://www.epa.gov/otaq/models/ngm/r02027.pdf)
25Koupal, J. Emission Analysis Approach for EPA 's Multi-scale Motor Vehicle & Equipment
Emission System (MOVES), U.S. EPA Office of Transportation and Air Quality, Proceedings
of 13th CRC On-road Vehicle Emissions Workshop, San Diego, CA, April 7-9, 2003.
Coordinating Research Council, Alpharetta, GA.
26 Nam, E. SpeedAnomolies in VSP Based Modeling, Ford Motor Company, Proceedings of 13th
CRC On-road Vehicle Emissions Workshop, San Diego, CA, April 7-9, 2003. Coordinating
Research Council, Alpharetta, GA.
27
Petrushov, V.A. Coast Down Method in Time-Distance Variables, SAE Paper No. 970408,
1997
28 See reference note 1
29 Joy, R. and J. Lee, I/MLookup Table Update. Report prepared for U.S. EPA Office of
Transportation and Air Quality by Sierra Research, Inc, October 2000 (available online at
http://www.epa.gov/otaq/regs/im/fmalrpt.pdf)
30 See reference note 27
31 Barm, M., G. Scora and T. Younglove. A Modal Emission Model for Heavy Duty Diesel
Vehicles. Transportation Research Board Annual Meeting, January, 2004. Washington, D.C.
32 Gillespie, T.D. Fundamentals of Vehicle Dynamics. Society of Automotive Engineers.
Warrendale, Pennsylvania, 1992
33 Kish, L. Survey Sampling. John Wiley & Sons, New York, New York, 1965
34 IM240 andEvap Technical Guidance, U.S. EPA Office of Transportation and Air Quality,
EPA Report No. 420-R-00-007, April 2000 (available online at
http://www.epa.gov/otaq/regs/im/r00007.pdf)
------- |