Exhaust Emission Rates for Light-Duty
Onroad Vehicles in MOVES4
£% United States
Environmental Protect
Agency
-------
Exhaust Emission Rates for Light-Duty
Onroad Vehicles in MOVES4
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 the exchange of technical information and to
inform the public of technical developments.
Assessment and Standards Division
Office of Transportation and Air Quality
U.S. Environmental Protection Agency
NOTICE
4>EPA
United States
Environmental Protection
Agency
EPA-420-R-23-028
August 2023
-------
Table of Contents
1 Introduction 13
1.1 Organization 13
2 Background 15
2.1 Development of the Motor Vehicle Emissions Simulator (MOVES) 15
2.1.1 Light-Duty Vehicles 16
2.2 Emissions Sources (sourceBinID) and Processes (polProcessID) 17
2.2.1 The emissionRateByAge Table 18
2.2.1.1 Age Groups (ageGroupID) 19
2.3 Exhaust Emissions for Running Operation 20
2.3.1 Operating Modes (opModelD) 20
2.4 Exhaust Emissions for Start Operation 21
2.4.1 Operating Modes for Start Emissions 22
2.4.2 Adjustments to Start Emissions 23
3 Gaseous Exhaust Emissions from Light-Duty Gasoline Vehicles (THC, CO, NOx) 23
3.1 Approach 23
3.2 Emission-Rate development: (Model years 1989-and-earlier) 24
3.2.1 Data Sources 24
3.2.1.1 Vehicle Descriptors 24
3.2.1.1.1 Track Road-Load Coefficients: Light-Duty Vehicles 24
3.2.1.1.2 Test Descriptors 25
3.2.1.1.3 Candidate Data Sources 26
3.2.1.2 Data Processing and Quality-Assurance 28
3.2.1.3 Sample-design reconstruction (Phoenix only) 29
3.2.1.4 Data Source Selection 30
3.2.2 Methods 31
3.2.2.1 Data-Driven Rates 31
3.2.2.1.1 Rates: Calculation of Weighted Means 31
3.2.2.1.2 Estimation of Uncertainties for Cell Means: 32
3.2.2.1.3 Model-generated Rates (hole-filling) 33
3.2.2.1.3.1 Rates 34
2
-------
3.2.2.1.3.1.1 Coast/Cruise/ Acceleration 35
3.2.2.1.3.1.2 Braking/Deceleration 39
3.2.2.1.3.2 Estimation of Model Uncertainties 40
3.2.2.1.3.3 Reverse Transformation 41
3.2.2.1.4 Table Construction 41
3.2.2.2 Adjustment for High-Power Operating modes 42
3.2.2.3 Stabilization of Emissions with Age 52
3.2.2.3.1 Non-I/M Reference Rates 57
3.3 MOVES2014 Emission-Rate Development (MY 2001-2016) 58
3.3.1 Data Sources 59
3.3.1.1 Vehicle Descriptors 59
3.3.2 Estimating Reference Rates 60
3.3.2.1 Averaging IUVP Results 60
3.3.2.2 Develop Phase-In Assumptions 65
3.3.2.3 Merge FTP Results and Phase-In Assumptions 69
3.3.2.4 Estimating Emissions by Operating Mode 75
3.3.2.4.1 Evaluation of MOVES2014 "High-Power" Emission Rates 78
3.4 MOVES2014 Emission-Rate Development (MY 2017 and later) 80
3.4.1 Averaging FTP Results (Step 1) 81
3.4.2 Develop Tier 3 Phase-In Assumptions (Step 2) 83
3.4.3 Merge Cycle Results and Phase-In Assumptions (Step 3) 87
3.4.4 Estimating Emissions by Operating Mode (Step 4) 91
3.4.5 Apply Deterioration (Step 5) 95
3.4.5.1 Recalculate the Logarithmic Mean 95
3.4.5.2 Apply a Logarithmic Age Slope 95
3.4.5.3 Apply the Reverse Transformation 95
3.4.5.4 Adjust to Account for Averaging with Electric Vehicles 96
3.4.6 Estimate Non-I/M References (Step 6) 96
3.4.7 Start Emissions 96
3.5 Estimating Rates for Non-I/M Areas 96
3.6 MOVES3 Running Exhaust Emission Rates (THC and NOx for MY 1990 and later)105
3.6.1 Data Source 105
3.6.2 Vehicle classes 106
3
-------
3.6.2.1 Clean Screen 108
3.6.3 DataReview 110
3.6.3.1 Oxides of Nitrogen (NOx) 110
3.6.3.2 Total Hydrocarbons (THC) 113
3.6.4 Model structure 116
3.6.4.1 Optimizing the Assignment of Knots 118
3.6.5 Model Results 121
3.6.5.1 Oxides of Nitrogen (NOx) 121
3.6.5.2 Total Hydrocarbons (THC) 126
3.6.6 Reverse Transformation 130
3.6.7 "Young Vehicle" Adjustments 137
3.6.7.1 Adjustments for NOx 137
3.6.7.1.1 Cars 137
3.6.7.1.2 Trucks 138
3.6.7.2 Calculating NOx Adjustments 141
3.6.7.3 Adjustments for THC 142
3.6.7.3.1 Cars at Age 2 142
3.6.7.3.2 Trucks at Age 2 143
3.6.7.4 Calculating THC Adjustments 145
3.6.8 Deterioration Adjustments 146
3.6.8.1 Running Process for NOx 146
3.6.8.2 Running Process for THC 149
3.7 Running Exhaust Emission Rates (CO for MY 1990 and Later) 151
3.7.1 Data Source 151
3.7.2 Vehicle Classes 152
3.7.3 DataReview 153
3.7.4 Model Structure 154
3.7.4.1 Optimizing Assignment of Knots 154
3.7.5 Model Results 158
3.7.6 Reverse Transformation 161
3.7.6.1 Translation from Fuel to Distance Bases 162
3.7.7 "Young Vehicle Adjustments" 163
3.7.7.1 Calculating Adjustments 164
4
-------
3.7.8 Deterioration Adjustments 165
3.7.8.1 Running Process for CO 165
3.8 Estimation of Emission Rates for Cold Starts 167
3.8.1 Subgroup 1: Vehicles manufactured in model year 1995 and earlier 167
3.8.1.1 Data Sources 167
3.8.1.2 Defining Start Emissions 168
3.8.2 Subgroup 2: Vehicles manufactured in MY1996 and later 169
3.9 Estimation of Emission Rates for Hot to Warm Starts 170
3.9.1 Subgroup 1: Model Years 2003 and earlier 170
3.9.1.1 Relationship between Soak Time and Start Emissions 170
3.9.2 Subgroup 2: Model Years 2004 and Later 171
3.9.2.1 Measuring Start Emissions using PEMS 171
3.9.2.2 Measuring Soak-time Relationships on the Dynamometer 176
3.9.2.3 Comparing Dynamometer and PEMS Measurements 176
3.9.3 Applying Deterioration to Starts 184
3.9.3.1 Assessing Start Deterioration in Relation to Running Deterioration 184
3.9.3.2 Translation from Mileage to Age Basis (MY 1989 and earlier) 197
3.9.3.3 Translation from Mileage to Age Basis (MY 1990 and later) 199
3.9.3.3.1 Start Process for NOx 199
3.9.3.3.2 Start Process for THC 200
3.9.3.3.3 Start Process for CO 201
3.10 Constructing Updated Rates (Model Years 1990 and Later) 202
3.10.1 Step 1: Extract LD gasoline rates from the Input database 202
3.10.2 Step 2: Apply Young-vehicle Adjustments to Running Rates 203
3.10.3 Step 3: Apply Deterioration Adjustments 203
3.10.4 Step 4: Apply Non-IM Ratios 203
3.10.5 Step 5: Replicate Rates for Additional Fuel Types 203
3.11 Final Results for Update for MOVES3 203
3.11.1 Trends with Vehicle-Specific Power 204
3.11.2 Trends with Soak Time 208
3.11.3 Trends with Age 212
3.11.4 Trends with I/M Status 216
3.12 Development of Emission Rates representing California Standards 220
5
-------
3.12.1 Averaging IUVP Results 222
3.12.2 Develop Phase-In assumptions 222
3.12.3 Merge FTP Results and Phase-In Assumptions 224
3.12.4 Scaling CA/177 Rates to Federal Rates 227
3.12.5 Extrapolating Phase-in Trends 233
3.12.6 Additional Steps 237
3.12.6.1 Apply Deterioration Adjustments 238
3.12.6.2 Apply Non-I/M ratios 238
3.12.6.3 Replicate Rates for additional Fuels 238
3.12.7 Availability 238
3.12.8 Early Adoption of National LEV Standards 238
3.13 Rates for E-85 Vehicles 239
4 Particulate-Matter Emissions from Light-Duty Gasoline Vehicles 239
4.1 Particulate-Matter Emission Rates for Model Year 2004 and Earlier Vehicles 240
4.1.1 Particulate Measurement in the Kansas City Study 240
4.1.2 New Vehicle or Zero Mile Level (ZML) Emission Rates 244
4.1.2.1 Longitudinal Studies 245
4.1.2.2 New Vehicle, or ZML Emission Rates and Cycle Effects 246
4.1.2.3 Aging or Deterioration in Emission Rates 251
4.1.3 Estimating Elemental Carbon Fractions 254
4.1.4 Modal Running Emission Rates 257
4.1.5 Modal Start Emission Rates 260
4.2 Particulate-Matter Emission Rates for Model Year 2004 and Later Vehicles 261
4.2.1 Introduction 261
4.2.1.1 Dataset Description 261
4.2.1.2 Fuel Considerations 262
4.2.2 Calculating FTP and LA92 Cycle Rates Using MOVES Emission Rates 264
4.2.3 Estimating Start Emissions for Particulate Matter 265
4.2.3.1 Start Emissions for Vehicles with Port Fuel Injection (PFI) 265
4.2.3.2 Start Emissions for Vehicles with Gasoline Direct Injection (GDI) 266
4.2.4 Estimating Running Emissions for Particulate Matter (PM) 267
4.2.4.1 Running Emissions for Vehicles with Port Fuel Inj ection (PFI) 267
4.2.4.2 Running Emissions for Vehicles with Gasoline Direct Injection (GDI) 268
6
-------
4.2.5 Developing Base Emission Rates for Model Year 2004 and Later 269
4.2.5.1 Additional Assumptions Used to Determine GDI Truck Scaling Factors 270
4.2.5.2 EC/NonECPM Fractions 271
4.2.6 Calculation of Fleet-Average PM Emission Rates by Model Year, Vehicle Age, and
PM component 271
4.2.6.1 Vehicle Population Data for Model Years 2004 and Later 272
4.2.6.2 Calculating Rates by Model Year, Vehicle Age, and PM Component 272
4.2.7 Incorporating Tier 3 Emissions Standards for Particulate Emissions 275
4.2.7.1 Apply Phase-in Assumptions 275
4.2.7.2 Apply Scaling Fractions 276
4.2.7.3 Simulate the Extended Useful Life 279
4.2.8 Incorporating the LEV-III Standard for Particulate Matter 281
4.3 Light-Duty PM Emission Rates Trends 281
5 Gaseous and Particulate Emissions from Light-Duty Diesel and Electric Vehicles (THC,
CO, NOx, PM) 284
5.1 Light Duty Diesel 284
5.2 Light Duty Electric Vehicles 285
6 Ammonia Emissions from Light-duty Vehicles 286
6.1 Light Duty Gasoline 286
6.1.1 Light-duty Model Year 1960 to 1980 Vehicles 286
6.1.2 Model Year 1981 and Later Vehicles 287
6.1.2.1 Remote Sensing Data 287
6.1.2.2 Average Fuel-based Emission Rates by Model Year Groups 289
6.1.2.3 Average Fuel-based Emission Rates by Model Year Group and Vehicle Age
291
6.1.2.4 Mass Rates by Operating Mode 296
6.1.3 Motorcycles 299
6.2 Light Duty E85 Vehicles 300
6.3 Light Duty Diesel Vehicles 300
6.4 Summary 303
7 Crankcase Emissions 304
7.1 B ackground 304
7.2 Modeling Crankcase Emissions in MOVES 304
7
-------
7.3 Light-duty Gasoline and E-85 Crankcase Emissions 304
7.4 Motorcycle Crankcase Emissions 305
7.5 Light-duty Diesel Crankcase Emissions 305
8 Nitrogen Oxide Composition 306
8.1 Light-Duty Gasoline Vehicles 306
8.2 Motorcycles 307
8.3 Light-duty Diesel Vehicles 307
9 Appendix A. Revisions to the Pre-2004 Model Year PM2.5 Emission Rates between
MOVES2010b and MOVES2014 308
10 References 309
8
-------
List of Acronyms
ALVW Adjusted Loaded Vehicle Weight
CAA Clean Air Act
CAFE Continuous Atlanta Fleet Evaluation
CARB California Air Resources Board
CA/S177 Clean Air Act Section 177 California LEV-II emissions standard
CD Compliance Division
CIC conventional internal combustion
CO carbon monoxide
CO2 carbon dioxide
CV coefficients of variation
CVS constant-volume sampler
DRI Desert Research Institute
DT DusTrak analyzer (measures mass and size fraction of PM)
EC elemental carbon
ECPM elemental carbon particulate matter
EGR exhaust-gas recirculation
EPA U.S. Environmental Protection Agency
EPAct Energy Policy Act of 2005
°F degrees Fahrenheit
FTP Federal Test Procedure
FUL full useful life
GDI Gasoline Direct Injection engines
g/hr Grams per hour
g/km Grams per kilometer
g/mi Grams per mile
g/sec Grams per second
g/SHO grams per source-hours operating
GVWR Gross Vehicle Weight Rating
HC Hydrocarbons
HLDT Heavy Light-Duty Truck
HNO2 nitrous acid (HONO)
9
-------
HP
horsepower
I/M
Inspection and Maintenance program
IM147
Inspection and Maintenance transient loaded-mode emissions test
IM240
Inspection and Maintenance transient loaded-mode emissions test
IUVP
In-Use Verification Program
KCVES
Kansas City Light-Duty Vehicle Emissions Study
kW
kilowatt
LA92
Unified driving schedule
lb
pound
LDT
Light-Duty Truck
LDT1
Light-Duty Truck LVW = 3750 lbs.
LDT2
Light-Duty Truck LVW > 3750 lbs.
LDT3
Light-Duty Truck ALVW =5750 lbs.
LDT4
Light-Duty Truck ALVW > 5750 lbs.
LDV
Light-Duty Vehicle
LEV
Low Emission Vehicle
LEV-I
California PC, LDT and MDV exhaust emission standards of 1990
LEV-II
California PC, LDT and MDV exhaust emission standards of 1999
LEV-III
California PC, LDT and MDPV exhaust emission standards of 2012
LLDT
Light Light-Duty Truck
LVW
Loaded Vehicle Weight
MDPV
Medium-Duty Passenger Vehicle
Mg
mass weight
MOBILE6
EPA Highway Vehicle Emission Factor Model, Version 6
MOVES
Motor Vehicle Emission Simulator
mpg
miles per gallon
MY
model year
NAAQS
National Ambient Air-Quality Standards
NLEV
National Low Emission Vehicle
NMHC
Non-Methane Hydrocarbons
NMOG
non-methane organic gases
NO*
oxides of nitrogen (NO + NO2 + HONO)
NTR
Northeast Trading Region
10
-------
NYIPA
New York Instrumentation/Protocol Assessment
OBD
On-Board Diagnostic (System)
OC
Organic Compounds/Organic Carbon
ORD
Office of Research and Development
OTAQ
Office of Transportation and Air Quality
PA
photoacoustic analyzer
PC
passenger cars
PCV
positive crankcase ventilation
PEMS
portable emissions measurement system
PFI
port fuel injection
PM
Particulate Matter
PM2.5
particulate matter <2.5 microns in diameter
PM10
particulate matter <10 microns in diameter
QCM
Quartz-crystal microbalance
RSD
Remote Sensing Device
RSE
relative standard error
SC03
SFTP chassis dynamometer test
SEMTECH-D
vehicle emissions measurement instrument produced by Sensors, Inc.
SFTP
Supplemental Federal Test Procedure
SHO
source-hours operating
SPSS
IBM Statistical Package for the Social Sciences software
SULEV
Super Ultra Low Emission Vehicle
SwRI
Southwest Research Institute
TC
Total Carbon
THC
Total Hydrocarbon (by flame ionization detector)
Tier 1
vehicle emissions certification standards phased in from 1994 - 1997
Tier 2
vehicle emissions certification standards phased in from 2004 - 2009
Tier 3
vehicle emissions certification standards phased in from 2017 - 2025
TLEV
Transitional Low Emission Vehicle
TOR
Thermal Optical Reflectance
TRLP
track road load power
UDDS
Urban Dynamometer Driving Schedule
ULEV
Ultra Low Emission Vehicle
11
-------
US06 Test cycle comprising one component of the Supplemental Federal Test
Procedure (SFTP)
VIN Vehicle Identification Number
VSP vehicle specific power
ZML zero-mile emission level
12
-------
1 Introduction
The United States Environmental Protection Agency's Motor Vehicle Emission Simulator
(MOVES) is a set of modeling tools for estimating air pollution emissions produced by onroad
(highway) and nonroad mobile sources. MOVES estimates the emissions of greenhouse gases
(GHGs), criteria pollutants and selected air toxics. The MOVES model is currently the official
model for use for state implementation plan (SIP) submissions to EPA and for transportation
conformity analyses outside of California. The model is also the primary modeling tool for
estimating the impact of mobile-source regulations on emission inventories.
This report describes the analyses conducted to generate exhaust emission for light-duty vehicles
in MOVES. Light-duty vehicles in MOVES are defined as any vehicle with a Gross Vehicle
Weight Rating (GVWR) up to 8,500 lbs. This report describes the development of ammonia
(NH3) emission rates for motorcycles and describes how motorcycle crankcase emissions and
nitrogen oxide speciation are handled in MOVES, but most information on MOVES motorcycle
emissions is described in a separate report.8
Emission rates for THC, CO, NOx, PM2.5 and NH3 are stored in the "EmissionRateByAge" table
in the MOVES database according to the following:
• Pollutant
• Emission process
• Fuel type
• Regulatory class
• Model year group
• Operating mode
• Vehicle age
Energy consumption and nitrous oxide rates for light-duty cars, light-duty trucks and
motorcycles, including electric cars and light trucks, are documented in a separate report.1
Particulate matter emissions from brake and tire wear are also documented separately.82
Unlike earlier versions of MOVES, MOVES4 includes electric vehicles in the default fleet2 and
accounts for projected increases in the THC and NOx emissions from conventional light-duty
vehicles assuming manufacturers take advantage of the fleet-wide averaging allowed by EPA
regulations. These adjustments are explained in the MOVES4 adjustments report.3
MOVES4 also includes updates to ammonia emission rates and the allocation of nitrogen species
for light-duty diesel vehicles as detailed in Sections 6 and 8.
1.1 Organization
This report is divided into eight chapters, including this introduction.
At the outset, Chapter 2 gives some background and history for the Motor Vehicle Emissions
Simulator (MOVES) model. It also introduces concepts and structures common to the various
emissions estimated by the model. Specifically, it defines the concepts "emissions source,"
"regulatory class," "emissions process" and "operating mode."
13
-------
Chapter 3 is by far the longest in the report. It describes the data and methods used to develop
light-duty gasoline emission rates for the "gaseous emissions," defined to include total
hydrocarbons (THC), carbon monoxide (CO) and oxides of nitrogen (NOx).
Sections 3.2, 3.3, and 3.4 describe the development of emission rates for "hot-running" vehicle
operation in MOVES2014. This MOVES2014 material has been retained because the MOVES3
rates were generated by applying adjustments to the MOVES2014 rates (which, in turn, were
based on rates developed for MOVES2010). The content in these sections is largely unchanged
from the corresponding sections in the MOVES2014 report. The exception to this rule concerns
the revisions to emission rates in "high-power operating modes," the basis for which is described
in 3.3.2.4.
Section 3.5 lays the foundation for how MOVES3 accounts for the effects of "Inspection and
Maintenance" (I/M) programs. It describes the data and methods used to develop the
proportional differences between default "I/M" and "non-I/M" base rates. This material is
identical to that in the MOVES2014 report, although it has been reorganized to be more
independent in relation to the sections that precede and follow it, as its applicability in the
emission rates is broad.
The most prominent and far-reaching change in the updates to emission rates in MOVES3 is the
reevaluation and modification of emissions deterioration. Sections 3.6 and 3.7 describe the
underlying data and analyses for NOx, THC and CO. New analysis of recently acquired large
data sets from the Denver Metropolitan Area was used to develop broad models of deterioration
covering 20 model years over 20+ years of age and to calculate adjustment ratios to apply to the
MOVES2014 rates.
The next two sections cover the development of emission rates for start operation. The
development of rates representing "cold-start" operation is covered in Section 3.8. The
development of rates representing "warm" or "hot" engines is presented in Section 3.9.
Important revisions in the estimation of these start emissions for "Tier-2" vehicles, i.e., vehicles
in model years 2004 and later, are presented in 3.9.2.
Section 3.10 describes specific steps taken to derive revised rates for MOVES3 by applying
adjustments to emission rates from MOVES2014. These revisions were applied to rates
representing model years 1990 and later. This section also covers the steps followed in
generating the full set of revised light-duty gasoline emission rates for MOVES3.
Section 3.11 presents selected results in which revised rates for MOVES3 are compared to each
other and to their MOVES2014 counterparts. Trends in emissions in relation to important
variables including power, age and time since key-off are included.
The rates presented in Sections 3.3 to 3.11 describe rates representing emissions from vehicles
compliant with "Federal" standards, i.e., standards developed and promulgated at the Federal
level by the US EPA. Section 3.12 describes development of a corresponding set of rates that
represent emissions from vehicles compliant with standards developed and promulgated by the
State of California and additional states that have adopted "California" emissions standards at
some point in the past 25 years.
Chapter 4 covers the development of emission rates for particulate matter (PM). In the emission
rates analysis, particulate matter is defined as particles < 2.5 microns in diameter (PM2.5). An
important update since MOVES2014 is the development of rates accounting for the expected
14
-------
transition from "port fuel injection" to "gasoline direct injection" between 2005 and 2030. These
analyses introduce recent data that supplements the "Kansas-City Vehicle Emissions Study" that
provided the sole basis for PM emission rates in MOVES2014.
Chapter 5 briefly describes how light-duty emission rates are assigned to represent fuels other
than gasoline, including diesel and "high-level" ethanol blends, i.e., E85. In addition, it
discusses briefly how MOVES3 treats hybrid and electric vehicles.
Chapter 6 documents the development of emission rates for ammonia (NH3). These rates were
updated for MOVES4.
Chapter 7 discusses how MOVES estimates crankcase emissions by relating them to tailpipe
emissions for running and start operation. We assume that the majority of light-duty vehicles
with properly functioning positive crankcase ventilation (PCV) have no crankcase emissions.
However, crankcase emissions are estimated for small fractions of vehicles that either lack PCV
or have malfunctioning PCV.
Chapter 8 describes the partition of nitrogen oxides and related compounds. The values for
diesel vehicles were updated for MOVES4.
2 Background
2.1 Development of the Motor Vehicle Emissions Simulator (MOVES)
MOVES calculates emission inventories by multiplying emission rates by the appropriate
emission-related activity, applying correction and adjustment factors as needed to simulate
specific situations and then summing the emissions from all sources and regions.
Thus, the material presented in this document is a component of a much larger effort, including
the estimation of emission rates for heavy-duty vehicles, estimation of evaporative emissions,
estimation of usage and activity patterns for vehicles, estimation of adjustments that account for
fuel parameters, ambient temperature and humidity, air conditioning effects and the impact of
various Inspection and Maintenance program designs, the compilation and storage of all types of
input data in the MOVES database, and the algorithms that combine and process input
information during model runs, translating inputs and modeling assumptions into inventory
estimates.
Readers not familiar with MOVES may find it useful to access additional documentation
providing a broader view of MOVES, the rationale for its development as a replacement for
MOBILE6, and broad overviews of its design.
• The "Initial ProposciF for MOVES describes the impetus behind the effort to design a
new inventory model from the ground up, with the goal of developing a tool both more
comprehensive and flexible than its predecessor.4
• A subsequent "Draft Design and Implementation Plan" describes the MOVES design and
introduces the reader to concepts and terminology developed for the new model.5
15
-------
• Readers wishing to further understand the development of the modal design for running
emissions can consult the "Methodology for Developing Modal Emission Rates, "6 as well
as the "Shoot Out"1 conducted among several candidate approaches.
A large volume of additional documentation and supporting materials can be obtained at
https://www.epa.gov/moves/moves-onroad-technical-reports. In general, the most recent and
relevant materials are at the top of the page, with older material further down. However, as the
previous references show, references posted throughout the page are still relevant to the MOVES
model and database in its most recent versions.
2.1.1 Light-Duty Vehicles
Light-duty vehicles are defined as cars and trucks with gross vehicle weight ratings (GVWR) of
less than 8,500 lbs. For purposes of emissions standards, "cars" are designated as "LDV" or
"passenger cars" (PC), and are distinguished from "light-duty trucks" (LDT) which are further
sub-classified as "light light-duty trucks" (LLDT) and "heavy light-duty trucks" (HLDT), on the
basis of GVWR < 6000 lbs. and GVWR > 6000 lbs., respectively. The two broad classes, LLDT
and HLDT, are further subdivided into LDT1/LDT2, and LDT3/LDT4. As these subdivisions are
highly specific and technical, we do not describe them here. Interested readers can find more
information at http://www.epa.gov/otaq/standards/weights.htm. As MOVES pools all light-duty
truck classes for purposes of inventory estimation, we will refer to "cars" and "trucks"
throughout. The development of emission rates for motorcycles are covered in a separate report.8
Exhaust emissions from light-duty vehicles have contributed substantially to urban air pollution,
and have received a great deal of scientific, political and regulatory attention over the past fifty
years. The Clean Air Act (CAA), passed in 1970 (and amended in 1977 and 1990), set "National
Ambient Air-Quality Standards" (NAAQS) for carbon monoxide (CO), lead (Pb), nitrogen
dioxide (NO2), ozone (O3) particulate matter (PM), and sulfur dioxide (SO2). The CAA provides
authority to the EPA to set emission standards for CO to help achieve the CO NAAQS, and for
THC and NO* largely for their roles in production of ground-level ozone. Regulations designed
to reduce automobile emissions to facilitate achievement of compliance with the NAAQS
include Tier-1 standards introduced in the mid 1990's, followed by National Low-Emission
Vehicle (NLEV) standards starting in 2001, Tier 2 standards starting in 2004, and Tier 3
standards starting in 2017. Concurrently, the state of California and additional states electing to
adopt "California" in lieu of "Federal" standards have implemented "LEV-I," "LEV-IE" and
"LEV-III" standards.
In addition to introducing more stringent tailpipe standards, requiring introduction of oxygenated
gasolines, and modifying test procedures, the 1990 CAA Amendments expanded requirements
for Inspection-and-Maintenance programs (I/M). The role played by I/M programs in many
urban areas over the past twenty years means that accounting for the existence of such programs
is an important consideration in modeling tailpipe emissions from light-duty vehicles.
Through a combination of regulation and improved technology, gaseous tailpipe emissions from
light-duty vehicles have declined substantially over the past several decades. Important
milestones in engine and emissions control technology have included the introduction of fuel
injection (replacing carburetion), positive crankcase ventilation (PCV), exhaust gas recirculation
(EGR), catalytic converters, electronic engine controls, on-board diagnostic systems (OBD) and
gasoline direct injection (GDI). Development of emission rates thus largely involves constructing
16
-------
a "quantitative" account of this history. A detailed account of these developments is beyond the
scope of this document which will focus on the development of emission rates as inputs to the
MOVES model. However, this history has been well described elsewhere, and we refer
interested readers to the USEPA website9 and to the peer-reviewed literature.101112131415161718
2.2 Emissions Sources (sourceBinID) and Processes (polProcessID)
In MOVES terminology, pollutants are emitted by "sources" via one or more "processes."
Within processes, emissions may vary by operating mode, as well as by age group. The relevant
gaseous criteria pollutants include: total hydrocarbons (THC), carbon monoxide (CO) and oxides
of nitrogen (MX). Relevant particulate criteria pollutants include elemental and organic carbon.
MOVES estimates other organic gas aggregations by ratio relative to THC emissions, including
methane (CH4), non-methane hydrocarbons (NMHC), volatile organic compounds (VOC), total
organic gases (TOG), and non-methane organic gases (NMOG). The definitions and methods for
estimating these other organic gas aggregations are documented in the MOVES speciation
report.19 THC emissions are intended to include all hydrocarbon emissions and are operationally
defined as measurements taken by flame ionization detector. In this report we also use the term
hydrocarbons (HC) emissions to refer specifically to emissions of compounds of carbon and
hydrogen (regardless of the measurement method), and more generally, to other measures of
organic gases that are primarily composed of hydrocarbons, including NMHC and NMOG.
The relevant processes are exhaust emissions emitted during engine start and running processes,
i.e., "exhaust start" and "exhaust running." Combinations of pollutant and process relevant to
this chapter are shown in Table 2-1. For start emissions, the meanBaseRate is expressed in units
of g/start, and for running emissions, the meanBaseRate is expressed in units of g/hr, which
MOVES terminology designates more specifically as "g/SHO," where SHO denotes "source-
hours operating."
Table 2-1 Combinations of pollutants and processes for gaseous pollutant emissions
pollutantName
pollutantID
processName
processID
polProcessID
THC
1
Running exhaust
1
101
Start exhaust
2
102
CO
2
Running exhaust
1
201
Start exhaust
2
202
NOx
3
Running exhaust
1
301
Start exhaust
2
302
Elemental Carbon
112
Running exhaust
1
11201
(ECPM)
Start exhaust
2
11202
Non-elemental Carbon
118
Running exhaust
1
11801
(non-ECPM)
Start exhaust
2
11802
17
-------
Note that this document describes only emission rates for exhaust hydrocarbons. Modeling of
emission rates for evaporative hydrocarbons is described in a separate report.20
Vehicle classes as "emissions sources" are described by a label known as the "sourceBinID".
The identifier is a 19-digit numeric label, of the form "IfftteeyysssswwwwOO" as described in
Table 2-2. Note that the engine-size and weight-class attributes are not used to classify vehicles
in MOVES3.
Table 2-2 Construction of sourceBins for exhaust emissions for light-duly vehicles
Parameter
MOVES Database Attribute1
Values
Fuel type
fuelTypelD
Gasoline =01
Diesel = 02
E85 = 05
Engine Technology
engtechid
01= "Conventional internal
Combustion"
Regulatory Class
regClassID
20 = "Car" (LDV)
30 = "Truck" (LDT)
Model-Year group
shortModY rGroupID
Varies2
Engine Size Class
engSizelD
Vehicle Test Weight
weightClassID
1 as used in the database table "emissionRateByAge."
2 as defined in the database table "modelYearGroup."
As an example, Table 2-3 shows the construction of sourceBin labels for light-duty gasoline
vehicles, manufactured in model years 1998 and 2010.
Table 2-3 Examples of sourceBinID construction for cars and trucks in model years 1998 and 2010
fuelTypelD
engTechID
regClassID
shortModY rGroupID
sourceBinID
1 (Gasoline)
1 (conventional)
20 (Car)
98 (MY 1998)
1 01 01 20 98 0000 0000 00
1
1
30 (Truck)
30 (MY 1998)
1 01 01 30 98 0000 0000 00
1
1
20 (Car)
98 (MY 2010)
1 01 01 20 30 0000 0000 00
1
1
30 (Truck)
30 (MY 2010)
1 01 01 30 30 0000 0000 00
2.2.1 The emissionRateByAge Table
The rates described in this document are stored in the MOVES emissionRateByAge table. This
table includes five fields, as shown in Table 2-4. 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 table includes rates for vehicles inside and outside of
Inspection-and-Maintenance areas. The uncertainty estimates, when present, are expressed as
coefficients of variation for the mean (meanBaseRateCV); this term is synonymous with the
18
-------
"relative standard error (RSE). In this section, we will describe the processes of data
classification by source bin and operating mode, calculation of mean emission rates, and
statistical evaluation of the results.
2.2.1.1 Age Groups (ageGroupID)
To account for emissions deterioration, MOVES estimates emission rates for vehicles in a series
of age ranges, identified as "age groups" (ageGroupID). Seven groups are used, as follows: 0-3,
4-5, 6-7, 8-9, 10-14, 15-19, and 20+ years. The values of the attribute ageGroupID for these
classes are 3, 405, 607, 809, 1014, 1519, and 2099, respectively. The resolution of these groups
is finest between 4 and 9 years of age, when the emission deterioration curves are steepest.
Table 2-4 Description of the EmissionRateByAge table
Field
Description
SourceBinID
Source Bin identifier. See Table 2-2
and Table 2-3.
PolProcessID
Combines pollutant and process. See
Table 2-1.
opModelD
Operating mode: defined separately
for running and start emissions. See
Table 2-5.
ageGroupID
Indicates age range for specific
emission rates.
meanBaseRate
Mean emission rates in areas not
influenced by inspection and
maintenance programs.
meanBaseRateCV
Coefficient of variation of the cell
mean (relative standard error, RSE),
for the meanBaseRate.
meanBaseRatelM
Mean emission rate in areas subject
to an I/M program with features
similar to the reference program.
meanBaseRatelM
CV
Coefficient of variation of the cell
mean (relative standard error, RSE),
for the meanBaseRatelM.
dataSourcelD
Numeric label indicating the data
source(s) and method(s) used to
develop specific rates.
19
-------
2.3 Exhaust Emissions for Running Operation
Running operation is defined as operation of internal-combustion engines after the engine and
emission control systems have stabilized at operating temperature, i.e., "hot-stabilized"
operation.
2.3.1 Operating Modes (opModelD)
For running emissions, the key concept underlying the definition of operating modes is "vehicle-
specific power" (VSP). This parameter represents the tractive power exerted by a vehicle to
move itself and its cargo or passengers.21 It is estimated in terms of a vehicle's speed and mass,
as shown in Equation 2-1.
VSPAv> t Eli t CA my'(a' + flsln w) Equation 2-1
t m
In this form, VSP at time t (kW/Mg) is estimated in terms of vehicles':
• speed at time t (vt, m/sec),
• acceleration at, defined as v, - v,-i, (m/sec2)
• road grade, where sin(6t) = fractional road grade at time /, and g is the acceleration
due to gravity (9.8 m/se°2), mass m (Mg) (usually referred to as "weight,")
• track-road load coefficients A, B and C, representing rolling resistance, rotational
resistance and aerodynamic drag, in units of kW-sec/m, kW-sec2/m2 and kW-sec3/m3,
respectively.3
For purposes of the data used in this analysis, the grade is assumed to be zero because the
vehicles were measured on chassis dynamometers. Note that during model operation, MOVES
accounts for grade when characterizing vehicle activity only in project-scale mode.
On the basis of VSP, speed and acceleration, a total of 23 operating modes are defined for the
running-exhaust process (Table 2-5). Aside from deceleration/braking, which is defined in terms
of acceleration, and idle, which is defined in terms of speed alone, the remaining 21 modes are
defined in terms of VSP within broad speed classes. Two of the modes represent "coasting,"
where VSP < 0, and the remainder represent "cruise/acceleration," with VSP ranging from 0 to
over 30 kW/Mg. For reference, each mode is identified by a numeric label, the "opModelD." In
cases where the deceleration/braking definition overlaps with other operating modes, the
deceleration/braking categorization takes precedence over other definitions.
20
-------
Table 2-5 Definition of MOVES operating modes for running-exhaust operation
opModelD
Description
Vehicle
Vehicle-
Vehicle Acceleration (at, mi/hr-
Speed
Specific
sec)
(vt, mi/hr)
Power (VSPr)
0
Deceleration/
Braking
Idle
-1.0 < vt <
1.0a
cit<-2.0d OR
(cit< -1.0e & at-i < -1.0 & cit-i < ¦
m
11
Coast
1 < v, < 25b
VSPf< 0
12
Cruise/Acceleration
0 < VSPf < 3
13
3 < VSPf < 6
14
6 < VSPf < 9
15
9 < VSPf < 12
16
VSPf < 12
21
Coast
25
-------
the engine and emissions control systems will be "cold," i.e., near ambient temperature.
Operationally, we define cold start as a start after the engine has been off ("soaked") for 12 hours
or more. The engine and catalyst heat up fairly quickly when the vehicle is driving, but if the
catalyst is cool, it will not effectively control emissions. In addition, to start the engine, it is
necessary to inject "excess" fuel into the cylinder to provide enough flammable vapor to ignite
when the spark plug fires. Incomplete combustion of this fuel, in addition to emissions control
systems being below operating temperature, yields a bolus of "excess" emissions during a brief
"start period" following key-on events. These emissions are referred to as "start emissions," in
contrast to the "hot running" emissions discussed in section 2.3.
Emission rates for start emissions are expressed as mass emitted for a single start event following
key on (mass/start).
In MOVES, start emissions for light-duty vehicles are defined in terms of the Federal Test
Procedure (FTP). The cycle includes three phases, or "bags," which are intended to represent,
"cold-start", "hot-running" and "hot-start" emissions, respectively. The first, or "cold-start"
phase, is 505 seconds (8.42 min.) in duration. The second, or "hot-running" phase is 867
seconds long. Following the second phase, the engine is turned off, and allowed to "soak" for 10
min., after which the engine is restarted and the third "hot-start" phase is performed, repeating
the first-phase driving cycle. To estimate true "cold-start" emissions, the mass emitted during
the third phase is subtracted from that emitted during the first phase, as described in more detail
below.
2.4.1 Operating Modes for Start Emissions
The "cold-start," as defined above, is represented as opModeID=108. An additional seven modes
are defined in terms of soak times ranging from 3 min up to 540 min (opModelD = 101-107), as
shown in Table 2-6.
22
-------
Table 2-6 Operating-mode definitions for start emissions, defined in terms of soak time
Nominal Soak Period
(min)
OpModelD
OpModeName
3
101
Soak Time < 6 minutes
18
102
6 minutes < Soak Time < 30 minutes
45
103
30 minutes < Soak Time < 60 minutes
75
104
60 minutes < Soak Time < 90 minutes
105
105
90 minutes < Soak Time <120 minutes
240
106
120 minutes < Soak Time < 360 minutes
540
107
360 minutes < Soak Time < 720 minutes
720
108
720 minutes < Soak Time
2.4.2 Adjustments to Start Emissions
Note that all discussion in this section applies to start conditions under "warm ambient"
conditions, i.e., for temperatures above 68°F. For start emissions at colder temperatures,
MOVES applies a separate "temperature adjustment." Note that the development and application
of temperature adjustments is discussed in a separate report.3 Start emissions are also adjusted to
account for fuel characteristics as explained in the Fuel Effects Report.22
3 Gaseous Exhaust Emissions from Light-Duty Gasoline Vehicles
(THC, CO, NO*)
This chapter describes the technical development of emission rates for gaseous exhaust
pollutants for light-duty vehicles. These pollutants include total hydrocarbons (THC), carbon
monoxide (CO) and oxides of nitrogen (NO*). The resulting model inputs are stored in the
emissionRateByAge table included in the MOVES input database. Ammonia emission rates are
described separately in Section 6.
3.1 Approach
In estimation of the regulated gaseous pollutants, it is essential to know with confidence whether
vehicles had been subject to inspection-and-maintenance (I/M) requirements at or previous to the
time of measurement. After reviewing data sources, it became clear that the volumes of data
collected within I/M areas vastly exceeded those collected in non-I/M areas. We also concluded
that I/M programs themselves could provide large and valuable sources of data. In consideration
of the demanding analytic tasks posed by the ambitious MOVES design, we elected to estimate
rates for vehicles in I/M areas first, as the "base-line" or "default" condition. Following
construction of a set of rates representing I/M "reference" conditions, the plan was to estimate
rates for non-I/M areas relative to those in I/M areas.
23
-------
In addition, the rates described below represent emissions on the FTP temperature range (68 - 86
°F) to provide a baseline against which temperature adjustments would be applied during model
runs.
3.2 Emission-Rate development: (Model years 1989-and-earlier)
MOVES3 updates light-duty gaseous emission rates for model years 1990-and-later, but the rates
for 1989-and-earlier are unchanged from MOVES2010 and MOVES2014.
3.2.1 Data Sources
For emissions data to be eligible for use in this analysis, several requirements were imposed:
• To derive rates for operating modes, it was essential to acquire data measured on
transient tests.
• Data had to be measured at a frequency of approximately 1 Hz, or higher, e.g.,
continuous or "second-by-second" measurements.
• To make allowance for application of temperature adjustments (developed
separately), it was necessary to know the ambient temperature at the time of test.
Vehicles were subject to I/M program requirements at the time of measurement.
3.2.1.1 Vehicle Descriptors
In addition to the requirements listed above, complete descriptive information for vehicles was
required. Vehicle parameters required for incorporation into MOVES are shown in Table 3-1.
Table 3-1 Required vehicle parameters
Parameter
Units
Purpose
VIN
Verify MY or other parameters
Fuel type
Distinguish gasoline vehicles
Make
Distinguish cars and light trucks
Model
Distinguish cars and light trucks
Model year
Assign sourceBinID, calculate age-at-test
Vehicle class
Assign sourceBinID
GVWR
lb
Distinguish trucks from cars (LDV)
Track road-load power
hp
Calculate track road-load coefficients A. B and C
3.2.1.1.1 Track Road-Load Coefficients: Light-Duty Vehicles
For light-duty vehicles, we calculated the track load coefficients from the "track road load power
at 50 mph" (TRLP, hp), based on Equation 3-1.23
24
-------
A = PFa.
f TRLHP • c '
B = PFn
C = PFn
V50~C2
TRLHP • Cj
(V50-C2)2
TRLHP • Cj
(V50 * ^2 )3
Equation 3-1
where:
PF i = default power fraction for coefficients 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),
c\ = a constant, converting TRLP from hp to kW (0.74570 kW/hp),
vso = 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
TRLP were obtained from the Sierra I/M Look-up Table.24
3.2.1.1.2 Test Descriptors
In addition, a set of descriptive information was required for sets of emissions measurements on
specific vehicles. Essential items for use in rate development are listed in Table 3-2.
Table 3-2 Required test parameters
Parameter
Units
Purpose
Date
Determine vehicle age at test
Time of day
Establish sequence of replicate tests
Ambient temperature
°F
Identify tests in target temperature range
Test Number
Identify 1st and subsequent replicates
Test duration
sec
Verify full-duration of tests
Test result
pass/fail
Assign tests correctly to pass or fail categories
Test weight
lb
Calculate vehicle-specific power
25
-------
3.2.1.1.3 Candidate Data Sources
In addition to the parameters listed in Table 3-1 and Table 3-2, datasets with historic depth and
large sample sizes were highly desirable to characterize the high variability typical of exhaust
emissions as well as trends with vehicle age.
When this analysis was conducted for MOVES2010 (2005-2008), a large volume of emissions
data was available, representing over 500,000 vehicles when taken together (Table 3-3). In some
cases, they could be combined as broadly comparable pairs representing I/M and non-I/M
conditions. While not all available data could receive detailed attention, due to limitations in
time and resources, a selection of likely candidates was subjected to a high degree of scrutiny
and quality-assurance, after which some were excluded from further consideration for specific
reasons.
Table 3-3 Datasets available for use in estimating running emissions from cars and trucks
Dynamometer
I/M
non-I/M
AZ (Phoenix)
IL (Chicago)
MO (St. Louis)
NY (New York)
Remote-Sensing
I/M
non-I/M
AZ (Phoenix)
IL (Chicago)
MO (St. Louis)
Maryland/N Virginia
VA (Richmond)
GA (Atlanta)
GA (Augusta/Macon)
NE (Omaha)
OK (Tulsa)
Several remote-sensing datasets received consideration. However, we elected not to use remote-
sensing data directly to estimate MY 1989-and-earlier rates, for several reasons: (1) For the most
part, at the time of the analysis, the remote-sensing datasets on hand had very restricted model-
year by age coverage (historic depth), which severely limited their usefulness in assigning
deterioration. (2) The measurement of hydrocarbons by remote sensing is highly uncertain. The
instruments are known to underestimate the concentrations of many hydrocarbon species relative
to other techniques, such as flame-ionization detectors. In inventory estimation, a multiplicative
adjustment of 2.0-2.2 is often applied to allow comparison to THC measurements by other
methods.25 (3) In MOVES, emissions are expressed in terms of mass rates (mass/time). While
fuel-specific rates (mass emissions/mass fuel) can be estimated readily from remote-sensing
data,26 mass rates cannot be calculated without an independently estimated CO2 mass rate. It
followed that remote-sensing would not provide rates for any MYxAge combinations where
dynamometer data were not available. In these cases, remote-sensing would be dependent on and
to some extent redundant with dynamometer data. (4) Because remote-sensing measurements are
typically sited to catch vehicles operating under light to moderate acceleration, results can
26
-------
describe emissions only in selected cruise/acceleration operating modes. However, remote-
sensing cannot provide measurements for coasting, deceleration/braking or idle modes.a
Table 3-4 Characteristics of candidate datasets
Criterion
Chicago
Phoenix
NYIPA
St. Louis
Type
Enhanced
Enhanced
Basic/Enhanced
Enhanced
Network
Centralized
Centralized
De-centralized
Centralized
Exempt MY
4 most recent
4 most recent
2 most recent
2 most recent
Collects random
sample?
YES
YES
n/a
NO
Program Tests
Idle, IM240, OBD-
II
Idle/SS, IM240,
IM147, OBD-II
IM240
IM240
Fast-pass/Fast-fail?
YES
YES
n/a
YES
Test type (for
random sample)
IM240
IM240, IM147
IM240
n/a
Available CY
2000-2004
1995-1999
2002-2005
1999-2002
2002-2005
Size (no. tests)
8,900
62,500
8,100
2,200,000
Dynamometer datasets that received serious consideration are described below and summarized
in Table 3-4.
Metropolitan Chicago. We acquired data collected over four calendar years (2000-04) in
Chicago's centralized enhanced program. In addition to routine program tests, the program
performed IM240 tests on two random vehicle samples. One is the "back-to-back" random
sample. This sample is relatively small (n ~ 9,000 tests), but valuable because each selected
vehicle received two full-duration IM240 tests in rapid succession, obviating concerns about
conditioning prior to conduction of IM240 tests. A second is the "full-duration" random sample,
in which selected vehicles received a single full-duration IM240. This sample is much larger (n >
800,000) but less valuable due to the lack of replication. Despite its size, the full-duration
sample has no more historic depth than the back-to-back sample, and thus sheds little additional
light on age trends in emissions. Both samples were presumably simple random samples,
indicating that in the use of the data, users must assume that the samples are self-weighting with
respect to characteristics such as high emissions, passing/failing test results, etc.
St. Louis. Another large program dataset is available from the program in St. Louis. While a
large sample of program tests is available, this program differed from the others in that no
random evaluation sample was available. Because vehicles were allowed to "fast-pass" their
routine tests, results contained many partial duration tests (31 - 240 seconds). At the same time,
a For revisons included in MOVES3, we have utilized remote sensing data in developing the CO
emissions for model year 1990 later vehicles as discussed in Section 3.7, and for estimating
ammonia emissions (see Section 6). In these cases, the benefits of using remote sensing data
outweighed the limitations.
27
-------
the lack of replication raised concerns about conditioning. Partial duration was a concern in itself
in that the representation of passing vehicles declined with increasing test duration, and also
because it compounded the issue of conditioning. In addition, while OBD-equipped vehicles
failing a scan received IM240s, those passing their scans did not. Because addressing the
interwoven issues of inadequate conditioning, "fast-pass bias" and "OBD-screening bias" proved
intractable, we excluded this dataset from further consideration.
Phoenix. At the outset, the random samples from the Phoenix program appeared attractive in
that they had over twice the historic depth of any other dataset, with model-year x age coverage
spanning 11 calendar years. Usage of these samples is somewhat complicated by the fact that no
random samples were collected for two years (2000-01) and by the fact that the sample design
employed changed in the middle of the ten-year period. During the first four years, a simple "2
percent random sample" was employed. During the last four years, a stratified design was
introduced which sampled passing and failing vehicles independently and at different rates. In
the stratified sample, failures were over-sampled relative to passing vehicles. Thus, using these
data to estimate representative rates and to combine them with the 2 percent sample, assumed to
be self-weighting, required reconstruction of the actual stratified sampling rates, as described
below.
New York Instrumentation/Protocol Assessment (NYIPA). This dataset differs from the others in
that while it was collected within an I/M area in New York City, it is not an I/M program dataset
as such. It is, rather, a large-scale research program designed to establish correlation between the
IM240 and an alternative transient test. It is not entirely clear whether it can be considered a
random sample, in part because estimation of representative averages was not a primary goal of
the study. All data that we accessed and used was measured on full-duration IM240s during a
four-year period. There was a high degree of replication in the conduction of tests, allowing
fully-conditioned operation to be isolated by exclusion of the initial test in a series of replicates.
While these data played a prominent role in development of energy consumption rates for
MOVES2004, the four-year duration of the program limits its usefulness in analysis of age
trends for gaseous pollutants.
3.2.1.2 Data Processing and Quality-Assurance
We performed several quality-assurance steps to avoid known biases and issues in using I/M data
to estimate mean emissions. One source of error, "inadequate conditioning" can occur when
vehicles idle for long periods while waiting in line. To ensure that measurements used reflected
fully-conditioned vehicles we excluded either portions of tests or entire tests, depending on test
type and the availability of replicates. If back-to-back replication was performed, we discarded
the first test in a series of replicates. If replication was not performed, we excluded the first 120
seconds of tests (for IM240s only).
Another problem occurs when calculation of fuel economy for tests yields values implausible
enough to indicate that measurements of one or more exhaust constituents are invalid (e.g., 300
mpg). To identify and exclude such tests, we identified tests with outlying measurements for
fuel economy, after grouping vehicles by vehicle make, model-year and displacement.
An issue in some continuous or second-by-second datasets is that cases occur in which the
emissions time-series appears to be "frozen" or saturated at some level, not responding to
28
-------
changes in power. We found that the occurrence of such problems was more or less evenly
distributed among the fleet regardless of age or model year, and that severe instances were rare.
We excluded tests in which 25 percent or more of the measurements were "frozen."
For a modal analysis assuming that emissions respond to power on short time scales, it is critical
that the emissions time-series be aligned to the power time-series. Consequently, we examined
alignment for all tests. As necessary, we re-aligned emissions time series to those for VSP by
maximizing correlation coefficients, using parametric Pearson coefficients for CO2 and NOx, and
non-parametric Spearman coefficients for CO and THC. For these two species, the trends with
respect to VSP were not linear, nor were distributions of emissions close to normal at any VSP
level. Consequently, we concluded that the Spearman coefficients, as measures of association,
rather than linear correlation, performed as well or better than Pearson coefficients for CO and
THC.
3.2.1.3 Sample-design reconstruction (Phoenix only)
For data collected in Phoenix during CY 2002-05, we constructed sampling weights to allow use
of the tests to develop representative means. The program implemented a stratified sampling
strategy, in which failing vehicles were sampled at higher rates than passing vehicles.
It is thus necessary to reconstruct the sample design to appropriately weight failing and passing
vehicles in subsequent analyses. After selection into the random sample, vehicles were assigned
to the "failing" or "passing" strata based on the result of their routine program test, with the
specific test depending on model year, as shown in Figure 3-1. Within both strata, sample
vehicles then received three replicate IM147 tests.
Based on test records, reconstructing sampling rates simply involved dividing the numbers of
sampled vehicles by the total numbers of vehicles tested, by model year and calendar year, for
failing (f) and passing (p) strata, as shown in Equation 3-2.
r _ Wf,MY,CY /¦ _ p,MY,C Y ^ ^
/i;my,cy ~~ /p,my,cy ~~ Equation 3-2
f.MY.CY p,MY,C Y
Corresponding sampling weights indicate the numbers of vehicles in the general fleet represented
by each sample vehicle. They were derived as the reciprocals of the sampling fractions, as shown
in Equation 3-3.
wf,MY,CY - r wp,my,cy - Equation 3-3
J f,MY,CY ./p,MY,CY
29
-------
Figure 3-1 Stratified sampling as applied in selection of the random evaluation sample in the Phoenix I/M
Program (CY 2002-05)
3.2.1.4 Data Source Selection
After excluding the St. Louis dataset, and comparing the Phoenix, Chicago and NY datasets, we
elected to rely on the Phoenix dataset for purposes of rate estimation and to use the other
datasets, including selected remote-sensing data, for purposes of comparison. This course was
chosen for several reasons.
For our purposes, the greater historic depth of the Phoenix data was a tremendous advantage. It
was the only set deep enough to allow direct and independent assessment of deterioration. The
limited depth of the other datasets would have meant that the subset of calendar years that could
be covered by pooled data would have been relatively limited. Only a single calendar year,
2002, is covered by all three datasets. Several years would be covered by two out of three.
Calendar 1999 is covered by Phoenix and NY; 2000 and 2001 would have been covered by NY
and Chicago, and 2003 and 2004 by Chicago and Phoenix. The remaining years, 1996-98 and
2005 could have been covered only by Phoenix in any case.
In addition, pooling the three datasets would have raised several difficult technical issues that
may not be apparent at first glance. Table 3-4 shows that the datasets were of greatly differing
sizes. Thus, if the datasets were pooled without some type of relative weighting, Phoenix would
have exerted much stronger influence than the others in most shared calendar years. To rectify
disparities in influence by assigning the different datasets similar or proportional influence would
have required development of some sort of a weighting scheme, but a rational basis for such
relative weighting is not immediately apparent.
The question of pooling is further complicated by the fact that use of the Phoenix data collected
in CY 2002 to 2005 requires use of sampling weights for passing and failing tests (as described)
above), whereas the Chicago and NYIPA datasets are assumed to be self-weighting. Again, no
rational basis for incorporating weighted and self-weighted tests from various programs in the
same CY was immediately apparent.
30
-------
Finally, the selection of the Phoenix data provided a relatively consistent basis for specification
of a "reference fuel," and development of associated fuel adjustments. 22
3.2.2 Methods
3.2.2.1 Data-Driven Rates
Where data was present, the approach was simple. We calculated means and other summary
statistics for each combination of sourceBinID, ageGroup and operating mode (i.e., table cell).
We classified the data by regulatory class (LDV="cars", LDT="trucks"), model-year group, age
group and operating mode (Table 2-5). The model-year groups used are shown in Table 3-5,
along with corresponding samples of passing and failing tests. Note that the analysis for 1990-
and-later model years was substantially updated as explained in Section 3.6 and Section 3.7
below.
Table 3-5 Test sample sizes for the Phoenix random evaluation sample (n = no. tests)
Model-year
group1
Cars
Trucks
fail2
pass
fail
pass
1981-82
562
539
340
495
1983-85
1,776
2,078
1,124
1,606
1980-89
3,542
6,420
1,745
3,698
1990-93
2,897
8,457
1,152
4,629
1994-95
997
4,422
703
3,668
1996-98
1,330
3,773
1996
526
1,196
1997-98
858
2,320
1999-2000
176
753
136
624
Total
11,285
26,478
6,589
18,254
1 Note that these are the model-year groups used for analysis; NOT the
model-year groups used in the MOVES database.
2 Note that 'failure' can indicate failure for CO, THC or NOr, as applicable.
We calculated means and other summary statistics for each combination of sourceBinID,
ageGroupID and opModelD. For simplicity, we will refer to a specific combination of
sourceBinID, and opModelD as a "cell" to be denoted by label '/?'.
3.2.2.1.1 Rates: Calculation of Weighted Means
The emission rate (meanBaseRate) in each cell is a (Eh) simple weighted mean
31
-------
Equation 3-4
YjW'
where Wi is a sampling weight for each vehicle in the cell, as described above, and Ri,t is the
"second-by-second" emission rate in the cell for a given vehicle at a given second t.
In the emissionRateByAge table, uncertainties for individual rates are stored in the
"meanBaseRateCV" fields (Table 2-4). To estimate sampling error for each cell, we calculated
standard-errors by weighted variance components. 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-test variability.27 To enable estimation of variances under this approach, we
calculated a set of summary statistics, as listed below:
Test mean (Et): the arithmetic mean of all measurements in a given test on a specific vehicle in a
given cell.
Test sample size (///,), the number of individual tests represented in a cell.
Measurement sample size (//,): the number of measurements in a cell representing an individual
test on an individual vehicle.
Cell sample size (m,i): the total number of individual measurements on all vehicles in a cell,
where each count represents a measurement collected at an approximate frequency of 1.0 Hz,
(i.e., "second-by-second").
2
Test variance (st ): the variance of measurements for each 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 test variance for each test in each cell.
2
Weighted Between-Test variance component (Sb ): the component of total variance due to
variability among tests in a cell, or stated differently, the weighted variance of the test means
about the cell mean, calculated as
3.2.2.1.2
Estimation of Uncertainties for Cell Means:
32
-------
nl
2>, -i
= —— Equation 3-5
i=1
2
Weighted Within-Test Variance Component (Sw ): the variance component due to variability
within tests, or the variance of measurements within individual tests (Ri,t) about their respective
test means, calculated in terms of the test variances, weighted and summed over all tests in the
cell:
si =
Y.w, ("y-ik2
7=1
Equation 3-6
h ^ . .
YjW, \nh,,-nh)
1=1
Variance of the cell mean (s|): 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
Sg = —H—— Equation 3-7
nH nKi
Coefficient-of-Variation of the Mean (CVa): 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
[P~
CV = * Eh Equation 3-8
Eh F
^h
Note that the term CVa is synonymous with the term "relative standard error" (RSE).
3.2.2.1.3 Model-generated Rates (hole-filling)
Following averaging of the data, it was necessary to impute rates for cells for which no data was
available, i.e., "holes." With respect to vehicle age, empty cells occur for age Groups not
covered by available data. As shown in Figure 3-2, "age holes" are represented by un-shaded
areas. Filling in these un-shaded areas required "back-casting" emissions for younger vehicles
for older model years, as well as "forecasting" deterioration of aging vehicles for more recent
model years. Empty cells occur as well in high-power operating modes not covered by the
IM147 or IM240, meaning operating modes with power greater than about 24 kW/Mg.
33
-------
MY
Vehicle Age at Test (years)
O 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1980
1981
1982
| | | |
1983
_r^ i
1984
i _| i _|
1985
i _i i _i
1986
i i _j
1987
i i i—- i
1988
i _j i _j
1989
i i i i
1990
95 96 97 98 99 00 0102 03 04 05
1991
| _J | _J
1992
| | | |
1993
| 1 | 1
1994
| | J 1
1995
1 I _|
1996
I | _J
1997
| | |
1998
| | 1
1999
| | |
2000
| |
Figure 3-2 Model-year by age structure of the Phoenix I/M random evaluation sample
3.2.2.1.3.1 Rates
To estimate rates in empty cells (holes), we constructed statistical models of emissions data to
extrapolate trends in VSP and age. For this purpose, we generated a series of models based on
the MOVES operating-mode/ageGroup structure. Note that the extrapolated values were
modified on a case-by-case basis.
As a preliminary step, data were averaged for each test within a set of classes for VSP and speed.
We averaged emissions by model-year-group, regClass, age, VSP class, speed class and test.
Classes for VSP followed intervals of 3.0 kW/Mg (e.g., 0-3,3-6, ... 27-30,30+). Speed classes
followed those used for the MOVES operating modes (e.g., 1-25 mph, 25-50 mph, 50+ mph).
The resulting dataset had a single mean for each test in each 6-way cell. The purpose for this
averaging was to give the resulting statistical model an appropriate number of degrees of
freedom for each of the class variables, i.e., the d.f. would be determined by the number of tests
rather than the number of individual "second-by-second" measurements. Note that the matrix
used for this purpose was finer than that represented in Table 2-5.
We fit separate models in three groups of operating modes. For all operating modes except
brake/deceleration and idle, we fit one model incorporating VSP. We call this group
"coast/cruise/acceleration." For braking/deceleration and idle, we fit two additional models not
incorporating VSP, as these modes are not defined in other terms (Table 2-5). Overall, we fit
three models for each combination of cars and trucks, for the model-year groups shown in Table
3-5, giving a total of 60 models.
Before fitting a model, we drew a sample of vehicle tests in each model-year group (n = 1,200 to
3,500, see Table 3-6). This sampling was performed to fit models on smaller volumes of data
that a standard desktop computer could handle at the time. The sample was stratified by test
result {pass, fail) and age, with allocation proportional to that in the sample pool. Within each
result age stratum, tests were drawn using simple random sampling, and sampling frequencies
and weights,/strat and Wstrat, calculated as
34
-------
Yl IN
r strat ,., _ strat _ _ _
Jstrat - ~—> strat ~ ~— ~ Equation 3-9
strat /strat ^ strat
where nstrat and Marat are the number of tests selected from a stratum and total number of tests in
the stratum, respectively. Then, for each test selected, a final weight was calculated as the
product of the stratum weight and the initial sampling weight (wresuit,MY,CY), as shown in Equation
3-3.
Wfmal = ^result, MY, CY^strat Equation 3-10
Table 3-6 Sample sizes for statistical modeling, by regulatory class and test resultb
Model-year
group
LDV
LDT
fail
pass
fail
pass
1981-82
645
554
476
723
1983-85
569
631
508
691
1980-89
375
828
343
856
Iwii-ij;
:wi
l>44
2iw
W|
IW4-US
4i)fi
1 w5
37S
2.021
I
I.73X
I uw,
34ft
S54
IW7-S
fi71
1.73d
Each model included two sub-models, one to estimate means and one to estimate variances, as
described below.
3.2.2.1.3.1.1 Coast/Cruise/Acceleration
Means model
For the means sub-model, the dependent variable was the natural logarithm of emissions
InEh - + PXPN + P2Py + /?3Py + /34a + fi5S + P6PNS + y1ti + S Equation 3-11
where :
b Note that model years 1990-and-later were subsequently updated as explained in later sections of this report.
35
-------
• 1 nEh = natural-logarithm transform of emissions (in cell h),
• Pv, Pv2, Pv3 = first-, second- and third-order terms for vehicle-specific power
(VSP, kW/Mg),
• a = vehicle age at time of test (years),
• 5 = speed class (1 -25 mph, 25-50 mph and 50+ mph),
• t = test identifier (random factor)
• e= random or residual error
• /? = regression coefficients for the intercept and fixed factors Pv, a and 5.
• y = regression coefficients for the random factor test.
The model includes first-, second- and third-order terms in Pvto describe curvature in the power
trend, e.g., enrichment for CO and the corresponding decline in NOx at high power. The age
term gives an ln-linear trend in age. The speed-class term allows for a modified intercept in each
speed class, whereas the power/speed-class interaction allows slightly different power slopes in
each speed class. The random factor term for test fits a random intercept for each test, which
does not strongly affect the mean estimates but does affect the estimation of uncertainties in the
coefficients.
After fitting models, we performed basic diagnostics. We plotted residuals against the two
continuous predictors, VSP and age. We checked the normality of residuals across the range of
VSP and age, and we plotted predicted vs. actual values.
Variances model
The purpose of this sub-model was to model the variance of \riEh, i.e., the logarithmic variance
.v/2, in terms VSP and age. To obtain a dataset of replicate variance estimates, we drew sets of
replicate test samples. Each replicate was stratified in the same manner as the larger samples
(Table 3-6). To get replicate variances, we calculated ln-variance for each replicate within the
VSP/age matrix described above.
Models were fit on set of replicate variances thus obtained. The dependent variable was
logarithmic variance
s] = a0 + axa + a2Pv + a3PYa + S Equation 3-12
where Pv and a are VSP and age, as above, and a are regression coefficients. After fitting we
examined similar diagnostics as for the means model.
3.2.2.1.3.1.1.1 Model Application
Application of the model involved several steps. The first step was to construct a cell matrix
including all emission rates to be calculated, as shown in Table 3-7.
36
-------
Table 3-7 Construction of emission-rate matrix for light-duty gasoline vehicles
Count
Category
MOVES Database attribute
1
Fuel (gasoline)
fuelTypelD = 01
X
2
Regulatory Classes (LDV,
LDT)
regClassID = 20, 30
X
10
Model-year groups
As in Table 3-6
X
21
Operating modes
opModelD = 11-16, 21-30, 33-
40
X
7
Age Groups
ageGroupID = 3, 405, 607,
809,1014,1519, 2099
X
3
Pollutant processes (running
THC, CO, NO*)
polProcessID = 101, 201, 301
=
9,660
TOTAL cells
Next, we constructed a vector of coefficients for the means sub-model (P) and merged it into the
cell matrix.
P lA A A A A A(0-25) A(25-50) A(50+) A
Equation 3-13
Then, for each table cell, we constructed a vector of predictors (X/,). Equation 3-14 shows an
example for an operating mode in the 1-25 mph speed class, e.g., the value for the 1-25 mph
class is 1 and the values for the 25-50 and 50+ speed classes are 0. To supply values for VSP
(Pv) and age group (a), cell midpoints were calculated and applied as shown in Table 3-8.
X, =
1 Pv Pv2Pv3a 1 0 0 Pv
Equation 3-14
37
-------
Table 3-8 Values of VSP used to apply statistical models
opModelD
Range
Midpoint
11,21
<0
-2.0
12, 22
0-3
-2.5
13,23
3-6
4.5
14, 24
6-9
7.5
15,25
9- 12
10.5
16
12 +
14.5
27,37
12- 18
15.0
28,38
18-24
21.0
29,39
24-30
27.0
30
30 +
34.0
40
30 +
34.0
33
<6
0.5
35
6- 12
9.0
The final step was to multiply coefficient and predictor vectors, which gives an estimated
logarithmic mean (ln/w;) for each cell h.
\ViEh =X/iP Equation 3-15
The application of the variances model is similar, except that the vectors have four rather than
nine terms
® = [ ^0 ^2 ^3 ] Equation 3-16
\h = [ 1 CI Pwa ] Equation 3-17
Thus, the modeled logarithmic variance in each cell is given by
sjh = Xha Equation 3-18
In some model-year groups, it was not always possible to develop plausible estimates for the age
slope /?4, because the data did not cover a wide enough range of calendar years. For example, in
the groups 83-85 and 81-82, the data covered vehicles at ages of 10 years and older but not at
younger ages. Simply deriving slopes from the available data would have given values that were
much too low, resulting in very high emissions for young vehicles. In these cases, we considered
it more reasonable to adopt an age slope from a subsequent model year group. When making this
38
-------
assumption, it is necessary to recalculate the intercept, based on the assumed slope and the
earliest available data point.
Intercepts, denoted as /?o*, were recalculated by rearranging Equation 3-11 to evaluate the model
in operating mode 24, using the age slope from the previous model-year group (fi4*) and an
estimate of ln-emissions from the available dataset at the earliest available age (1 nEa*) at age a*.
In operating mode 24, the midpoint of the VSP range (6-9) is 7.5 kW/Mg and the speed class is
25-50 mph.
Pi = ln£fl* -7.5#-1.52P2 -7.53/?3 -/?>*-/?5(25_50) -7.5/?6 Equation3-19
On a case by case basis, age slopes were adopted from earlier or later model-year groups. In a
similar way, ln-variance models or estimates could be adopted from earlier or later model years.
3.2.2.1.3.1.2 B raki ng/D ecel erati on
3.2.2.1.3.1.2.1 Means Model
We derived models similar to those used for coast/cruise/accel erati on. For these operating
modes, however, the models were much simpler, in that they did not include VSP or the speed
classes used to define the coast/cruise/accel operating modes. Thus, emissions were predicted
solely in terms of age, although random intercepts were fit for each test as before:
In Eh= PQ+ Pft + y1ti + £ Equation 3-20
3.2.2.1.3.1.2.2 Variances Model
In addition, we fit variances models for these operating modes, which were also simple functions
of age.
sf =(X0+ OCfl + S Equation 3-21
3.2.2.1.3.1.2.3 Model Application
In these operating modes, rates were to be modeled for a total of 840 cells. This total is
calculated as in Table 3-7, except that the number of operating modes is 2, rather than 21. We set
up coefficient and predictor vectors, as before.
For the means and variances sub-models the vectors are
P = [AA] Equation 3-22
and
= [ 1 ^ ] Equation 3-23
respectively.
For the variances model the coefficients vector is
39
-------
OH = <2| ] Equation 3-24
and the predictor vector is identical to that for the means model.
As with coast/cruise/accel modes, we considered it reasonable in some model-year groups to
adopt a slope or ln-variance from a previous or later model-year group. In model-year groups
where the purpose was to backcast rates for younger vehicles, rather than forecast rates for aging
vehicles, it was again necessary to recalculate the intercept based on a borrowed age slope and an
estimate of 1 riEh calculated from the sample data for the youngest available age class. In this
case, Equation 3-25 is a rearrangement of Equation 3-20.
/?0 = In Ea* - * Equation 3-25
After these steps, the imputed values of 1 riEh were calculated, as in Equation 3-17.
3.2.2.1.3.2 Estimation of Model Uncertainties
We estimated the uncertainty for each estimated 1 nEh in each cell. During each model run, we
saved the covariance matrix of the model coefficients (.v//2). This matrix contains covariances of
each of the nine coefficients in relation to the others, with the diagonal containing variances for
each coefficient.
s2r =
Equation 3-26
5(0-25)
5(25-50)
'5(50+)
Using the parameter vectors Xi, and the covariance matrix .v//2, the standard of error of estimation
for each cell was calculated as
sIe„ = Equation 3-27
The standard error of estimation in each cell represents the uncertainty of the mean estimate in
the cell, based on the particular values of the predictors defining the cell.28 The pre- and post-
multiplication of the covariance matrix by the parameter vectors represents the propagation of
uncertainties, in which the parameters represent partial derivatives of each coefficient with
respect to all others and the co-variances represent the uncertainties in each coefficient in relation
to itself and the others.
40
-------
3.2.2.1.3.3
Reverse Transformation
To obtain an estimated emission rate Eh in each cell, the modeled means and variances are
exponentiated as follows
The two exponential terms use the results of the means and variances sub-models, respectively.
The left-hand "means" term represents the geometric mean, or the center of the implied log-
normal distribution, whereas the right-hand "variance" term reflects the influence of the "high-
emitting" vehicles representing the tail of the distribution.
The estimate of ln-variance could be obtained in several different ways. The first and preferred
option was to use the modeled variance as described above. A second option was to use an
estimate of variance calculated from the available sample of ln-transformed data. A third option,
also based on available data, was an estimate calculated from averaged emissions data and the
mean and variance of ln-transformed emissions data. This process involves reversing Equation
3-28 to solve for si2 If the mean of emissions data is xa and mean of ln-transformed data is x/,
then the logarithmic variance can be estimated as
In practice one of these options was selected based on which most successfully provided model
estimates that matched corresponding means calculated from the data sample.
The uncertainties mentioned above represent uncertainties in InEh. Corresponding standard errors
for the reverse-transformed emission rate Eh were estimated numerically by means of a Monte-
Carlo process. At the outset, we generated a pseudo-random set of 100 variates of 1 riEh, based on
a normal distribution with a mean of 0.0 and variance equal to sx-ag1. We applied Equation 3-28
to reverse-transform each variate, and then calculated the variance of the reverse-transformed
2
variates. This result represented the variance-of-the-mean for Eh($Eh), as in Equation 3-7.
Finally, we calculated the CV-of-the-mean (CV#/,) for each modeled emission rate, as in
Equation 3-8.
3.2.2.1.4 Table Construction
After compilation of the modeling results, the subset of results obtained directly from the data
(Equation 3-4 to Equation 3-8), as shown in the shaded area in Figure 3-2 and the complete set
generated through modeling (Equation 3-11 to Equation 3-29) were merged. A final value was
selected for use in the model data table. The value generated from data was retained if two
criteria were met: (1) a subsample of three or more individual vehicles must be represented in a
given cell (m > 3), and (2) the CVa (relative standard error, RSE) of the data-driven Eh must be
less than 50 percent (CV£ < 0.50). Failing these criteria, the model-generated value was
substituted. For purposes of illustration, results of both methods are presented separately.
At this point, we mapped the analytic model-year groups onto the set of model-year groups used
in the MOVES database. The groups used in the database are designed to mesh with heavy-duty
Eh = elnEhex5s'
Equation 3-28
Equation 3-29
41
-------
standards and technologies, as well as those for light-duty vehicles. To achieve the mapping, we
replicated records as necessary, in cases where the analytic group was broader than the database
group. Both sets of groups are shown in Table 3-9.
Table 3-9 Mapping "analytic" model-year groups onto MOVES-database model-year groupsc
"Analytic"
"MOVES database"
modelYearGroupID
shortModYrGroupID
Cars
Trucks
1981-82
1981-82
1980 and previous
19601980
1
1981-82
1981-82
1981-82
19811982
61
1983-85
1983-85
1983-84
19831984
62
1983-85
1983-85
1985
1985
85
1986-89
1986-89
1986-87
19861987
63
1986-89
1986-89
1988-89
19881989
64
1990-93
1990-93
1990
1990
90
1990-93
1990-93
1991-1993
19911993
65
1994-95
1994-95
1994
1994
94
1994-95
1994-95
1995
1995
95
1996-98
1996
1996
1996
96
1996-98
1997-98
1997
1997
97
1996-98
1997-98
1998
1998
98
1996-98
1997-98
1999
1999
99
1996-98
1997-98
2000
2000
20
3.2.2.2 Adjustment for High-Power Operating modes
The rates described were derived from data measured on IM240 or IM147 cycles, which are
limited in terms of the ranges of speed and vehicle-specific power that they cover. Specifically,
these cycles range up to about 50 mph and 24 kW/Mg for speed and VSP, respectively. Some
data does exist outside these limits but can be sporadic and highly variable. The operating modes
outside the I/M window include modes 28,29,30, 38, 39 and 40, which we'll refer to as the 'high-
power' operating modes. For these modes, the statistical models described in 3.2.2.1.3 above
were used to extrapolate up to about 34 kW/Mg.
Based on initial review and comment on this aspect of the analysis, for MOVES2010, we gave
additional scrutiny to the high power extrapolation. To obtain a framework for reference, we
examined a set independently measured data, collected on drive cycles more aggressive than the
IM cycles, namely, the US06 and the "Modal Emissions Cycle" or "MEC." Much of the data
0 Note that model years 1990-and-later were subsequently updated as explained in later sections of this report.
42
-------
was collected in the course of the National Cooperative Highway Research Program (NCHRP)29
and the remainder on selected EPA programs, all stored in OTAQ's Mobile-Source Observation
Database (MSOD). Unlike the US06, which was designed specifically to capture speed and
acceleration not captured by the FTP, the MEC is an "engineered" cycle, designed not to
represent specific driving patterns, as does the FTP, but rather to exercise vehicles through the
ranges of speed, acceleration and power comprising the performance of most light-duty vehicles.
Several variants of the MEC were developed to provide a database to inform the development of
the Comprehensive Modal Emissions Model (CMEM).29 Driving traces for the US06 and MEC
cycles are shown in Figure 3-3 and Figure 3-4. Both cycles range in speed up to over 70 mph and
in VSP up to and exceeding 30 kW/Mg.
Time (sec)
Drive Cycle mec us06
Figure 3-3 Example speed traces for the US06 and MEC cycles
43
-------
c
c
g.
CL
e
40"
30
20
V
-10
-20
-30 i
1 1 1 1 i 1 1 1 1 i 1
0 200 400
I 1 1 1 1 I 1 1 1 1 I 1 1 1 1 I 1 1 1 1 I 1 1 1 1 I 1 1 1 1 I 1 1 1 1 I
600 800 1000 1200 1400 1600 1800 2000
Time (sec)
— mec us06
Drive Cycle
Figure 3-4 Example vehicle-specific-power (VSP) traces for the US06 and MEC cycles
Table 3-10 summarizes the numbers of available tests by regulatory class, model-year group and
drive cycle, with numbers of tests differing in each model-year group. Samples were somewhat
larger for cars for both cycles, which represented a broad range of model-years.
Table 3-10 Sample sizes for US06 and MEC cycles (No
tests)
Model-year group
Car
Truck
Total
US06
MEC
US06
MEC
1980 & earlier
4
14
6
24
1981-85
15
23
8
19
65
1986-89
21
24
13
31
89
1990-93
54
57
22
36
169
1994-95
49
45
22
30
146
1996-99
58
28
56
17
159
Total
201
191
121
139
652
Figure 3-5, Figure 3-6 and Figure 3-7 show trends in emissions vs. VSP for CO, THC and NO*
for LDV and LDT by model year group. Both cycles were averaged and plotted as aggregates.
44
-------
VSP (kW/tonne)
Reg/MYG LDT-0080 LDT-8185 LDT-8689 —^ LDT-9093
LDT-9495 LDT-9699 LDV-0080 LDV-8185
LCA/-8689 — — LCV-9093 — — LDV-9495 LDV-9699
Figure 3-5 CO emissions (g/sec) on aggressive cycles, vs. VSP, by regulatory class and model-year group
0.15
0.14
0.13
0.12
0.11
0.10
0)
009
o>
OCR
<1>
t
0.07
CO
0.06
F
0.0b
0.04
0.03
0.02
0.01
0.00
20
VSP (kW/tonne)
Reg/MYG
' LDT-OOGO
1 LDT-9495
LW—8689
' LDT-8185
1 LDT-9699
LCW-9093
' LDT-8689
LDV-0080
LDV-9495
LDT-9093
LW—8185
LW—9699
Figure 3-6 THC emissions (g/sec) on aggressive cycles, vs. VSP, by regulatory class and model-year group
45
-------
Figure 3-7 NO* emissions (g/sec) on aggressive cycles, vs. VSP, by regulatory class and model-year group
3
o>
0.21
0,20;
0.19;
0.18;
0.17;
0.16 ¦
0.15;
0.14;
0.13;
0.12;
0.11;
0.10;
0.09;
0.08;
0,07;
0.06;
0.05;
0.04;
0,03;
0.02;
0.01;
0.00 4,
-10
Reg/MYG
0 10 20 30
VSP (kW/tonne)
LDT-0C6O LDT-8185 LDT-8689
LDT-9495 LDT-9699 — — LDV-0080
LCV-8689 — — LW—9093 LDV-9495
To construct a basis for reference, we averaged the data by regulatory class, model-year group
and operating mode, using the model-year groups shown in Table 3-10. After averaging, we
calculated ratios from high-power operating modes to a selected reference mode. Specifically,
we selected two modes covered by the IM cycles (27 and 37) to serve as reference points. The
midpoint VSP for each is -15 kW/Mg. With mode 27 as a reference, we calculated ratios to
modes 28, 29 and 30.
E, .
Rr.n=-jr^-> for/' =28,29,30 Equation 3-30
h,21
and with mode 37 as a reference, we calculated ratios to modes 38, 39 and 40.
E,
^:37=TTi£_' for/ =38>39>40 Equation 3-31
/j,37
After calculating the ratios, we calculated ratio-based emissions estimates (ER) as the products of
their respective ratios and the initial rate for modes 21 ox hi
T?R D rpinitial rpR d rpinitial _ . _ _ _
h,i ~ t.Tl h,21 > 01 ^h,i ~ ^v37^h,37 Equation 3-32
46
-------
respectively, where Ehmitial is the initial data-driven or model-generated rate calculated as
previously described.
The next step, the process by which ratio-based rates were selected as rates for particular
operating modes on a case-by-case basis changed substantially for the final rates used in
MOVES2010 and later. In the draft, we calculated upper and lower confidence limits for ER and
replaced the initial rate with ER if it fell outside the confidence band, i.e., if the initial rate was
greater than the upper bound or lower than the lower bound. Evaluation of the results of this
approach showed, however, that it gave spurious results in many cases. We found it impossible
to assign a confidence level for the band that would work in all cases, i.e., sufficiently sensitive
to identify and correct problem cases, but not so sensitive so as to make unnecessary
modifications.
For the final rates, we developed a different logic for applying the ratio-based rates. One change
from the draft is that ratio-based rates were considered only for modes 29,30, 39 and 40, i.e.,
modes spanning the range of VSP beyond the IM147. Modes 28 and 38 are partially covered by
the I/M cycles, and the differences among the data, model and ratios were generally much
smaller than for the four highest modes. The steps in the revised process are:
1) Identify acceptable candidate values (data, model or ratio). The data values were considered
acceptable if (1) a value was present, (2) it met the acceptability criteria (described above) and
(3) it was greater than the value in the next lowest mode. Similarly, predicted values were
acceptable if they exceeded the value for the preceding operating mode.
Following these evaluations, the final value was selected as the minimum of the acceptable
candidates. These criteria were applied sequentially to prevent declining emissions trends with
increasing power. As a first step, values were selected for operating modes 29 and 39, relative to
modes 28 and 38. In a successive step, values were selected for 30 and 40, relative to those
selected for 29 and 39, respectively. We present some examples below, showing differences
between the draft and final rates.
In the THC example (Figure 3-8), the final values are substantially reduced, particularly for
modes 29 and 30. In the draft (a), the initial rates fall outside the confidence intervals for the
ratio-based rates for three out of six possible cases, i.e., in modes 30, 39 and 40. The resulting
rate is higher for modes 30 and 40, but lower for 39. In the final rates, the results vary. For
modes 29 and 30, the data values meet the criterion of the minimum value giving an increasing
trend from mode 28 - 30. However, for modes 39 and 40, the ratio and the model give the
values meeting the criterion, as shown in (c).
The example for CO shows different behavior in the draft, but a similar outcome in the final
(Figure 3-9). In the draft (a), the initial values for modes 28-30 all fall within the confidence
intervals for the ratio-based value and are thus retained. The values for 39 and 40, fall outside the
band on the low side and are replaced by the ratio-based rates. For operating modes 29 and 30,
the data is selected as the minimum option available, as with THC. For modes 39 and 40, the
model is similarly selected. In the final rates, the ratio-based values are not adopted for this
example, as they had been in the draft, and the net result is a decrease in CO rates in the affected
operating modes.
Finally, in the NOx example (Figure 3-10) the initial rates are replaced in five out of six cases in
the draft (a). The initial values for 28-30 and 40 all fall below the lower confidence limit,
47
-------
whereas that for 30 falls above the upper confidence limit. In the final, the ratio is used more
sparingly, as in the THC and CO examples. Model values are used in two cases (modes 30 and
40) and the ratio in one case (mode 39).
These examples highlight the uncertainty of projecting emissions at high power and of projecting
beyond the range of the IM147. Uncertainties are much smaller for opModes 28 and 38 than for
29, 30, 39 and 40. This pattern may be due to the fact that, for modes 28 and 38, the power range
for the IM147 overlaps somewhat the range of the aggressive cycles. For this reason, the degree
of extrapolation is lower and the power trends are similar.
48
-------
80
70
60
50
40
30
20
10
fal Draft
--
[[
I
1
t/'
4
~ «**~~~*~~~~~* ~ ~ ~
Initial
- ratio
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
60
40
5
"<5
CL 30
00
20
E
10
(b) Final: options available
~
A
~
• r
m
i
A
£ 6
. - * '
1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
opModelD
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
opModelD
Source • • • Data a a a Model n n n Rat,0
Figure 3-8 THC emission rates (g/hr), vs. VSP for MY 1998 cars at ages 4-5 years: (a) options for draft rates,
(b) options for final model (data, model and ratio) and (c) options selected for final rates
49
-------
5,000
4,500
4,000
3,500
3,000
2,500
2,000
1,500
1,000
500
0
i
r
)
r
J
L A
~ Initial
-H— ratio
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
(b) Final: options available
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
opModelD
1400
1300
1200
1100
1000
900
800
700
600
500
400
300
200
100
01
•
(c) Final: options selected
•
A
i
•
•
• •
•
•
• • • •
• • •
•
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
opModelD
Source • • • Data a a a Model
Figure 3-9 CO emission rates (g/hr), vs. operating mode for MY-1998 trucks at ages 6-7: (a) options for draft
rates, (b) options for final model (data, model and ratio and (c) options selected for final rates
50
-------
500
450
400
350
300
250
200
150
100
50
r
* X"
1 J1
~
~ ~ «. , ~
. ~ . ~ *
~ ~~~* , , ~~ ~
~ Initial
-B— ratio
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
c 300
1
¦*6
Di 200
LU
c
"3
E
Of
(b) Final: options available
¦
&
A
A
•
A
0 6
A
*
A
A • A
• 6 *
* A #
, 4 * . a •
i • ~
•
£
&
1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
opModelD
3 200
(c) Final: options selected
0 1 11 12 13 14 15 18 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
opModelD
Source • • • Data a a a Model Ratio
Figure 3-10 NOv emission rates (g/hr) vs. operating mode for MY-1995 Cars at ages 8-9: (a) options for draft
rates, (b) options for final model (data, model and ratio and (c) options selected for final rates
51
-------
3.2.2.3 Stabilization of Emissions with Age
One characteristic of the data is that fleet-average emissions do not appear to increase
indefinitely with age, but rather tend to stabilize at some point between 12 and 15 years of age.
This behavior is visible in datasets with enough historical depth for age trends to be observable,
including the Phoenix random sample and long-term remote-sensing studies.14
Figure 3-11 and Figure 3-12 show age trends by model year for cars and trucks, respectively.
The values shown are aggregate mass rates over the IM147 expressed as g/sec for CO, THC and
NO,,
At the time that emission trends with age were determined for the 1989-and-earlier vehicles, no
data was available at ages older than 15 years for model years older than 1990. Thus it was
necessary to project emissions.
However, it is not appropriate to simply extrapolate the statistical models past about 8-10 years.
As described above, emissions were modeled as ln-linear with respect to age, which implies
exponential trends for reverse-transformed values. However, exponential trends will increase
indefinitely if extrapolated much beyond the range of available data, which obviously does not
describe observed patterns of fleet emissions. To compensate for this limitation, we employed a
simple approach to represent the decline and stabilization of the rates.
We calculated ratios of means between the 10-14 and the 15-19 year age groups, each relative to
the 8-9 year age group, using the 1986-89 and 1990-93 model-year groups, which contain data
for vehicles as old as 19 years. For this purpose we used Phoenix data averaged by MOVES
model-year and age groups, as shown in Figure 3-13. Data points in the figure represent
aggregate tests (g/mi). After averaging by model-year group and ageGroup, we calculated ratios
of means for the 10-14 and 15-19 ageGroups.
= t^14 , ^age=^dl Equation 3-33
8-9 8-9
We calculated modified rates for the 10-14 and 15-19 year ageGroups as the product of the rate
for the 8-9 year ageGroup and the corresponding ratio (i?age). Assuming that emissions would be
fully stable by 20 years, we set the rate for the 20+ year ageGroup equal to that for the 15-19
year ageGroup. We calculated variances for the ratios as in Equation 3-37.
52
-------
Table 3-11 Ratios used to stabilize emission rates for the 10-14 and 15-19 year ageGroups, calculated
relative to the 8-9 year ageGroup
Regulatory Class
ageGroup
Ratios (Rage)
Variances (Vr)
THC
CO
NOx
THC
CO
NOx
Cars
10-14
1.338
1.226
1.156
0.000000032
0.000160
0.00000009
Cars
15-19
1.571
1.403
1.312
0.00000411
0.00268
0.00000261
Trucks
10-14
1.301
1.220
1.156
0.00000173
0.000758
0.00000138
Trucks
15-19
1.572
1.479
1.312
0.0000518
0.0666
0.0000499
53
-------
LDV WEIGHTED
CO vs. Age (years)
Vehicle age (years)
1992 r! ~ ~ 1993 AAA 1994 AAA 1996 AAA 1996
19S8 AAA lag© 2000 ' ' ' 2001 ' ' ' 2002
LDV, WEIGHTED
ThC vs. Age (years), LDV
\fehicle age (years)
LDV WEIGHTED
NOx \e. Age (years), LDV
\fehicle age (years)
1S60 o Q Q 1981 OOP 1S82 1263 O O O 1S64 O O O 1S66
1S8S ~ ~ ~ 1S87 ~ ~ ~ 1S88 ~ ~ D ises 1990 ~ ~ ~ 1991
1902 ~ ~ ~ 1993 AAA 1994 AAA 19S6 AAA 1996 1997
Figure 3-11 Aggregate IM147 emissions (g/sec) for cars, by model year and age, for the Phoenix random
evaluation sample
54
-------
IDT. WEIGHTED
CO vs. Age (years)
©
®
i
£
8
(a) CO
TSfl
tA*/
ty«s
© j= a sa * to
Vfehacte ago (yoarf;)
XMZ
T3bU
raw
Tsfcb
B91
190/
2003
191*.
1932
TaW
24XH
LDT WEIGHTED
THC vs. Aga (years), LDT
tse/
n n n ^
tag®
© 5= fti S M- Hi
Vfehicta ago (yr»m)
rate
T9UJ
Wl
^Sk»
TSSO
TS9i>
2002
t9«b
1907
2003
LDT, WEIGHTED
N3x vs. Aga (years), LDT
(c) NOx
-a S--PT' '
¦ ; ¦ r ¦ i ¦ i
Model Year o o o wi o o o
law n " n
n n n ism
&-A-6 iggg AAA g«»
¦ I ¦ 1 ' I ¦ I ¦ I
©ssui2s.wnifcs
\frtfiicta ag© (yiTflifi)
tae ,Tr ° ° «w *
ia® " n n
ISSk A-A-A T3st: f
2001 ' 1 * 2002
ISfc o Q o 19tfc
BS>1 " n n 1932
190/ ¦¦' 1,1 J- T9Sti
2003 1 * 1 2004
Figure 3-12 Aggregate IM147 Emissions (g/sec) for trucks, by model year and age, for the Phoenix random
sample
55
-------
30.0
20.0
a>
E 15.0
ot
8 1
6 i
4 \
2 \
E
oi0
1°
v) 0
-------
3.2.2.3.1
Non-I/M Reference Rates
The ratios developed in 3.2.2.3 are assumed to apply in I/M areas, as the underlying data was
collected in the Phoenix I/M area. It is therefore plausible that the patterns observed may be
reflective of I/M areas. However, in the absence of a program, high-emitting vehicles are not
identified and owners have less incentive to repair or replace them. Thus, the question arises as
to whether deterioration patterns would necessarily be identical in non-I/M as in I/M areas. Two
plausible scenarios can be proposed. In the first, the pattern of deterioration followed by
stabilization is similar in non-I/M as in I/M areas, but emissions stabilize at a higher level, and
perhaps at a later age. In the second, emissions continue to increase in non-I/M areas, but at a
slower rate after 10-15 years.
Data that sheds light on these questions are very limited, as the datasets with sufficient history
were collected within I/M areas. Thus, given the absence of information, we adopted an
assumption that, absent the existence of a program, emissions would increase after 19 years. We
applied this assumption by assuming that the ratio observed between the 10-14 and 15-19 year
ageGroups would persist in linear fashion from the 15-19 to the 20+ year ageGroups.
Table 3-12 shows the deterioration stabilization ratios for both the I/M and non-I/M reference
rates. As mentioned above, the ratios are applied by multiplying them by the values for the 8-9
year age group in all operating modes. The ratios for I/M areas (i?age,i/M) are identical to those in
Table 3-11. The center column shows the ratio of values of i?age,i/M for the 15-19 to the 10-14
year ageGroups. Ratios for the non-I/M references (i?age,non-i/M ) are identical to those for I/M in
the 10-14 and 15-19 year ageGroups. In the 20+ year ageGroup, the non-I/M ratio is equal to the
product of the 15-19 value and the ratio of the 15-19 and the 10-14 values.
57
-------
Table 3-12 Deterioration-stabilization ratios as applied to I/M and non-I/M reference rates
Pollutant
Regulatory
Class
ageGroup
¦Rage,I/M1
Ratio (15-19:10-14)
./?age,non-I/M
10-14
1.338
1.338
Cars
15-19
1.571
1.174
1.571
THC
20+
1.571
1.845
10-14
1.301
1.301
Trucks
15-19
1.572
1.206
1.572
20+
1.572
1.898
10-14
1.226
1.226
Cars
15-19
1.403
1.144
1.403
CO
20+
1.403
1.606
10-14
1.220
1.220
Trucks
15-19
1.479
1.213
1.479
20+
1.479
1.795
10-14
1.159
1.159
Cars
15-19
1.312
1.132
1.132
NO,
20+
1.312
1.486
10-14
1.159
1.159
Trucks
15-19
1.312
1.132
1.132
20+
1.312
1.486
1 Values in this column are identical to those in Table 3-11.
2 Calculated as the ratio of the values in the current and previous rows.
3 For 10-14 and 15-19 year ageGroups, values in this column identical to the I/M column; for the 20+ year ageGroup, values
in this column equal the product of the value in the previous row (15-19) and the value in the center column.
3.3 MOVES2014 Emission-Rate Development (MY 2001-2016)
This section describes methods used in developing model rates for MOVES2010 and
MOVES2014, representing emissions from vehicles certified to National LEV and Tier-2
standards, in model years 2001 and later. This material is retained because the MOVES2014
58
-------
rates provide the basis for the updated MOVES3 rates, after being modified by adjustments as
described in Sections 3.6, 3.7, and 3.10 below.
3.3.1 Data Sources
Data for vehicles in model years 2001 and later was acquired from results of tests conducted
under the In-Use Verification Program (IUVP). This program, initiated in 2003, is run by
manufacturers and administered by EPA/OTAQ through the Compliance Division (CD).
To verify that in-use vehicles comply with applicable emissions standards, customer-owned
vehicles at differing mileage levels are tested on an as-received basis with minimal screening.
Emissions are measured on the Federal Test Procedure, US06 and other cycles. The FTP is most
relevant to our purposes, but the US06 is also important.
3.3.1.1 Vehicle Descriptors
In addition to the parameters listed above in Table 3-2, the IUVP data provides test-group
(formerly engine family) information. Using test group, the IUVP files can be merged with
certification test records by model year. The certification test records provide information on
standard level and specific emissions standards applicable to each vehicle. The standard level
refers to the body of standards to which vehicles were certified (Tier 1, NLEV, LEV-I, LEV-II),
and the standards refer to specific numeric standards for THC, CO or NO*, where THC are
represented by non-methane hydrocarbons (NMHC) or non-methane organic gases (NMOG),
depending on combinations of standard level and vehicle class (LDV, LDT1-4).
Table 3-13 Vehicle descriptors available in IUVP files and certification test records
Parameter
Source
Purpose
IUVP
Cert. Records
VIN
Y
Verify MY or other parameters
Fuel type
Y
Make
Y
Y
Model
Y
Y
Model year
Y
Y
Assign sourceBinID, calculate age-at-test
Test group1
Y
Y
Tier
Y
Emissions Standard
Y
Assign Vehicle Class
'Formerly "engine family."
Combining data from both sources allows individual test results to be associated with the correct
standard level and emissions standard, allowing inference of the correct vehicle class.
59
-------
3.3.2 Estimating Reference Rates
The goal of this process is to represent "with I/M" reference rates for young vehicles, i.e., the
first ageGroup (0-3 years). The rates are estimated by Tier, model year and regulatory class. The
process involves six steps, each of which is discussed in more detail in Sections below.
1. Average II/VP results by standard level and vehicle class.
2. Develop phase-in assumptions for MY 2001 - 2017, by standard level, vehicle class and
model year.
3. Merge FTP results and Phase-in assumptions. For running emissions, calculate weighted
ratios of emissions in each model year to those for Tier 1 (MY2000). We assumed that the
emissions control at high power (outside ranges of speed and acceleration covered by the FTP)
would not be as effective as at lower power (within the range of speed and acceleration covered
by the FTP).
4. Estimate Emissions by Operating Mode. Then calculate emissions by operating mode in each
model year by multiplying the MY2000 emission rates by the weighted ratio for each model
year.
5. Apply Deterioration to estimate emissions for three additional age Groups (4-5, 6-7 and 8-9).
We assume that NLEV and Tier 2 vehicles will deteriorate similarly to Tier-1 vehicles, when
viewed in logarithmic terms. We therefore apply ln-linear deterioration to the rates developed in
steps 1-4. For the remaining three groups, emissions are assumed to stabilize as described above
on page 52.
6. Estimate non-LMreference rates. The rates in steps 1-6 represent I/M references.
Corresponding non-I/M references are calculated by applying the ratios applied to the Tier-1 and
pre-Tier-1 rates (see Section 3.5, page 96).
Each of these steps is described in greater detail in the sub-sections below.
3.3.2.1 Averaging IUVP Results
In using the IUVP results, "cold-start" emissions are represented as "Bag 1 - Bag 3" i.e., the
mass from the cold-start phase less that from the corresponding hot-start phase. Similarly, "hot-
running" emissions are represented by the "Bag 2," or the "hot-stabilized" phase, after the initial
cold-start phase has conditioned the engine.
The first step is to average the IUVP results by Tier and vehicle Class. Results of this process
are shown below. In the figures, note that the HC values represent non-methane hydrocarbons
(NMHC) for Tier 1 and non-methane organic gases (NMOG) for NLEV and Tier 2. Figure 3-14
shows FTP composite results in relation to applicable certification and useful-life standards. For
THC and NO*, the data show expected compliance margins in the range of 40-60 percent in most
cases. For CO, compliance margins are even larger, ostensibly reflecting the concomitant effects
of HC or NO,; control on CO emissions.
Figure 3-15 shows results for separate phases of the FTP, to examine differential effects of
standards on start and running emissions. As mentioned, the "cold-start" emissions are
represented by the difference between Bags 1 and 3, divided by the nominal bag distance (3.59
miles) which expresses the values as a "start rate" in g/mi. The "hot-running" emissions are
represented by Bag 2 emissions, also divided by the appropriate distance to obtain an aggregate
60
-------
rate, in g/mi. Additionally, Figure 3-16 shows composite, start and running values normalized to
their respective Tier-1 levels, which clearly displays the greater relative levels of control for
running as opposed to start emissions. Not surprisingly then, distinguishing start and running
emissions shows that composite FTP values for HC and CO are strongly influenced by start
emissions. Starts are also important for NOx, but to a lesser degree. In any case, the results show
that sole reliance on composite results in projecting future emissions declines would give
misleading results in projecting either start or running emissions. Hence, the method described
below emphasizes treating them separately.
Figure 3-16 shows composite, start and running emissions each normalized to their Tier 1 levels.
These ratios are applied in a subsequent step to estimate running emission rates.
61
-------
0.350
0.300
0.250
0.200
0.150
0.100
0.050
0.000
T1 TLEV
T1 NLEV
LEV ULEV
NLEV NLEV
bin8
T2
bin7 bin6
T2 T2
bin5
T2
bin4 bin3
T2 T2
bin2
T2
E
D)
in
c
o
'
-------
Cold start (g/mi) M Hot Running (g/mi)
T1
T1
¦
LEV
—¦—
ULEV
—¦—
bin8
—¦—
bin7
—¦—
bin6
—¦—
bin5
—¦—
bin4
—¦—
bin3
NLEV
NLEV
12
12
12
12
12
12
bin2
12
¦—
—-¦
¦
¦—
—¦—
T1
TLEV
LEV
ULEV
bin8
bin7
bin6
bin5
bin4
bin3
bin2
T1
NLEV
NLEV
NLEV
12
12
12
12
12
12
12
1.000
0.900
0.800
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.000
—¦—
—¦—
—
—¦—
—
11
TLEV
LEV
ULEV
bin8
bin7
bin6
bin5
bin4
bin3
bin2
T1
NLEV
NLEV
NLEV
12
12
12
12
12
12
12
Figure 3-15 Cold-start (Bag 1 - Bag 3) and hot-running (Bag 2) FTP emissions for Tier 1, NLEV and Tier 2
passenger cars (LDV), as measured by IUVP (g/mi)
63
-------
1.000
0.900
0.800
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.000
T1
T1
TLEV LEV
NLEV NLEV
ULEV bin8 bin7
NLEV T2 T2
bin6 bin5 bin4 bin3 bin2
T2 T2 T2 T2 T2
1.000
a) 0.900
5 0.800
-1 0.700
i 0.600
j2 0.500
o 0.400
q 0.300
'.5 0.200
UL 0.100
0.000
T1
T1
TLEV
LEV
ULEV
bin8
bin7
bin6
bin5
bin4
bin3
NLEV
NLEV
NLEV
T2
T2
T2
T2
T2
T2
bin2
T2
1.000
0.900
0.800
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.000
TLEV
LEV
ULEV
bin8
bin7
bin6
bin5
bin4
bin3
NLEV
NLEV
NLEV
T2
T2
T2
T2
T2
T2
bin2
T2
Figure 3-16 Composite, cold-start (Bag 1 - Bag 3) and hot-running (Bag 2) FTP emissions for Tier 1, NLEV
and Tier 2 passenger cars (LDV), as measured by IUVP, normalized to respective Tier-1 levels
-Composite —¦—Cold Start A Hot Running
64
-------
3.3.2.2
Develop Phase-In Assumptions
To estimate emissions levels for specific model years, we developed assumptions describing the
phase-in of new emissions standards after model year 2000. For rates stored in the MOVES
default database, we developed assumptions intended to apply to vehicles sold in states where
Federal, rather than California standards applied. Thus, the phase-is designed to represent the
phase-in of National-Low-Emission-Vehicle (NLEV) and Tier 2 standards.
To achieve these steps, we obtained certification records and test results for a selection of model
years.30 These records contain information on certified vehicles, including model year, test
group (engine family), standard level (Tier-1, LEV, Bin 5, etc.), and sales area, as well as
numerical standards used for certification on the Federal Test Procedure (e.g., 0.05 g
NMOG/mile, etc.). For each engine family, we inferred the vehicle class (LDV, LDT1-LDT4)
based on combination of standard and numerical values. Examples illustrating this process are
shown in Table 3-14.
After compiling lists of engine families by standard, model year and vehicle class, we obtained
estimates of final sales from the EPA VERIFY database for MY 2001-2007.31d We merged the
certification records with the sales estimates, by model year and engine family.
Then to estimate the default "Federal" phase-in, we summed the sales by model year, standard
level and vehicle class, for a subset of sales areas in which Federal or California standards
applied, excluding those sales areas in which only California standards applied. Estimates of
numbers of engine families certified for various sales areas are listed in Table 3-15. Sales-
weighted phase-in scenarios for each vehicle class are shown in Figure 3-17 through Figure 3-20.
As noted, the results in the Figures reflect the certifications in the "Fed" or "Both" groups shown
in Table 3-15.
Proportions of each standard represent actual phase-in history for MY 2001-2007. We projected
phase-in assumptions through MY2010, after which we held assumptions constant, under
assumption that the Tier 2 phase-in would be complete.
The National LEV (NLEV) standards apply only to LDV, LDT1 and LDT2 vehicle classes, for
which Tier 1 certification ended in MY 2000. Certification to NLEV standards began in 2001
and ended in 2006, however, NLEV vehicles dominate the (Federal) fleet between 2001 and
2003. Tier 2 vehicles enter the fleet in 2003 and completely comprise new sales by 2010.
The phase-in for LDV, LDT1 and LDT2 are broadly similar in that LEV and Bin 5 vehicles
dominate certifications and sales. There are relatively small differences in that LDV-T1 contains
higher fractions of ULEV and Bin 8.
The phase-in for heavy light-duty trucks is simpler in that Tier-1 certifications continue through
2004, after which Tier 2 standards are introduced. After 2003, certifications are dominated by
Bin 8, Bin 5 and Bin 4.
d Note that this database has been renamed as the "Engines and Vehicles Compliance Information System" (EV-
CIS).
65
-------
Table 3-14 Examples of information obtained from certification test records, with vehicle class inferred from
combinations of standard, and FTP certification values
Standard
Engine Family
Sales Area
FTP Standard
Vehicle-Class
50,000-mi
100,000-mi
120,000-mi
LEV
2HNXV02.0VBP
NLEV all states
0.075
0.09
LDV, LDT1
LEV
2MTXT02.4GPG
NLEV all-states
0.100
0.13
LDT2
Tier 1
2CRXT05.95B2
Federal all-altitude
0.32
0.46
LDT3
Tier 1
2CRXT05.96B0
Federal all-altitude
0.39
0.56
LDT4
Table 3-15 Approximate numbers of engine families certified, by model year and age group, for model years
2001-2007
Sales Area
Code
Group1
Model Year
Total
2001
2002
2003
2004
2005
2006
2007
California
CA
CA
114
116
118
240
251
275
255
1,369
Clean Fuel Vehicle
CF
Fed
38
46
81
76
69
61
55
426
California + NLEV
(all states)
CL
Both
149
140
129
418
Federal All Altitude
FA
Fed
79
75
86
209
219
271
274
1,213
Federal + CA Tier 2
FC
Both
16
81
41
33
16
187
Clean Fuel Veh +
NLEV(ASTR)2
+ CA
NF
Both
57
56
45
158
NLEV (All States)
NL
Fed
31
47
74
152
TOTAL
468
480
549
606
580
640
600
3,923
1 "Fed" denotes areas for which vehicles were certified to Federal Tier 1, NLEV or Tier 2 standards, "CA" denotes vehicles
certified to California LEV-I or LEV-II standards, including the "section 177" states, "Both" denotes vehicles certified for
Federal or California Sales Areas.
2 "ASTR" = "All-state trading Region."
66
-------
100%
90%
80%
70%
60%
c
Q.
40%
30%
20%
10%
0%
lTier-2(Bin 2)
lTier-2(Bin 3)
Uier-2(Bin 4)
lTier-2(Bin 5)
~ Tier-2(Bin 6)
lTier-2(Bin 7)
lTier-2(Bin 8)
LEV-II (ULEV)
ILEV-II (LEV)
I NLEV(ULEV)
I NLEV(LEV)
INLEV(TLEV)
l Tier 1
Model Year
Figure 3-17 Phase-in assumptions for Tier 1, NLEV, and Tier 2 standards, for LDV and LDT1
100%
90%
80%
70%
60%
c
O)
o 50%
0)
Q.
40%
30%
20%
10%
0%
lTier-2(Bin 3)
ITier-2(Bin 4)
lTier-2(Bin 5)
lTier-2(Bin 7)
¦ Tier-2(Bin 8)
I NLEV(ULEV)
I NLEV(LEV)
I NLEV(TLEV)
¦ Tier 1
Model Year
Figure 3-18 Phase-in assumptions for Tier 1, NLEV and Tier 2 standards, for LDT2
67
-------
100%
90%
80%
70%
60%
c
Q)
o 50%
Q)
Q_
40%
30%
20%
10%
0% -
¦ Tier-2(Bin 4)
¦ Tier-2(Bin 5)
¦ Tier-2(Bin 8)
¦ Tier 1
Model Year
Figure 3-19 Phase-in assumptions for Tier 1 and Tier 2 standards, for LDT3
100%
¦ Tier-2(Bin 5)
¦ Tier-2(Bin 8)
¦ Tier 1
Model Year
Figure 3-20 Phase-in assumptions for Tier 1 and Tier 2 standards, for LDT4
68
-------
3.3.2.3
Merge FTP Results and Phase-In Assumptions
The goal of this step is to calculate weighted averages of the FTP cold-start and running results
for all standards in each model year, with the emissions results weighted by applicable phase-in
fractions. We do this step for each vehicle class separately, then we weight the four truck classes
together using a set of fractions also derived from the weighted sales estimates. Through MY
2007, where we had actual history, these fractions vary by model year, but are held stable after
2008. See Figure 3-21.
Figure 3-22 shows an example of the phase-in calculation for NO* from cars between model
years 2000 and 2010. The figure shows cold start and running FTP values for Tier 1, NLEV and
Tier 2 standards, as well as the phase-in fractions for each standard in each model year. Start and
running emissions in each model year are simply calculated as weighted averages of the
emissions estimates and the phase-in fractions. The resulting weighted start estimates are used
directly to represent cold-start emissions for young vehicles in each model year (ages 0-3). For
running emissions, however, the averages are not used directly; rather, each is expressed as a
ratio to the corresponding Tier-1 value.
Table 3-16 shows weighted average values for model-years 2001-2010 for simulated FTP
composites, cold-start and hot-running emissions. The start values, expressed as the cold-start
mass increment (g), are used directly in the MOVES emission rate table to represent cold-start
emissions (for operating mode 108). The composites and running emissions, expressed as rates
(g/mi), are presented for comparison. For running emissions, however, the averages shown in
the table are not used directly; rather, each is expressed as a ratio to the corresponding Tier-1
value, as shown in Figure 3-23 to Figure 3-25 below.
Model Year
Figure 3-21 Relative fractions of truck classes, by model year
¦ LDT1
¦ LDT2
¦ LDT3
¦ LDT4
69
-------
Standard Cold start Hot Running Phase-in by Model Year
(9)
(g/mi)
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Tier 1
Tier 1
0.888
0.127
1
0.011
0.004
0.002
0
0
0
0
0
0
0
TLEV
0.888
0.127
0
0.052
0.018
0.011
0
0
0
0
0
0
0
NLEV
LEV
0.566
0.040
0
0.801
0.752
0.613
0.175
0.110
0.132
0.103
0.070
0.035
0
ULEV
0.566
0.040
0
0.136
0.226
0.192
0.042
0
0
0
0
0
0
bin8
0.418
0.035
0
0
0
0.115
0.251
0.163
0.095
0.002
0
0
0
bin7
0.364
0.052
0
0
0
0.017
0.004
0.005
0.004
0
0
0
0
Tier 2
bin5
0.165
0.008
0
0
0
0.049
0.491
0.682
0.698
0.799
0.830
0.855
0.890
bin4
0.090
0.005
0
0
0
0
0.016
0.021
0.033
0.042
0.050
0.060
0.060
bin3
0.071
0.004
0
0
0
0
0.008
0.009
0.003
0.013
0.010
0.010
0.010
bin2
0.067
0.000
0
0
0
0
0
0.010
0.011
0.014
0.015
0.015
0.015
LEV-//
LEV
0.165
0.008
0
0
0
0
0.0052645
0.000
0.000
0.000
0.000
0.000
0.000
ULEV
0.071
0.004
0
0
0
0
0.0074988
0.000
0.024
0.026
0.025
0.025
0.025
Start (g)
0.888
0.586
0.573
0.530
0.314
0.248
0.237
0.199
0.185
0.170
0.156
Running (g/mile)
0.127
0.046
0.042
0.039
0.022
0.016
0.015
0.011
0.010
0.009
0.008
RATIO to Tier 1
1.00
0.36
0.33
0.31
0.17
0.13
0.12
0.087
0.079
0.070
0.061
Figure 3-22 Example of phase-in calculation, for NO* from cars (LDV), for MY 2000-2010
70
-------
regClass
Table 3-16 Weighted average FTP values for trucks and cars for MY 2001-2010
MY
Reference1 2000
Trucks
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Cars
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
CO
Comp.
(g/mi)
Start
(g)
Running
(g/mi)
1.62
11.4
0.805
1.43
12.6
0.566
1.41
12.4
0.552
1.47
12.7
0.586
0.923
7.92
0.393
0.783
7.05
0.315
0.697
6.12
0.296
0.664
5.85
0.281
0.647
5.75
0.270
0.632
5.67
0.260
0.618
5.58
0.251
0.8561
7.68
0.287
0.8206
7.27
0.284
0.8076
7.05
0.300
0.7141
6.16
0.298
0.6716
5.91
0.274
0.6566
5.85
0.257
0.6210
5.63
0.234
0.6114
5.55
0.232
0.6011
5.47
0.230
0.5915
5.38
0.229
THC
Comp.
(g/mi)
Start
(£)
Running
(g/mi)
NO*
Comp.
(g/mi)
Start
(g)
Running
(g/mi)
0.126
1.53
0.0571
0.209
0.888
0.127
0.0965
1.23
0.0400
0.171
0.843
0.0876
0.0942
1.21
0.0376
0.169
0.836
0.0865
0.1004
1.25
0.0424
0.181
0.863
0.0934
0.0535
0.786
0.0123
0.0849
0.473
0.0434
0.0440
0.703
0.00574
0.0596
0.367
0.0291
0.0378
0.612
0.00511
0.0381
0.264
0.0183
0.0361
0.587
0.00490
0.0315
0.226
0.0148
0.0356
0.580
0.00479
0.0285
0.208
0.0130
0.0350
0.571
0.00470
0.0258
0.192
0.0115
0.0345
0.564
0.00461
0.0233
0.177
0.0101
0.0361
0.954
0.00508
0.0948
0.586
0.0457
0.0333
0.893
0.00451
0.0898
0.573
0.0421
0.0340
0.839
0.00462
0.0824
0.530
0.0394
0.0360
0.664
0.00488
0.0461
0.315
0.0220
0.0358
0.634
0.00477
0.0351
0.248
0.0161
0.0350
0.633
0.00462
0.0335
0.239
0.0150
0.0341
0.608
0.00443
0.0271
0.201
0.0112
0.0341
0.592
0.00443
0.0248
0.187
0.0101
0.0339
0.574
0.00442
0.0224
0.172
0.00896
0.0339
0.557
0.00442
0.0201
0.158
0.00784
'The reference level for calculating ratios is MY 2000, representing cars (LDV) for Tier 1.
71
-------
to
cc
0.704 0.686
—Composite
—Cold Start
— Hot Running
0392 0368 °'349 0.335 0.322 0.311
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Model Year
1.2
1.1
1.0
0.9
0.8
O 0.7
* 0.6
^ 0.5
0.4
0.3
0.2
0.1
0.0
(b) Cars
1\
—V 1 A
—*—"—¦
0.357 0.353 0372
°'371 0.340 o 320
0.290 0.288 0.286 0.284
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Model Year
Figure 3-23 Weighted ratios for composite, start and running CO Emissions, for (a) trucks and (b) cars
72
-------
to
cc
1.2
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
1
(a) Trucks
0 701 0.74\
0 659 \\v
~ Composite
m Co Id Sta rt
¦ Hot Running
....
1 1 1
1 i 0.101 i 0.090 1 U.U8b 1 U.U84 1 U.U82
1 U.U81
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Model Year
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Model Year
Figure 3-24 Weighted ratios for FTP composite, start and running THC emissions, for (a) trucks and (b) cars
73
-------
1.2
1.1
1.0
0.9
0.8
O 0.7
* 0.6
oe 0.5
0.4
0.3
0.2
0.1
0.0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Model Year
Model Year
Figure 3-25 Weighted ratios for FTP composite, start and running NO.v emissions, for (a) trucks and (b) cars
1.2
1.1
1.0
0.9
0.8
O 0.7
* 0.6
^ 0.5
0.4
0.3
0.2
0.1
0.0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
74
-------
3.3.2.4 Estimating Emissions by Operating Mode
With the introduction of the NLEV standards, new emissions requirements were imposed, in
addition to standards defined in terms of the Federal Test Procedure. The new requirements,
under the "Supplemental Federal Test Procedure" (SFTP), imposed more stringent emissions
control under conditions of high speed and power (through the US06 cycle), and with air-
conditioning running (through the SC03 cycle).
In developing rates for use with MOVES2010 and MOVES2014, we attempted to explicitly
account for the effects of the SFTP standards. Due to a lack of "second-by-second" data on
vehicles certified to the NLEV (or Tier 2) standards at the time, distinct sets of "US06-based"
scaling factors were developed to represent emissions during "high-power" operation, which was
assumed to occur in a subset of six operating modes (28-30, 38-40).
This approach implied that the interaction of increasing stringent FTP standards (e.g., Bin 3, Bin
2, SULEV) with non-changing SFTP standards would increase the steepness of emissions trends
with increasing VSP over approximately 18 kW/Mg.
More recently, the availability of second-by-second data measured on vehicles certified to Tier-2
standards has enabled us to reassess this assumption (see Section 3.3.2.4.1). Our review of these
data suggests that the expected offsets in emissions trends with power are not observable.
Accordingly, we have modified rates for the current release by removing the 'US06-based'
scaling factors.
Thus, in MOVES3 to estimate emissions by operating mode, the approach was to multiply the
emission rates for MY 2000, representing Tier 1, by a specific ratio for each model year from
2001 to 2010, to represent emissions for that year. For all operating modes, we applied a single
"hot-running" ratio as listed in Table 3-16 above.
Figure 3-26 and Figure 3-27 show application of the ratios to selected hot-running operating
modes in model years 2000 (the reference year), 2005, and 2010, both calculated with respect to
2000. In these figures, the results are presented on both linear and logarithmic scales. The linear
plots more clearly display the differences at high-power, but obscure those at lower power. The
logarithmic plots supplement the linear plots by making visible the relatively small differences
between MY 2005 and 2010 in the lower power modes.
75
-------
1,800
1,600
1,400
1,200 --
1,000
ra
% 800
i/)
"J
m 600 4-
c
ra
g 400
200
0
: (a) CO
2000
2005
-B-2010
i ll
**¦¦¦¦ 1
rr.TTi
¦ ¦ ¦ ¦ 1
1 ¦ ¦ ¦ ¦ 1
1 ¦ ¦ ¦ ¦
5 10 15 20 25 30
Vehicle-Specific Power (kW/Mg)
35
40
20.00
S 12.00 ¦¦
cu
ra
oe
cu
I/)
ro
CO
c
ro
QJ
E
o.oo
: (b)THC
-~-2000
-¦-2005
-B 2010
: »—
h-TT .1
0, ¦ ."]
¦!
111111
¦ ¦ ¦ ¦
10 15 20 25 30
Vehicle-Specific Power (kW/Mg)
35
40
120
^100 --
ro
DC
0>
I/)
ra
CQ
c
ra
cu
E
: (c) NOx
—~— zOOO
2005
-B-2010
I I LI ¦
rrfin
¦ ~ 11
S=i=F^
h-rr.-
~
5 10 15 20 25 30
Vehicle-Specific Power (kW/Mg)
35
40
Figure 3-26 Projected emission rates for cars, vs. VSP, for three model years (LINEAR SCALE). (NOTE: rates
pictured represent operating modes 21-30 for ages 0-3 years)
76
-------
10,000
(13
en
-------
3.3.2.4.1 Evaluation of MOVES2014 "High-Power" Emission Rates
The removal of the MOVES2014 "US06 based" scaling factors described above was based on
analysis of recently available second-by-second high-power emission data for vehicles certified
to Tier 2 or equivalent standards. While the evaluation was not sufficient to develop new rates, it
demonstrated clearly that MOVES would better estimate high-power emissions without the
"US06-based" scaling.
One such dataset includes measurements on a set of light-duty vehicles collected by faculty and
students at North Carolina State University between 2008 and 20 1 8.32 The sample includes 205
vehicles. The vehicles range in model year from 1996 to 2018 and incorporate multiple
standards, including Tier 1 through Tier 3 and their LEV equivalents. Age and mileage at the
time of measurement range from 0 years or miles to 18 years and over 300,000 miles,
respectively. Gaseous emissions were measured using Clean Air Technologies (CATI) Montana
or Axion portable emissions measurement systems (PEMS) over a set of drive routes in the
Raleigh area covering approximately 110 miles.
A second dataset includes measurements on a set of light-duty vehicles taken at the USEPA
National Vehicle and Fuel Emissions Laboratory in Ann Arbor, MI. This vehicle sample is much
smaller, including 4 cars and 6 trucks. Gaseous emissions were measured using Sensors
SEMTECH portable instruments over routes comprising a variety of road types and driving
conditions around Ann Arbor, including freeway driving.
Binning and averaging the continuous data by vehicle specific power allows comparison of the
results with corresponding trends in MOVES2014 emission rates (see Figure 3-28). The trends
for "NCSU" represent a subsample of vehicles certified to Tier-2 or LEV-II standards, whereas
those for "EPA" represent several vehicles each.
For cars, the "NCSU" trend is noticeably lower than the MOVES2014 trend at "high power,"
i.e., above 20 kW/Mg, and shows a gentler increase in this range. The "EPA" trend increases
more aggressively than the "NCSU" trend but is still lower than the MOVES2014 trend. At "low
power," i.e., between ~3 to 20 kW/Mg, the "NCSU" trend shows a positive "convex" curve,
which is due to the presence of a single "high-emitting" vehicle. Absent this vehicle, the trend
would look very similar to that for trucks in this range.
For trucks, the MOVES trend is markedly higher than either of the PEMS trends, although the
"EPA" trend is more aggressive than the "NCSU" trend above -30 kW/Mg.
78
-------
Passenger Cars Passenger Trucks
Figure 3-28 Trends in NOx emissions for Cars in trucks, as measured by two PEMS instruments and as
represented in MOVES2014 emission rates.
Based on the results of this initial comparison, we compared the "real-world" PEMS results,
represented by the "NCSU" trends, to MOVES2014 rates and to a set of MOVES rates modified
by removal of the "US06-based" scaling factors, as shown in Figure 3-29 below for CO and NO*
for cars and trucks. In all cases, removal of "US06-based" scaling improved agreement with the
PEMS results.
Based on this finding, for MOVES3 we elected to revise the rates by removal of the "US06
scaling" for the subset of "high-power" modes, as described above. Work previously performed
to evaluate the projection of N0.v emissions by MOVES2014 had showed that high-power
operation contributes substantially to the light-duty NOT inventory in the National Emissions
Inventory.33 34 Note that some uncertainty remains in the estimation of emissions at high power,
as evidenced in part by the differences between measurements by the two PEMS instruments.
This topic requires additional evaluation after the release of MOVES3 as more data becomes
available.
79
-------
300
o>200
£
O
o
100
Cars
• MOVES2014
• MOVES2014-US06 reduction
¦ PEMS
CO
Trucks
: ¦
~
15
V10
CT)
E
x
O
z
Cars
• MOVES2014
• MOVES2014-US06 reduction
¦ PEMS
. t f
NOx
I
Trucks
Jt
10
20
30 0
VSP (kW/ton)
10
20
30
Figure 3-29 Trends in emissions with VSP for two vehicles classes and two pollutants, for MOVES2014 rates,
modified MOVES rates and NCSU PEMS results
3.4 MOVES2014 Emission-Rate Development (MY 2017 and later)
This section describes methods used in developing model rates for MOVES2014, representing
emissions from vehicles certified to Federal Tier-3 standards, in model years 2017 and later. This
material is retained because the MOVES2014 rates provide the basis for the updated MOVES3
rates, after being modified by adjustments as described in Sections 3.6, 3.7, and 3.10 below.
Methods used to develop rates to represent emissions for vehicles certified to Tier 3 standards
were identical to those used to develop rates for vehicles certified to NLEV and Tier 2 standards,
as described in Section 3.3 above, with several specific modifications. Where no modifications
to methods were made, we will refer the reader to the appropriate section of this report.
As previously described, the goal of this process is to represent I/M reference rates for the 0-3
year ageGroup. The rates are estimated by Tier, model year and regulatory class. The process
involves six steps previously described, repeated below for convenience.
1. Average FTP results by standard level and vehicle class. As before, we made use of data
measured on the FTP cycle in the course of the In-use Verification Program (IUVP).
2. Develop phase-in assumptions for MY 2017 - 2031, by standard level, vehicle class and
model year.
80
-------
3. Merge FTP results and Phase-in assumptions. For running emissions, calculate weighted
ratios of emissions in each model year relative to those for cars (LDV) in MY2000, which
represent Tier-1 LDV.
4. Estimate Emissions by Operating Mode. We calculated emissions by operating mode in each
model year by multiplying the MY2000 emission rates by the weighted ratio for each model
year.
5. Apply Deterioration to estimate emissions for three additional age Groups (4-5, 6-7 and 8-9).
for Tier 3 vehicles, we modified deterioration to represent an extended useful life of 150,000
miles, as opposed to the 120,000 mile duration assumed for NLEV and Tier 2 vehicles. We
therefore apply ln-linear deterioration to the rates developed in steps 1-4. For the remaining
three age groups, emissions are assumed to stabilize as previously described in Section 3.2.2.3
(page 52).
6. Estimate non-1ZMreference rates. The rates in steps 1-5 represent I/M references.
Corresponding non-I/M references are calculated by applying the ratios applied to the pre-Tier 3
rates.
We followed steps 1-6, with specific modifications to represent Tier 3 rates. In step 1, we
developed estimates of FTP results under Tier 3, including composite results, "cold-start"
emissions" (Bagl-Bag3) and "hot-running" emissions (Bag 2 FTP). For step 2, we developed
phase-in assumptions representing the introduction of Tier 3 standards. Each of these steps and
modifications is described in greater detail in the sub-sections below.
3.4.1 Averaging FTP Results (Step 1)
Projecting emissions for Tier 3 vehicles is driven by the NMOG+NO* standard, set at 30 mg/mi.
However, because MOVES projects NO* and THC emissions separately, we apportioned the
aggregate standard into NMOG and NO* components, which we will refer to as the "effective
standards" for each pollutant. For purposes of apportionment, we assumed that NMOG control
would pose a greater technical challenge than NO* control. Accordingly, we assumed "effective
standards" for NMOG and NO* of 20 mg/mi and 10 mg/mi, respectively. To implement this
assumption, we further assumed that for NO*, vehicles would be effectively brought into Tier 2
Bin 2, and that for NMOG, vehicles would be brought to a level between Bin 2 and Bin-3, but
closer to Bin 2.
In addition, MOVES models start and running processes separately. It is therefore necessary to
translate the composite standard into start and running components. One component represents a
"cold start" on the FTP cycle, represented as "Bagl - Bag3" emissions. A second component
represents "hot-running" emissions, represented by the hot-running phase of the FTP (Bag 2).
Estimated FTP emissions levels for hydrocarbons are shown in Table 3-17 for several Tier 2
Bins and for Tier 3. Values for all standards except Tier 3 are identical to those used to develop
rates in the default database. The cold start and hot running values for Tier 3 are calculated as a
weighted average of those for Bins 2 and 3, using Equation 3-34 with the bin weighting factors
selected such that they give the required value for the Tier 3 FTP composite.
81
-------
T3 = 0.775-B2 +0.225-B3
Equation 3-34
Table 3-17 Hydrocarbons (HC): useful-life FTP standards and associated cold-start and hot-running results
Bin
Useful-life Standard
FTP Composite1
FTP Cold Start1
FTP hot Running1
(mg/mi)
(mg/mi)
(mg)
(Bag 2)
(mg/mi)
8
125
41.3
591
3.56
5
90
35.5
534
2.63
4
70
24.8
383
2.28
3
55
21.5
329
1.74
2
10
5.6
87
0.42
Tier 32
20
9.2
142
0.7
1 Values represent "non-methane organic gases" (NMOG).
2 Values for Tier 3 calculated using Equation 3-34.
Under a general assumption that CO standards are not forcing, but that CO emissions tend to
track NMOG emissions, corresponding values for CO were calculated in the same manner, and
are presented in Table 3-18.
Table 3-18 CO: Useful-life FTP standards and associated cold-start and hot-running results on the FTP and
Bin
Useful-life Standard
(mg/mi)
FTP Composite
(mg/mi)
Cold Start
(mg)
FTP hot Running
(Bag 2)
(mg/mi)
8
4,200
861
6,680
451
5
4,200
606
5,510
238
4
4,200
537
5,500
201
3
2,100
463
3,470
119
2
2,100
235
1,620
70
Tier 31
2,100
286
2,040
81
1 Values for Tier 3 calculated using Equation 3-34.
Corresponding results for NOx are presented in Table 3-19. In contrast to HC and CO, the values
for Bin 2 were adopted for Tier 3, as the FTP composite of 5.5 mg/mi suggests that Bin-2
vehicles gives a compliance margin of about 50 percent with respect to the "effective standard"
of 10 mg/mi.
82
-------
Table 3-19 NO*: Useful-life FTP standards and associated cold-start and hot-running results on the FTP and
US06 cycles
Bin
Useful-life Standard
FTP Composite
Cold Start
FTP hot Running
(mg/mi)
(mg/mi)
(mg)
(Bag 2)
(mg/mi)
8
200
64.2
418
35.1
5
70
21.2
165
8.2
4
40
8.7
90
4.7
3
30
5.7
71
3.8
2
20
5.5
67
0.4
Tier 3
10
5.5
67
0.4
3.4.2 Develop Tier 3 Phase-In Assumptions (Step 2)
We designed phase-in assumptions so as to project compliance with the Tier 3 fleet average
NMOG+NOi requirements. The requirements are shown in Table 3-20 for cars and trucks. The
phase-in begins in model year 2017 and ends in model year 2025.
Table 3-20 Target NMOG+NO^fleet average requirements for the Federal Test Procedure
Model year
FTP Composite, NMOG+NO* (g/mi)
LDV/T1
LDT21
2017
0.086
0.101
2018
0.079
0.092
2019
0.072
0.083
2020
0.065
0.074
2021
0.058
0.065
2022
0.051
0.056
2023
0.044
0.047
2024
0.038
0.038
2025
0.030
0.030
1 Throughout these results applied to Federal truck classes
LDT2, LDT3 and LDT4.
These results are also pictured in Figure 3-30. Note the sharp drop in emissions at the outset of
the Tier 3 phase-in, also that the truck requirements (LDT2,3,4) are slightly higher than those for
the lighter vehicles (LDV-T1). After 2017, the reduction in the fleet average is linear, and at the
completion of the phase-in, the fleet averages for cars and trucks no longer differ.
83
-------
Figure 3-30 NMOG+NO* FTP fleet average requirements during phase-in of the Tier 3 exhaust emissions
standards for light-duty vehicles
CUD
05
~o
c
05
+-<
(/)
-------
100%
90%
80%
c
o
70%
"+j
"i/i
60%
o
Q.
E
50%
o
u
40%
at
>
"+j
30%
ns
at
cc
20%
10%
0%
II
T3
I Bin 2
Bin 3
I Bin 4
Bin 5
II
rsi
O
rsi
rsi
O
rsi
Model Year
Figure 3-31 Phase-in assumptions, by standard and bin, for LDV-T1 vehicles
100%
90%
80%
c
o
70%
"+j
"i/i
60%
o
a.
E
50%
o
u
40%
at
>
"+j
30%
ns
at
cc
20%
10%
0%
T3
Bin 3
I Bin 4
Bin 5
Cs|
O
rsi
LD
rsi
O
rsi
Model Year
Figure 3-32 Phase-in assumptions, by standard and bin, for LDT2 vehicles
85
-------
SO
oS
C
O
o
Q.
E
o
u
cu
,>
*+-»
_ro
cu
SO
oS
C
O
1/1
o
Q.
E
o
u
cu
>
cu
Q£
Model Year
Figure 3-33 Phase-in assumptions, by standard and bin, for LDT3 vehicles
Model Year
Figure 3-34 Phase-in assumptions, by standard and bin, for LDT4 vehicles
86
-------
3.4.3 Merge Cycle Results and Phase-In Assumptions (Step 3)
The goal of this step is to calculate weighted averages of the FTP (cold-start and hot-running)
results for all standards in each model year, with the emissions results weighted by applicable
phase-in fractions. We do this step for each vehicle class separately, then weight the four truck
classes together using a set of fractions also derived from the weighted sales estimates. See
Figure 3-21 (page 69).
Figure 3-35 shows an example of the phase-in calculation for NO* from cars between model
years 2016 and 2025. The figure shows cold-start and hot-running FTP values for Tier-1, Tier 2
and Tier 3 standards, as well as the phase-in fractions for each standard in each model year. Start
and running emissions in each model year are simply calculated as weighted averages of the
emissions estimates and the phase-in fractions. The resulting weighted start estimates are used
directly to represent cold-start emissions for young vehicles in each model year (ages 0-3). For
running emissions, however, the averages are not used directly; rather, each is expressed as a
ratio to the corresponding Tier-1 value.
Table 3-21 shows weighted average values for model-years 2016-2025 for simulated FTP
composites, cold-start and hot-running emissions. The start values, expressed as the cold-start
mass increment (g), are used directly in the MOVES emission rate table to represent cold-start
emissions (operating mode 108). The composites and running emissions, expressed as rates
(g/mi), are presented for comparison. For running emissions, however, the averages shown in
the table are not used directly; rather, each is expressed as a ratio to the corresponding Tier-1
value, as shown in Figure 3-36 to Figure 3-38 below.
Standard
Cold
Hot
Phase-In by Model Year
Start Running
(ms)
(nig/nii)
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
Tier 1
Tier 1
888.00
127.00
0
0
0
0
0
0
0
0
0
0
Bill 8
417.87
35.07
0
0
0
0
0
0
0
0
0
0
Bill 5
165.42
8.21
0.890
0.407
0.356
0.305
0.254
0.204
0.153
0.102
0.058
0
Tier 2
Bill 4
89.72
4.69
0.060
0.027
0.024
0.021
0.017
0.014
0.010
0.007
0.004
0
Bill 3
70.89
3.78
0.010
0.016
0.014
0.012
0.010
0.008
0.006
0.004
0.002
0
Bill 2
67.18
0.38
0.015
0.007
0.006
0.005
0.004
0.003
0.003
0.002
0.001
0
Tier 3
Tier 3
67.18
0.38
0.000
0.543
0.600
0.657
0.714
0.771
0.829
0.886
0.935
1.000
r
Cold Start (nig)
r
154.32
r
107.85
r
102.76
r
97.69
r
92.60
r
87.52
r r r
82.43 77.35 72.99 67.18
Hot Running (nig/nii)
7.64
3.74
3.32
2.90
2.48
2.06
1.64 1.22 0.86 0.38
RATIO to Tier 1
0.0601
0.0295
0.0262
0.0228
0.0195
0.0162
0.0129 0.00961 0.00677 0.00299
Figure 3-35 Example of phase-in calculation, for NO* from LDV-T1, for MY 2016-2025
87
-------
Table 3-21 Weighted average FTP values projected for trucks and cars for MY 2017-2025
regClass
MY
CO
THC
NO*
Composite
Start
Running
Composite
Start
Running
Composite
Start
Running
(mg/mi)
(mg)
(mg/mi)
(mg/mi)
(mg)
(mg/mi)
(mg/mi)
(mg)
(mg/mi)
Ref.1
2000
1,620
11,400
805
126
1,530
57.1
209
888
127
Trucks
2017
541
4,749
213
28.3
462
3.82
19.6
154
8.24
2018
434
3,625
155
20.7
341
2.76
13.1
114
4.45
2019
412
3,395
144
19.0
314
2.54
12.0
108
3.86
2020
391
3,164
134
17.3
287
2.32
10.9
101
3.27
2021
369
2,934
123
15.7
260
2.10
9.82
93.9
2.68
2022
348
2,704
112
14.0
233
1.88
8.72
87.0
2.09
2023
327
2,474
101
12.3
206
1.66
7.62
80.2
1.50
2024
305
2,246
91
10.7
179
1.45
6.53
73.4
0.91
2025
286
2,037
81
9.2
154
1.25
5.54
67.2
0.38
Cars
2017
426
3,566
149
20.5
339
2.70
12.0
108
3.74
2018
408
3,375
140
19.1
316
2.52
11.2
103
3.32
2019
391
3,184
132
17.7
293
2.34
10.4
97.7
2.90
2020
373
2,993
123
16.3
270
2.16
9.60
92.6
5.48
2021
356
2,802
115
14.8
247
1.97
8.77
87.5
2.06
2022
338
2,610
106
13.4
224
1.79
7.96
82.4
1.64
2023
321
2,419
98
12.0
201
1.61
7.16
77.4
1.22
2024
306
2,255
91
10.8
181
1.46
6.46
73.0
0.86
2025
286
2,037
81
9.2
154
1.25
5.54
67.2
0.38
1 The reference level represents Tier-1 LDV-T1.
88
-------
0.45
0.40
0.35
0.30
.2 0.25
2 0.20
0.15
0.10
0.05
0.00
(a) Trucks
0.265
0.193
0.179
0.166
0.153
0.139
0.126
¦Composite
¦Cold Start
¦ Hot Running
0.113
0.101
1 1 1 1 1 1 1 1
2017 2018 2019 2020 2021 2022 2023 2024 2025
Model Year
0.45
0.40
0.35
0.30
.2 0.25
5 0.20
0.15
0.10
0.05
0.00
(b) Cars
0.132
0 122 0.113
0.101
1 1 1 1 1 1 1 1
2017 2018 2019 2020 2021 2022 2023 2024 2025
Model Year
Figure 3-36 Weighted ratios for composite, start and running CO emissions, for (a) trucks and (b) cars
89
-------
0.35
0.30
0.25
0 0.20
1 0.15
0.10
0.05
0.00
2017 2018 2019 2020 2021 2022 2023 2024 2025
Model Year
0.35
0.30
0.25
O 0.20
'+¦»
CO
CC 0.15
0.10
0.05
0.00
Figure 3-37 Weighted ratios for composite, start and running THC emissions, for (a) trucks and (b) cars
(b) Cars
2017 2018 2019 2020 2021 2022 2023 2024 2025
Model Year
90
-------
(a) Trucks
¦Composite
Cold Start
¦ Hot Running
2017 2018 2019
2020 2021 2022
Model Year
2023
0.007 QX]Q2_
2024
0.20
0.18
0.16
0.14
O °"12
0.10
K 0.08
0.06
0.04
0.02
0.00
Model Year
Figure 3-38 Weighted ratios for composite, start and running NO* emissions, for (a) trucks and (b) cars
3.4.4 Estimating Emissions by Operating Mode (Step 4)
To project modal emissions for Tier 2 and Tier 3 vehicles, the approach was to multiply the
emission rates for MY 2000, representing Tier 1, by a specific ratio for each model year from
2016 to 2025, to represent emissions for that model year. For all operating modes, we applied
the ratios shown in the three figures immediately above.
91
-------
Figure 3-39 and Figure 3-40 show application of the ratios for cars in model years 2000, 2005,
2010, 2017, and 2025, representing Tier 1 standards, partially phased-in Tier 2 standards, fully
phased-in Tier 2 standards, an interim year during the Tier 3 phase-in, and the fully phased-in
Tier 3 standards, respectively. Rates for all five model years are calculated with respect to rates
for cars in model-year 2000 (using reduction ratios described above, applied to selected
operating modes for running operation. In these figures, the results are presented on both linear
and logarithmic scales. The linear plots display the differences in the high-power modes, but
obscure those in the low-power modes. The logarithmic plots supplement the linear plots by
making visible the relatively small differences in the lower power modes. In addition, the
logarithmic plots include the level for MY2000, which represents Tier-1 standards. Thus, these
plots display the degree of running-emissions reduction between Tierl and Tier 2 (MY2000:
MY2010), and between Tier 2 and Tier 3 (MY2010: MY2025), across the full range of vehicle-
specific power. Note that for simplicity, these figures represent rates for operating modes 21-30,
covering a wide range of power for vehicles operating at speeds between 25 and 50 mph.
92
-------
10 15 20 25 30
Vehicle-Specific Power (kW/Mg)
ao
20.00
18.00
16.00
14.00
12.00
10.00
8.00
6.00
4.00
2.00
0.00
i" (b) THC
-~-2000
—¦—2005
-0-2010
—•—2017
-*-2025
11 iTi
¦ ¦ ¦ ¦
10
15
20
25
30
35
40
Vehicle-Specific Power (kW/Mg)
120
T- 100 --
ra
cc
01
(0
CQ
c
(0
01
E
80 --
60 --
40 --
20 --
: (c) NOx
-~-2000
2005
-0-2010
—•—2017 ^
-*-2025
*
—a—
~
~ . .
F. . .*H
-5
5 10 15 20 25 30
Vehicle-Specific Power (kW/Mg)
35
40
Figure 3-39 Projected emission rates for cars in operating modes 21-30, vs. VSP, in ageGroup 0-3 years, for
five model years, for (a) CO, (b) THC and (c) NO* (LINEAR SCALE)
93
-------
10,000
1,000 --
ro
DC
-------
3.4.5 Apply Deterioration (Step 5)
Based on review and analysis of data from the Phoenix Inspection-and-Maintenance Program,
we assume that deterioration for different technologies is best represented by a multiplicative
model, in which different technologies, represented by successive model-year groups, show
similar deterioration in relative terms but markedly different deterioration in absolute terms. We
implemented this approach by translating emissions for the 0-3 age Group, as calculated above,
into their respective logarithmic means and applying uniform logarithmic age trends to all
model-year groups. We derived logarithmic deterioration slopes for Tier-1 vehicles (MY 1996-
98) and applied them to Tier 2 vehicles. In this process we applied the same logarithmic slope to
each operating mode, which is an extension of the multiplicative deterioration assumption.
3.4.5.1 Recalculate the Logarithmic Mean
Starting with the values of the arithmetic mean (xt1) calculated as described in step 4 above, we
calculate a logarithmic mean (x/), as previously shown in Equation 3-28 (page 41).
3.4.5.2 Apply a Logarithmic Age Slope
After estimating logarithmic means for the 0-3 age class (X/.0-3), we estimate additional
logarithmic means for successive age classes (x/.age), by applying a linear slope in ln-space (mi),
using Equation 3-29 (page 41).
The values of the logarithmic slope are adapted from values developed for the 1996-98 model -
year group. The values applied to Tier 2 and Tier 3 vehicles are shown in Table 3-22. The
reduced slopes for Tier 3 were calculated by reducing the Tier 2 values by 27 percent for THC
and CO and by 14 percent for NO*. These values were estimated empirically so as to implement
the assumption of reduced deterioration for the extended useful life. When calculating the age
inputs for this equation, we subtracted 1.5 years to shift the intercept to the midpoint of the 0-3
year age Group.
Table 3-22 Values of the logarithmic deterioration slope applied to running-exhaust emission rates for MY
Pollutant
Logarithmic Slope (mi)
Tier 2
Tier 3
CO
0.13
0.0949
THC
0.09
0.0657
NO,
0.15
0.129
3.4.5.3 Apply the Reverse Transformation
After the previous step, the values of x/.age were reverse-transformed, as shown in Equation 3-28
(page 41).
95
-------
3.4.5.4 Adjust to Account for Averaging with Electric Vehicles
Unlike MOVES3, MOVES4 includes electric vehicles in the default fleet2 and accounts for
projected increases in the running and start THC and NOx emissions from conventional light-
duty vehicles assuming manufacturers take advantage of the fleet-wide averaging allowed by
EPA Tier 3 regulations. These adjustments are not included in the EmissionRateByAge table;
instead, they are applied later in MOVES processing as explained in the MOVES4 adjustments
report.3
3.4.6 Estimate Non-I/M References (Step 6)
Completion of the preceding steps provided a set of rates representing I/M reference rates for
MY 2016-2025. As a final step, we estimated non-I/M reference rates by applying the same
ratios applied to the I/M references for default rates (Section 3.5, page 96).
3.4.7 Start Emissions
The values for "FTP Cold-start" shown in Table 3-16 (page 71) and Table 3-21 (page 88) were
used to represent cold-start emissions (opModeID=108). Rates for "warm" or "hot" starts
following a range of soak periods were estimated as for the default rates (see Section 3.9.2, page
171 ). Deterioration was applied to start emissions, relative to that for running emissions, also as
described (see below in Section 3.9.3.3, page 199).
3.5 Estimating Rates for Non-I/M Areas
In modeling emission inventory for light-duty vehicles, it is necessary at the outset to consider
the question of the influence of inspection-and-maintenance (I/M) programs. In MOVES, two
sets of rates are stored in the input table (emissionRateByAge). One set represents emissions
under "I/M conditions" (meanBaseRatelM) and the other represents rates under "non-I/M
conditions" (meanBaseRate). The first set, representing vehicles subject to I/M requirements, we
call the "I/M reference rates". The second, representing vehicles not subject to I/M requirements,
we call the "non-I/M reference rates."
For the I/M reference rates, the term "reference" is used because the rates represent a particular
program, with a specific design characteristics, against which other programs with differing
characteristics can be modeled. Thus, the I/M references are, strictly speaking, regional rates, and
not intended to be (necessarily) nationally representative.
Our approach is to derive the non-I/M rates relative to the I/M references, by adjustment. One
reason for adopting this approach is that, as mentioned, the volumes of data available in I/M
areas vastly exceed those collected in non-I/M areas. An additional practical reason is that major
work-intensive steps such as "hole-filling" and projection of deterioration need only be
performed once.
In contrast to the I/M references, the non-I/M reference rates are designed to be nationally
representative. Broadly speaking, they are intended to represent all areas in the country without
I/M programs. In general, estimating the influence of I/M areas on mean emissions is not trivial,
and efforts to do so commonly follow one of two broad approaches. One approach is to compare
emissions for two geographic areas, one with and one without I/M, as shown in Figure 3-41(a).
A second and less common approach is to compare emissions between two groups of vehicles
96
-------
within the same I/M area, but with one group representing the main fleet ostensibly influenced
by the program, and the second, far smaller, representing vehicles measured within the program
but presumably not yet influenced by the program, as shown in Figure 3-42(b).
Figure 3-41 General approaches to estimating differences attributable to I/M programs: (a) comparison of
subsets of vehicles between two geographic areas, and (b) comparison within a program area
For convenience, we refer to the first approach as the "between-area" approach, and the second
as the "within-area" approach. Neither approach attempts to measure the incremental difference
attributable to a program from one cycle to the next.
The approach we adopted emphasizes the "within-area" approach, based on a sample of vehicles
"migrating" into Phoenix and entering that program. Characteristics of the Phoenix program
during 1995-2005 are listed below.
• A four-year exemption period,
• transient tailpipe tests for MY 81-95,
• OBD-II for MY 96+,
• biennial test frequency.
To lay the basis for comparison, the primary goal was to identify a set of vehicles that had been
measured by the program after moving into the Phoenix area, but that had not yet been
influenced by the program. The specific criteria to identify particular migrating vehicles are
presented in Table 3-23.
(a) Comparison between a program Area
and a non-program area
(b) Comparison within a program
A-ea
97
-------
Table 3-23 Criteria used to identify vehicles migrating into the Phoenix program
logic
Criterion
The vehicle comes from out-of-state
OR
from a non-I/M county in AZ
AND NOT
from other I/M areas
AND
receiving very first test in Phoenix program
AND
selected for the random evaluation sample
After applying these criteria, we identified a sample of approximately 1,400 vehicles. The origin
of vehicles entering the Phoenix Area was traced by following registration histories of a set of
approximately 10,000 candidate vehicles. The last registered location of vehicles was identified
prior to registration in Phoenix or the vehicle's first test in the Phoenix program. Vehicles were
excluded if their most recent registration location was in a state or city with an I/M program.35
Figure 3-42 shows the distribution of incoming non-I/M vehicles by Census Region. Most
vehicles migrating to Phoenix came from the Midwest (47 percent), followed by the South (32
percent), the West (20 percent) and the Northeast (1 percent). The low incidence from the NE
may be attributable to the large number of I/M programs in that region.
98
-------
WEST
MIDWEST
NORTHEAST
Pacific
Mountain
West
North Central
East Middle New
North Central Atlantic England
South
Atlantic
South Central South Central
SOUTH
Figure 3-42 Geographic distribution of non-I/M vehicles migrating into the Phoenix I/M area, 1995-2005
To assess the emissions differences between migrating (non-I/M) and "local" (I/M) vehicles, we
adopted a simple approach. We calculated ratios between means for the migrating and local
groups, as shown in Equation 3-35. We used aggregate tests, after preliminary analyses
suggested that the ratios did not vary significantly by VSP. Because the sample was not large in
relation to the degree of variability involved, we also aggregated tests for cars and trucks in all
model years. However, we did calculate ratios separately for three broad age groups (0-4, 5-9,
and 10+) years.
For purposes of verification, we compared our results to previous work. An initial and obvious
comparison was to previous work by Thomas Wenzel based on an out-of-state fleet migrating
into Phoenix that provided a model for our own analysis.36 This previous effort identified a
migrating fleet, and analyzed differences between it and the program fleet for vehicles in model
years 1984-1994 measured during calendar years 1995-2001. To adapt the previous results for
our purposes, we translated averages for migrating and program fleets into ratios as in Equation
Another valuable source for comparison was remote-sensing data collected in the course of the
Continuous Atlanta Fleet Evaluation (CAFE) Program.37 38 Unlike our own analysis, this
program involves a comparison between two geographic areas. The "I/M area" is the thirteen-
Ratio = K--,sl
f
1 J t '\ r
:J/y:
Equation 3-35
3-35.
99
-------
county Atlanta area, represented by measurements for approximately 129,000 vehicles. The other
area is the twelve-county "non-I/M area," surrounding Atlanta, represented by measurements for
approximately 28,000 vehicles. Both areas had been under a low-sulfur fuel requirement since
1999. Results used for this analysis were collected during CY 2004. The non-I/M: I/M ratios
calculated from the remote-sensing are based on concentrations, rather than mass rates.
A third source was an additional remote-sensing dataset collected in N. Virginia/D.C. area. The
I/M area was the "northern-Virginia" counties, and the non-I/M area was Richmond. The I/M
and non-I/M areas were represented by about 94,000 and 61,000 vehicles, respectively, collected
in CY 2004. In this case, the molar ratios were converted to mass rates, with use of fuel-
consumption estimates derived from energy-consumption rates in MOVES2004. After this step,
non-I/M:I/M ratios were calculated using the mass rates.
Results are shown in Figure 3-43. The charts show mean ratios for the three age groups for our
migrating vehicle analysis, as well as the remote-sensing studies. The diamonds represent
approximate values from Wenzel's earlier work with the Phoenix data. For our analyses (solid
bars), the ratios are generally lower for the 0-4 year age group, and larger for the 5-9 and 10+ age
groups, but differences between the two older groups are small. The Atlanta results show a
similar pattern for THC and NOx, but not for CO, for which the ratios are very similar for all
three age groups. The Virginia results are the other hand, show increasing trends for CO and
THC, but not for NOx. The ratios in Atlanta are slightly higher than those for Phoenix in the 0-4
year age group. This difference may be attributable to the shorter exemption period in Atlanta (2
years) vs. the four-year period in Phoenix, but it is not clear that these differences are statistically
significant. In all three programs, ratios for the two older age classes generally appear to be
statistically significant.
In interpreting the ratios derived from the Phoenix data, it is important to note that they assume
full program compliance. In the migrating vehicle analysis this is the case because all emissions
measurements were collected in I/M lanes. Thus, vehicle owners who evaded the program in one
way or another would not be represented. On the whole, results from multiple datasets, using
different methods, showed broad agreement.
If we calculate non-IM reference rates from the I/M references by ratio, with the ratios constant
by model-year group and VSP, it follows that the absolute differences must increase with power.
Similarly, absolute differences increase with age, for two reasons. The first reason is the same as
that for VSP, that for a constant ratio, the absolute difference increases as emissions themselves
increase, and in addition, the second reason is that the ratios themselves increase with age (Figure
3-43). And, because these ratios are applied to calculate non-I/M rates for all model year groups
in MOVES, a third implication is the absolute differences would be smaller for successive
model-year groups as tailpipe emissions decline with more stringent standards.
100
-------
0-4
5-9
Age Class
10+
i AZ l/M ¦& GA RSD (CY04) = VA RSD (CY04)
1.80
i.eo-|-(b)-mc
1.40
i AZ l/M _ GA RSD (CY04) ¦ VA RSD (Cy04)
1.60
0-4 5-9 10+
Age Class
i AZ l/M ¦ GA RSD (Cy04) - VA RSD (Cv04)
Figure 3-43 Non-I/M: I/M ratios for CO, THC and NO.v for the Phoenix area (this analysis) compared to
remote-sensing results for Atlanta and N. Virginia, and previous work in Phoenix (diamonds)
A final practical step is to translate these results into terms corresponding to the MOVES age
groups. As mentioned, the program in Phoenix has a four-year exemption period for new
vehicles. However, it is not uncommon for other programs to have shorter exemptions; for
example, both the Atlanta and N. VA programs have two-year exemptions.
An additional factor is that the coarser age groups used for the migrating-vehicle analysis don't
mesh cleanly with the MOVES age groups. It was therefore necessary to impute values to the
101
-------
first two MOVES age groups (0-3 and 4-5 years). We achieved this step by linearly interpolating
the value for the 5-9 year age Group to a value of 1.0 at 0 years of age, as shown in Figure 3-44.
To anchor the interpolation, we associated the value of the ratio for the 5-9 year age group with
the midpoint of the group (7.5 years). Then, based on a straight line interpolation, we imputed
values for the 0-3 and 4-5 MOVES age groups, by taking the value on the line associated with
the midpoint of each class, 1.5 and 5 years, respectively.
0 - 4 years
5 - 9 >
'ears
/?5-9
i 1 1 1 1 1 1 1 1 1 1 1
1.5 5.0 7.5
0
|1 |2 |3
|4 |5 |6 |7 |8 |9
0-3 years
4-5 years 6-7 years 8-9 years
Figure 3-44 Imputation of non-I/M ratios for the 0-3 and 4-5 year MOVES ageGroups by linear interpolation
from the midpoint of the 5-9 year analysis age group
Figure 3-45 shows final values of the non-I/M ratios for CO, THC and NO*, with error-bars
representing 95 percent confidence intervals. The values for each pollutant start at 5.0 percent
and increase with age, stabilizing at maximum values at 6 years (for MX) and 10 years (for THC
and CO).
102
-------
0-3 4-5 6-7 8-9 10-14 15-19 20+
Age Class
0-3 4-5 6-7 8-9 10-14 15-19 20+
Age Class
-(c) NQx
T
~
T
T
T
T
r J
JL
""
10
CO
O
-
1
CNI
CNI
CNI
1—
CN|
CN|
0-3 4-5 6-7 8-9 10-14 15-19 20+
Age Class
Figure 3-45 Final non-I/M ratios for all model years for CO, THC and NO.v, by MOVES ageGroups, with 95
percent confidence intervals
103
-------
The ratios shown in Figure 3-45 are applied to the I/M reference rates to derive non-I/M reference
rates.
^7i,non-I/M ~ R^tiO* A/l[ x[
Equation 3-36
In addition, starting in MOVES3.1, we applied the ratio of the I/M and non I/M rates from
gasoline light-duty trucks to compute the HC, CO and NOx running and start emission I/M rates
for gasoline LHD2b3 trucks as detailed in the MOVES4 HD report.39
The uncertainty in Eh,wa-vu was calculated by propagating the uncertainty in the Ratio with that
of the corresponding I/M rate Emm.
8E
A,non-I/M
dR
SR +
r)F
/;,non-I/M
r)F
F s +
R£s
2 2
Equation 3-37
Thus, for any given cell h, the uncertainty in the non-I/M reference rate is larger than that for the
corresponding I/M reference rate, which is reasonable and appropriate given the additional
assumptions involved in developing the non-I/M reference rate.
104
-------
3.6 MOVES3 Running Exhaust Emission Rates (THC and NOxfor MY 1990
and later)
For MOVES3, light-duty THC and NO* emission rates for MY 1990-and-later were updated by
applying adjustments to the rates used in MOVES2014.40 We developed and applied two sets of
adjustments for these model years. The first is a set of adjustments that we applied to rates in the
first ageGroup, 0-3 years. For convenience, we will refer to the rates in the 0-3 year ageGroup as
"young vehicles," and the adjustments applied to them as "young-vehicle" adjustments. The
second set was applied to adjusted rates for young vehicles to project modified deterioration
assumptions for the remaining six ageGroups. Thus, the second set of adjustments will be
referred to as "deterioration adjustments."
We chose to modify the existing rates by adjustment so that the update could be completed in
time for release with MOVES3. The key motivators for this update was to reevaluate and modify
the deterioration assumptions in the MOVES2014 rates, which are very aggressive in some
cases.
However, at the time this update was initiated, the relevant datasets were not ready for use in
directly developing modal rates, i.e., the supporting analyses to evaluate time series alignment,
calculate vehicle-specific power and assign operating modes had not been completed.
Nonetheless, it was possible to analyze deterioration in these datasets on a non-modal basis, and
using the results, to propose modifications to the existing rates. These analyses and their
application are described in Section 3.6 for THC and NO*, and in Section 3.7 for CO.
3.6.1 Data Source
While the MOVES2014b rates for MY 1990-and-later were based on the same data and analysis
described above for the 1989-and-earlier model years, the MOVES3 updates for MY 1990-and-
later are based primarily on the Evaluation Sample for the Denver Metropolitan I/M program.
This source is recent, having been collected during the past decade, and includes a large body of
data directly measured on vehicles certified to Tier-2 standards. In addition, the Denver program
remains one of the very few programs that performs transient tailpipe testing and that has
compiled a random evaluation sample over a period long enough to enable a deterioration
analysis. During the past decade, most programs have transitioned to use of scans of the onboard
diagnostic system (OBD) as the basis for I/M tests. For example, the program in Chicago, which
was considered for MOVES2010, discontinued tailpipe testing by 2010.
As the name implies, "evaluation samples" are collected to provide a basis for evaluation of a
program's effectiveness. They involve the collection of vehicle samples at random, to ensure
representativeness, and the application of "full duration" tests with replication, to ensure that
results represent "hot running," or "fully conditioned" operation.
In addition, full-duration tests, in which the test cycle is run to completion (e.g., 240 sec on the
IM240 cycle) are needed to avoid the bias inherent in program test data in which the duration of
the test is proportional to vehicles' emissions levels. Such "fast-pass" or "fast-fail" bias is a
major obstacle to the use of program data, and precluded the use of data from the St. Louis
program in MOVES2010 (see 3.2.1, page 24).
105
-------
The data from the Denver evaluation sample used in the MOVES3 update was collected between
CY 2008 and 2017. The vehicle sample includes model years ranging from the early nineties
through 2010. The sample incorporates vehicle emission standards from Tier 1, National LEV,
and Tier 2, as well as their California counterparts LEV-I and LEV-II. In the evaluation sample,
vehicles selected at random receive two additional full-duration transient tests on the IM240
cycle, in addition to their official test. For purposes of analysis, we used only the second
replicate, to ensure that the data represented fully conditioned vehicles.
3.6.2 Vehicle classes
We analyzed emissions results for three classes of vehicles, which include passenger cars and
two classes of trucks, distinguished on the basis of gross-vehicle weight. These vehicle classes
are defined in Table 3-24.
Table 3-24 Definitions for Vehicle Classes in the Denver Evaluation Sample
Category
Vehicle Class
Description
GVWR (lb)
No. Tests
Cars
LDV
Light-Duty Vehicles
55,506
Trucks
LLDT
Light Light-Duty Trucks
0 < GVW <= 6,000
43,901
Trucks
HLDT
Heavy Light-Duty Trucks
6,000 < GVW <= 8,500
17,184
Total
116,591
The table shows totals numbers of vehicles in each class, for the subsets of data used for
analysis, spanning model years 1990-2010. These totals include hot-conditioned "second
replicates" only, following some exclusions for purposes of quality assurance. The total samples
are largest for cars, followed by the trucks, with HLDT having the smallest sample, roughly one
third of the total for cars.
The model-year by age distributions of the vehicle samples for each of these classes are shown in
Table 3-25 to Table 3-27 below. For each model year, the sample spans an age trend of nine
years. Note that the sampling effort is uneven throughout, but is highest for model years 2004
and later, during calendar years 2012-2016.
106
-------
Table 3-25 Sample of Passenger Cars (LDV) in the Denver Evaluation Sample
Sample of Passenger Cars from Denver l/M used for modeling
1990-
1991 -
1992-
1993-
1994-
1995-
1996-
1997-
fc 1998-
£ 1999-
03 2000-
E 2001 -
^ 2002-
2003-
2004-
2005-
2006-
2007-
2008-
2009-
2010-
114 113 123 12
110 144 90
143 151 150 19
121 156 88 17 21
169 145 150 19 9 12
205 148 27 19 18 54
32
253
100 I
2116
1690
1752
12
35
184
1Q10j
2383
1106
964
2
9
19
{10041693
444
1445
2
3
296
620
107
1438
204]
1
101
249
1246
24
92
446
> 1584
623
No. vehicles
¦ 2500
™ 2000
r 1500
1000
500
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
Age
Table 3-26 Sample of Light Light-Duty Trucks in the Denver Evaluation Sample
Sample of Light Trucks from Denver l/M used for modeling
1990-
1991 -
1992-
1993-
1994-
1995-
1996-
1997-
jg 1998-
1999-
¦gj 2000-
"g 2001-
^ 2002-
2003-
2004-
2005-
2006-
2007-
2008-
2009-
2010-
12
29
106
92 128
130 103 129
72 140 95 16
81 136 8 18
12 10 34 33 24
57
34
17 14 32 78 52
2 174
76 205
No. vehicles
I 2000
P 1500
1000
500
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
Age
107
-------
Table 3-27 Sample of Heavy Light-Duty Trucks in the Denver Evaluation Sample
Sample of Heavy Trucks from Denver l/M used for modeling
1990-
8
5
5 1
CM
1991 -
7
8
10
1
1
4 2
1992-
7
8
7
2
1
4
7
5
1993-
7
13
24
3
3
11
1
1994-
22
16
17
2
1
10
4
3
1995-
31
22
32
2
3
3
11
11 8
1996-
25
23
20
1
2
2
3
10
7
1997-
37
28
38
1
7
4
14
11
10
1998-
36 48
30
9
4
6
9
27
9
1999-
50
51 57
5
11
2
15
22
24
2000-
28
55
33 9
8
7
22
29
13
2001-
37
45
46
6 4
2
19
27
17
2002-
40
33
28
6
6
9
22
18
2003-
42
30
57
4
7
2 30
28
21
2004-
40
59
44
847
581
693
466 580
213
No.
vehicles
2005-
67
26
39
400
705
393
257
334 191
¦
800
2006-
10
78
44
663
515
522
175
411
160
600
400
2007-
4
12
50
390
780
396
282
280
171
2008-
7
10
786
406
607
146
442
132
2009-
1
76
519
154
23
315
63
200
2010-
1 31
103
491
12
34
139
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
Age
3.6.2.1 Clean Screen
In the Denver metropolitan area, a 'clean-screen' program is used to reduce the testing burden in
the inspection stations. This goal is achieved by performing remote sensing throughout the area
on an ongoing basis. Vehicles identified as "clean" are eligible to forgo the emissions inspection
at their next scheduled registration. Thus, the net effect of "clean-screen" should be to bias the
mean emissions levels for the measured fleet somewhat high, as "clean" vehicles are
preferentially screened out of the fleet reporting to the centers for regularly scheduled biennial
inspections.
We accounted for "clean-screen" by treating it as a secondary de facto sampling process, in
which the selection would be proportional to vehicles' emissions levels as measured by remote
sensing.
We estimated counts of vehicles eligible for clean screen that show up for testing. There are
two classes of such eligible vehicles. The first class includes vehicles in the evaluation sample
identified as "clean-screen" eligible but whose owners are intentionally not notified. This "hold-
back" sub-sample is intended to allow estimation of the emissions levels of eligible vehicles. The
second class includes eligible vehicles whose owners were notified that they need not report for
emissions inspections but who nonetheless reported to lanes and received inspections, i.e., "came
in anyway."
Within the evaluation sample, the fractions of eligible vehicles in a given model year that receive
emissions tests, out of the total of clean-screen eligible vehicles, is given by Equation 3-38,
TlH + 7lc
f = Equation 3-38
nH + nc + rip
where:
tin = eligible vehicles retained for program evaluation, i.e., "holdback" vehicles,
tic = eligible vehicles that received tests, i.e., "came in anyway,"
tip = eligible vehicles exempted from testing, i.e., "clean-screen participants."
108
-------
After calculating the fractions of clean-screen eligible vehicles undergoing tests, clean-screen
weights (wc) are calculated as their reciprocals, as shown in Equation 3-39:
1
wc = — Equation 3-39
These calculations were performed on the basis of model year, as shown in Table 3-28. This
reciprocal sample-weighting approach can be seen as an analog to non-response weighting in
analysis of a sample survey. The weights represent the numbers of eligible vehicles represented
by each eligible vehicle that underwent emissions measurements. For example, for model years
since 2004, each measured eligible vehicle, in group m + «c, represents approximately five
eligible vehicles that were exempted from the emissions inspection and were thus not measured.
All other vehicles in the evaluation sample not designated as clean-screen eligible were assigned
weights of 1.0, i.e., they represent "only themselves."
Table 3-28 Clean-Screen fractions and weights constructed for use with the Denver Evaluation Sample
Model Year
Clean-screen Fraction (f)
Clean-screen Weight (wc)
1990
0.182
5.49
1991
1.000
1.00
1992
0.167
5.99
1993
0.206
4.85
1994
0.023
43.5
1995
0.222
4.50
1996
0.108
9.26
1997
0.120
8.33
1998
0.112
8.93
1999
0.099
10.10
2000
0.113
8.85
2001
0.099
10.10
2002
0.095
10.53
2003
0.095
10.53
2004
0.221
4.52
2005
0.189
5.29
2006
0.200
5.00
2007
0.180
5.56
2008
0.208
4.81
2009
0.198
5.05
2010
0.197
5.08
109
-------
3.6.3 Data Review
Prior to analysis, we plotted the data for each combination of pollutant and vehicle class on both
linear and logarithmic scales. Review of the plots informs the process of model building and
selection.
The plots show four views of the data. In subplots (a) and (b), we plot individual measurements
for the entire dataset on linear and logarithmic scales, with simple linear trendlines by model
year. These trend lines give a sense of central tendency, i.e., where the means are situated within
the clouds of points, which are very broad.
The plots on linear scale demonstrate the strong degree of right skew within the emissions data.
They also display that small fractions of extreme high-emitting vehicles report to the lanes for
testing. Despite the undoubted tendency of some fraction of drivers to avoid or evade I/M
testing, large numbers of vehicle owners report to the lanes with high to very high emissions.
The plots also show that extremely high emissions can occur in vehicles that are quite young,
certified to low standards, and ostensibly within their regulatory useful lives.
The plots on logarithmic scale are more informative for modeling purposes. They display the
remarkably high degree of variability in emissions data, spanning several orders of magnitude. In
addition, the trendlines show the general parallelism in trends for successive model years.
Another important feature is that the trendlines give a broad indication of the shapes of long-term
emissions trends, showing how emissions first increase with age and then gently decline with
increasing age.
Review of these plots, supplemented by preliminary modeling of smaller subsets of model years,
led to the formulation of the spline model described below.
We also average the data by model year and age (panels (c) and (d)) and present the averages.
On the whole, the trends in means also reflect the broad picture in the plots of all measurements.
However, the trends in individual means are erratic, due to variation in sample sizes, and treating
each model-year x age combination as independent.
3.6.3.1 Oxides of Nitrogen (NO.v)
For NOi, sets of plots for the three vehicle classes are shown in Figure 3-46 for Passenger Cars,
Figure 3-47 for Light Light-Duty Trucks, and Figure 3-48 for Heavy Light-Duty trucks.
110
-------
Mod*) Year
— 1990
2001
-»• 1991
2002
* 1992 —
2003
2.5
— 1993 —
2004
g
— 1994 —
2006
Z
1994 —
2006
* 1M« ~
2007
#20
— 1997 —
2008
* 1998 —
2009
~ 1999 —
2010
— 2000
1 5
0
to 20
Age (years)
3
(d) Group means: log scale
Figure 3-46 N0X for Passenger Cars (LDV): IM240 Emissions (mg/mi) vs. age: (a) full data set, linear scale,
with simple trendlines by model year; (b) full dataset, common logarithmic scale, with simple trendlines by
model year; (c) means by model year and age, linear scale; (d) means by model year and age, common
logarithmic scale
111
-------
Figure 3-47 N0T for Light Light-Duty Trucks (LLDT): IM240 Emissions (mg/mi) vs. age: (a) full data set,
linear scale, with simple trendlines by model year; (b) full dataset, common logarithmic scale, with simple
trendlines by model year; (c) means by model year and age, linear scale; (d) means by model year and age,
common logarithmic scale
112
-------
(a) Full data: linear scale
Figure 3-48 N0T for Heavy Light-Duty Trucks (HLDT): IM240 Emissions (mg/mi) vs. age: (a) full data set,
linear scale, with simple trendlines by model year; (b) full dataset, common logarithmic scale, with simple
trendlines by model year; (c) means by model year and age, linear scale; (d) means by model year and age,
common logarithmic scale
3.6.3.2 Total Hydrocarbons (THC)
For THC, sets of plots for the three vehicle classes are shown in Figure 3-49 for passenger cars,
Figure 3-50 for Light Light-Duty Trucks and Figure 3-51 for Heavy Light-Duty Trucks.
113
-------
(a) Full data: linear scale
; 11 j i i I i i i ij j i i i
tool
— 2CKW
W — 2005
~ 20W
tOOT
W * ?9C®
Figure 3-49 THC for Passenger Cars (LDV): IM240 Emissions (mg/mi) vs. age: (a) full data set, linear scale,
with simple trendlines by model year; (b) full dataset, common logarithmic scale, with simple trendlines by
model year; (c) means by model year and age, linear scale; (d) means by model year and age, common
logarithmic scale
114
-------
30CCC
(a) Full data: linear scale
(c) Group means: linear scale
aooco*
3000
MocM Year
— 1990 — 2001
1991 2002
30CQ3-
— 1992 2003
|
— 1993 —- 20C4
5 2000
|
•
— 1994 — 2005 «
1995 -W- JQQ£ *
\ I
U
^ 20000
— 1996 — 2007 jE
1
— 1997 — 2006
| I
¦ ;
— 1999 — 2010
IS
WOQ0-
• t m
— 2000
.,;: 111 i I i m i i 1! i i; 11: i s.
0
0
0 to 20
o to
20
Age
Age (years)
Figure 3-50 THC for Light Light-Duty Trucks (LLDT); IM240 Emissions (mg/mi) vs. age: (a) full data set,
linear scale, with simple trendlines by model year; (b) full dataset, common logarithmic scale, with simple
trendlines by model year; (c) means by model year and age, linear scale; (d) means by model year and age,
common logarithmic scale
115
-------
(a) Full data: linear scale
i¦i i jijIU-
Model Year
«•» 1990 — 2001
— 1991 2008
—- (992 20OJ
— >990 — 2004
— I9M — 20OS
~ 1996 -•» 2006
— 1996 — 2007
— 1907 — 2006
— 1908 — 2009
*— 1999 * 2010
— 2000
I
-------
For a single model year, the three-piece model is defined by the equation
lny = b0 + b1m + b2a + b3(a — fc1)d2 + b4(a — k2)d3 + e Equation 3-40
where:
lny = natural logarithm of IM240 cycle aggregate emissions (mg/mi),
bo = grand intercept for a reference model year, assigned as the most recent model year, 2010,
6i.m = incremental intercept coefficient for model year m, as difference from bo,
m = model year as a class or categorical (dummy) variable,
62 = coefficient for age at test (a) as a continuous predictor (yr),
ki,k2 = knots where linear segments meet,
63 = incremental difference in slope for predictor a between k\ and £2,
64 = incremental difference in slope for predictor a above £2,
d2 = 1, if a > k\, else = 0,
d?, = 1, if a > £2, else = 0.
The predictor variables are age and model year, with age (a) fit as a continuous variable and
model year (m) fit as a class or categorical variable. The model structure assumes that the
logarithmic age trend is uniform across model years within a segment, and that each model year
has a distinct intercept.
Resolving the equation for each segment gives expressions for intercepts and slopes within each
of the three segments, as shown in Table 3-29.
Table 3-29 Expressions for intercept and slope parameters in the three-piece spline model
Segment
d\
Intercept
Slope
0 < a
-------
3.6.4.1 Optimizing the Assignment of Knots
We fit the models repeatedly to test series of combinations for values of the two knots k\ and ki.
For each model in the search grid, k\ x ki, we compiled information for goodness of fit (/• -
statistics or -2 log likelihood) and tests of effect (Y-tests for individual coefficients).
We found that criteria typically used for model selection based on overall goodness of fit, such
as partial F tests, were not helpful in that the differences in F statistics among the various models
were not large enough to be meaningful. Accordingly, we devised an alternative criterion for
selecting models with the optimal assignment of knots.
The criterion we settled upon was to sum the ^-values for the /-tests for the three slope
coefficients, as shown in Equation 3-41. In each case, we selected the model with the minimum
value of the summed ^-values, as the model with the most significant values for the slope terms.
criterion = pb2 + pb3 + pb4 Equation 3-41
Values of the criterion for all models fit during optimization are shown in Table 3-30 for NOx
and Table 3-31 for THC. In each table the minimum value of the criterion is indicated. The
assignments of knots for each vehicle class for NOx and THC is summarized in Table 3-32.
118
-------
Table 3-30 NO*: Optimization of knot assignments
Passenger Cars (LDV)
k2
k
l
6
7
8
9
10
11
11
0.42
0.014
0.00035
0.0012
0.0019
...
12
0.37
0.010
0.00021
0.0060
0.019
0.18
13
0.33
0.0093
0.00034
0.017
0.061
0.36
14
0.29
0.011
0.0010
0.053
0.21
0.92
15
0.26
0.020
0.0044
0.181
0.67
0.69
Light Li
ght-Duty Trucks (LLDT)
k2
k
l
6
7
8
9
10
11
11
0.0044
0.0031
0.19
0.14
0.24
...
12
0.00087
0.00037
0.030
0.017
0.56
0.51
13
0.00034
0.00010
0.0093
0.0041
0.17
0.20
14
0.000092
0.000048
0.0015
0.0033
0.023
0.51
15
0.000050
0.00012
0.00058
0.0059
0.021
0.52
Heavy Light-Duty Trucks (HLDT)
k2
k
l
5
6
7
8
9
10
11
0.74
0.54
0.54
0.25
0.84
...
12
0.58
0.38
0.22
0.13
0.52
...
13
0.48
0.29
0.12
0.078
0.33
...
14
0.39
0.21
0.079
0.038
0.15
...
15
0.33
0.16
0.044
0.019
0.073
...
16
0.26
0.10
0.026
0.0087
0.026
...
17
0.19
0.068
0.011
0.015
0.035
...
119
-------
Table 3-31 THC: Optimization of knot assignments
Passenger Cars (LP V)
k 2
k i
6
7
8
9
10
11
11
0.43
0.117
0.0018
0.048
0.035
...
12
0.49
0.087
0.0024
0.067
0.082
0.36
13
0.49
0.083
0.0010
0.036
0.034
0.070
14
0.51
0.074
0.0012
0.048
0.055
0.12
15
0.56
0.063
0.0027
0.10
0.15
0.40
Light Li
ght-Duty Trucks (LLDT)
k 2
k i
6
7
8
9
10
11
11
0.14
0.13
0.97
0.80
0.14
NA
12
0.049
0.055
0.29
0.268
0.58
0.04
13
0.037
0.033
0.22
0.1570
0.38
0.52
14
0.042
0.063
0.15
0.2513
0.479
0.32
15
0.072
0.12
0.16
0.3860
0.653
0.29
Heavy Light-Duty Trucks (HLDT)
k 2
k i
5
6
7
8
9
10
11
0.92
0.24
0.071
0.22
0.79
0.98
12
0.61
0.11
0.081
0.21
0.94
0.78
13
0.56
0.090
0.030
0.076
0.41
0.36
14
0.52
0.077
0.017
0.043
0.24
0.22
15
0.48
0.065
0.012
0.027
0.14
0.14
16
0.44
0.055
0.0082
0.018
0.083
0.091
17
0.38
0.043
0.018
0.034
0.12
0.11
Table 3-32 Assignment of knots for three-piece spline models, by emission and vehicle class
Vehicle Class
k i
k 2
NOx
Passenger Cars (LDV)
8
12
Light Light-Duty Trucks
7
14
Heavy Light-Duty Trucks
8
16
THC
Passenger Cars (LDV)
8
13
Light Light-Duty Trucks
7
13
Heavy Light-Duty Trucks
7
16
120
-------
3.6.5 Model Results
Model fitting results for NO* and THC are presented below. The left-hand portions of the tables,
"Coefficients," present coefficients, standard errors and tests of effect (i.e., /-tests) as output by
the model fitting procedure. In this case, the models were fit by ordinary least squares (OLS)
using the lm() function in R.
The "Intercept" parameter represents the intercept for the reference model year, assigned as the
most recent model year, 2010. The intercept parameter for all other model years is fit as an
incremental difference between the reference model year and the given model year.
The tables also present slope terms. The slope parameter for "Age" is the 62 coefficient in
Equation 3-40 and represents the slope below the first knot {a < k\). The second slope
parameter, which applies to the term (a - k\)d\, is the 63 coefficient and represents an
incremental difference in slope between the two knots. The third slope parameter, which applies
to the term (a - ki)ch, is the 64 coefficient and represents an incremental difference in slopes
above the second knot.
The upper right-hand portions of the tables, "Intercepts," presents intercepts (at age = 0) for each
of the three linear segments of the model, calculated as defined in Table 3-32. As shown in the
table, the calculated intercepts for the first segment {a < k\), are simply the sums of the intercept
coefficients bo and b\. Those for the second segment (k\
-------
erratic subsamples for the oldest vehicles. Alternatively, the declines may indicate that older
vehicles with higher emission rates are dropping from the population over time.
For cars, however, the slope is steeper in the middle segment (7-14 years) than in the first. In the
third segment, the slope is still positive, but very gentle. Reasons for these differences are not
clear. They may be artifacts of particular subsets of data.
Table 3-33 NO*for Passenger Cars (LDV): Intercept and slope coefficients for the selected spline model
Coefficients
Parameter
Estimate
Std Err
t-value
Pr> H|
Intercept
2.7013
0.0378
71.4980
< 2e-16
Model Year = 1990
3.8343
0.1279
29.9810
< 2e-16
Model Year = 1991
3.6080
0.1212
29.7770
< 2e-16
Model Year = 1992
3.6118
0.1125
32.1120
< 2e-16
Model Year = 1993
3.6946
0.0956
38.6320
< 2e-16
Model Year = 1994
3.3822
0.0897
37.7150
< 2e-16
Model Year = 1995
3.2332
0.0797
40.5830
< 2e-16
Model Year = 1996
2.9455
0.0759
38.8200
< 2e-16
Model Year = 1997
2.9309
0.0667
43.9730
< 2e-16
Model Year = 1998
2.6974
0.0685
39.4000
< 2e-16
Model Year = 1999
2.5712
0.0619
41.5610
< 2e-16
Model Year = 2000
2.2527
0.0588
38.3330
< 2e-16
Model Year = 2001
1.5781
0.0560
28.1860
< 2e-16
Model Year = 2002
1.6699
0.0573
29.1360
< 2e-16
Model Year = 2003
1.4395
0.0544
26.4640
< 2e-16
Model Year = 2004
1.0891
0.0369
29.5200
< 2e-16
Model Year = 2005
0.7816
0.0361
21.6630
< 2e-16
Model Year = 2006
0.5565
0.0348
15.9920
< 2e-16
Model Year = 2007
0.2810
0.0345
8.1540
0.0000
Model Year = 2008
0.2463
0.0332
7.4270
0.0000
Model Year = 2009
0.1965
0.0353
5.5650
0.0000
Model Year = 2010
0.0000
Age
0.02052
0.00536
3.82600
0.00013
Age (a -k^d^
0.03384
0.00858
3.94500
0.00008
Age (a -k2)d2
-0.04943
0.01060
-4.66500
0.00000
Intercepts
Model Year
bii + bl
bis + bl-bikl
b0 + b1-b}k1-bik2
1990
6.5356
6.2649
6.8580
1991
6.3093
6.0386
6.6317
1992
6.3131
6.0424
6.6355
1993
6.3959
6.1252
6.7183
1994
6.0835
5.8128
6.4059
1995
5.9345
5.6638
6.2569
1996
5.6469
5.3761
5.9692
1997
5.6322
5.3615
5.9546
1998
5.3987
5.1280
5.7211
1999
5.2725
5.0018
5.5949
2000
4.9540
4.6833
5.2764
2001
4.2794
4.0087
4.6018
2002
4.3712
4.1005
4.6936
2003
4.1409
3.8701
4.4633
2004
3.7904
3.5197
4.1128
2005
3.4829
3.2122
3.8053
2006
3.2578
2.9871
3.5802
2007
2.9823
2.7116
3.3047
2008
2.9477
2.6769
3.2701
2009
2.8978
2.6271
3.2202
2010
2.7013
2.4306
3.0237
Slopes
b2
bl + by
b2 + b3 + b4
0.02052
0.05436
0.00493
122
-------
Table 3-34 NO*for Light Light-Duty Trucks (LLDT): Intercept and slope coefficients for the selected spline
model
Coefficients Intercepts
Parameter
Estimate
Std Err
t-value
Pr> H|
Intercept
2.3289
0.0443
52.5590
< 2e-16
Model Year = 1990
3.9166
0.1760
22.2560
< 2e-16
Model Year = 1991
3.6634
0.1422
25.7570
< 2e-16
Model Year = 1992
3.6749
0.1450
25.3450
< 2e-16
Model Year = 1993
3.9974
0.1158
34.5110
< 2e-16
Model Year = 1994
3.6904
0.1171
31.5190
< 2e-16
Model Year = 1995
3.6680
0.0933
39.3090
< 2e-16
Model Year = 1996
3.2522
0.0832
39.0800
< 2e-16
Model Year = 1997
3.2091
0.0773
41.5150
< 2e-16
Model Year = 1998
3.1372
0.0716
43.8040
< 2e-16
Model Year = 1999
2.7283
0.0664
41.0850
< 2e-16
Model Year = 2000
2.7060
0.0638
42.4170
< 2e-16
Model Year = 2001
1.9002
0.0617
30.7730
< 2e-16
Model Year = 2002
1.5970
0.0611
26.1530
< 2e-16
Model Year = 2003
1.4452
0.0579
24.9530
< 2e-16
Model Year = 2004
1.1163
0.0373
29.9090
< 2e-16
Model Year = 2005
0.7889
0.0366
21.5490
< 2e-16
Model Year = 2006
0.6042
0.0359
16.8510
< 2e-16
Model Year = 2007
0.3873
0.0352
11.0020
< 2e-16
Model Year = 2008
0.0713
0.0335
2.1290
0.0332
Model Year = 2009
0.0230
0.0379
0.6080
0.5430
Model Year = 2010
0.0000
Age
0.08814
0.0074
11.8510
< 2e-16
Age (a -k^d^
-0.04130
0.0095
-4.3320
0.0000
Age (a -k2)d2
-0.05328
0.0128
-4.1500
0.0000
Model Year
b„ + b i
btt + bl -b3k1
btl + b1-b3k1-b4k2
1990
6.2455
6.5346
7.2804
1991
5.9923
6.2814
7.0272
1992
6.0038
6.2929
7.0387
1993
6.3263
6.6154
7.3612
1994
6.0193
6.3084
7.0542
1995
5.9969
6.2860
7.0318
1996
5.5811
5.8702
6.6160
1997
5.5379
5.8270
6.5729
1998
5.4661
5.7552
6.5010
1999
5.0572
5.3462
6.0921
2000
5.0349
5.3240
6.0698
2001
4.2291
4.5181
5.2640
2002
3.9259
4.2150
4.9608
2003
3.7741
4.0632
4.8091
2004
3.4452
3.7343
4.4801
2005
3.1178
3.4069
4.1528
2006
2.9331
3.2222
3.9680
2007
2.7162
3.0053
3.7512
2008
2.4002
2.6893
3.4351
2009
2.3519
2.6410
3.3869
2010
2.3289
2.6180
3.3638
b2
b2 + b3
b2 + b} + bi
0.08814
0.04684
-0.006434
123
-------
Table 3-35 NO*for Heavy Light-Duty Trucks (HLDT): Intercept and slope coefficients for the selected spline
model
Coefficients Intercepts
Parameter
Estimate
Std Err
t-value
Pr> H|
Intercept
2.3721
0.0695
34.1140
< 2e-16
Model Year = 1990
4.4626
0.3546
12.5850
< 2e-16
Model Year = 1991
4.1824
0.3129
13.3660
< 2e-16
Model Year = 1992
4.3503
0.2893
15.0390
< 2e-16
Model Year = 1993
4.2614
0.2315
18.4090
< 2e-16
Model Year = 1994
3.8143
0.2058
18.5360
< 2e-16
Model Year = 1995
3.9178
0.1690
23.1810
< 2e-16
Model Year = 1996
3.3442
0.1789
18.6920
< 2e-16
Model Year = 1997
3.2845
0.1468
22.3790
< 2e-16
Model Year = 1998
3.2281
0.1323
24.3950
< 2e-16
Model Year = 1999
2.8004
0.1223
22.8920
< 2e-16
Model Year = 2000
2.4822
0.1307
18.9880
< 2e-16
Model Year = 2001
2.0237
0.1143
17.7130
< 2e-16
Model Year = 2002
2.4084
0.1237
19.4690
< 2e-16
Model Year = 2003
1.7058
0.1078
15.8280
< 2e-16
Model Year = 2004
0.9004
0.0685
13.1520
< 2e-16
Model Year = 2005
0.9378
0.0681
13.7690
< 2e-16
Model Year = 2006
0.6373
0.0647
9.8480
< 2e-16
Model Year = 2007
0.7342
0.0634
11.5900
< 2e-16
Model Year = 2008
0.4001
0.0600
6.6670
0.0000
Model Year = 2009
0.0208
0.0679
0.3060
0.7597
Model Year = 2010
0.0000
Age
0.09457
0.01054
8.97300
< 2e-16
Age (a -k^d^
-0.04084
0.01525
-2.67800
0.00741
Age (a -k2)d2
-0.10121
0.03139
-3.22400
0.00127
Model Year
b„ + b i
btt + bl -b3k1
btl + b1-b3k1-b4k2
1990
6.8347
7.1614
8.7808
1991
6.5545
6.8812
8.5006
1992
6.7224
7.0491
8.6685
1993
6.6335
6.9603
8.5796
1994
6.1864
6.5131
8.1325
1995
6.2899
6.6167
8.2360
1996
5.7163
6.0431
7.6624
1997
5.6566
5.9833
7.6027
1998
5.6002
5.9270
7.5463
1999
5.1725
5.4992
7.1186
2000
4.8543
5.1810
6.8004
2001
4.3958
4.7226
6.3419
2002
4.7806
5.1073
6.7266
2003
4.0779
4.4046
6.0240
2004
3.2725
3.5992
5.2186
2005
3.3099
3.6367
5.2560
2006
3.0094
3.3361
4.9555
2007
3.1064
3.4331
5.0524
2008
2.7722
3.0990
4.7183
2009
2.3929
2.7196
4.3390
2010
2.3721
2.6988
4.3182
b2
b2 + b3
b2 + b} + bi
0.09457
0.05373
-0.047480
124
-------
10,000
(a) Passenger Cars (LDV)
10 . . . . 1 . . . . 1 . . . . 1 . . . . 1 . . . .
0 5 10 15 20 25
Vehicle Age at Test (years)
Vehicle Age at Test (years)
0 5 10 15 20 25
Vehicle Age at Test (years)
—19 90 ——19 91 —19 92 —19 93 19 94 —•—19 95 ——19 96 —19 97 —19 98 —19 99 —*-2000
__2001—•—2002—*—200320042005-*-2006—*—2007—*—2008-*—20092010
Figure 3-53 NO*: Three-piece linear spline deterioration models for three vehicle classes: (a) Passenger cars,
(b) Light Light Duty Trucks, and (c) Heavy Light-Duty Trucks. Note that emissions are expressed on common
logarithmic scale
125
-------
3.6.5.2 Total Hydrocarbons (THC)
Model fitting results for THC for the three vehicle classes are shown in Table 3-36, Table 3-37
and Table 3-38 above, respectively. Trends are also shown graphically in Figure 3-54.
The figures are depicted in logarithmic scale. However, for clarity, they are presented as
common logarithms, i.e., base 10, despite having fit the models as natural logarithms. At
logarithmic scale, the parallelism of trends by model year within the three segments is easy to
see.
However, the sequencing of trends by MY is not always monotonic. For cars, the sequencing is
generally consistent throughout. For LLDT and HLDT, there are cases where model years do not
always decrease in sequence.
Patterns of steepness in the slopes by segment are similar to the NO* models. The slope in the
youngest segment is very gentle, and that in the second steeper. Slopes in the third segment are
gently positive for cars and LLDT, and negative for HLDT.
Table 3-36 THC for Passenger Cars (LDV): Intercept and slope coefficients for the selected spline model
Coefficients
Parameter
Estimate
Std Err
t-value
Pr> H|
Intercept
1.7881
0.0460
38.8920
< 2e-16
Model Year = 1990
3.4132
0.1534
22.2550
< 2e-16
Model Year = 1991
3.2188
0.1449
22.2200
< 2e-16
Model Year = 1992
2.8936
0.1340
21.5960
< 2e-16
Model Year = 1993
2.8477
0.1137
25.0350
< 2e-16
Model Year = 1994
2.6462
0.1064
24.8730
< 2e-16
Model Year = 1995
2.5205
0.0945
26.6780
< 2e-16
Model Year = 1996
2.0756
0.0902
23.0150
< 2e-16
Model Year = 1997
1.9756
0.0798
24.7520
< 2e-16
Model Year = 1998
1.7102
0.0827
20.6690
< 2e-16
Model Year = 1999
1.4418
0.0745
19.3530
< 2e-16
Model Year = 2000
1.1036
0.0707
15.6010
< 2e-16
Model Year = 2001
0.6765
0.0679
9.9640
< 2e-16
Model Year = 2002
0.4631
0.0690
6.7110
0.0000
Model Year = 2003
0.2334
0.0661
3.5310
0.0004
Model Year = 2004
0.2075
0.0448
4.6280
0.0000
Model Year = 2005
0.1400
0.0439
3.1900
0.0014
Model Year = 2006
0.1392
0.0423
3.2890
0.0010
Model Year = 2007
0.0786
0.0419
1.8750
0.0608
Model Year = 2008
0.0676
0.0404
1.6750
0.0939
Model Year = 2009
0.0181
0.0430
0.4210
0.6738
Model Year = 2010
0.0000
Age
0.0237
0.0065
3.6450
0.0003
Age (a -kl)dl
0.0348
0.0100
3.4810
0.0005
Age (a - k2)d2
-0.0491
0.0133
-3.6830
0.0002
Intercepts
Model Year
bii + bl
b{l + b1 - b3k1
bll + b1-b3k1-bik 2
1990
5.2013
4.9232
5.5610
1991
5.0069
4.7288
5.3666
1992
4.6817
4.4036
5.0415
1993
4.6358
4.3576
4.9955
1994
4.4343
4.1561
4.7940
1995
4.3086
4.0305
4.6683
1996
3.8637
3.5856
4.2234
1997
3.7637
3.4856
4.1234
1998
3.4983
3.2202
3.8580
1999
3.2299
2.9518
3.5896
2000
2.8917
2.6135
3.2514
2001
2.4646
2.1865
2.8243
2002
2.2512
1.9730
2.6109
2003
2.0215
1.7434
2.3813
2004
1.9956
1.7175
2.3553
2005
1.9281
1.6499
2.2878
2006
1.9273
1.6491
2.2870
2007
1.8667
1.5886
2.2264
2008
1.8557
1.5776
2.2155
2009
1.8062
1.5281
2.1659
2010
1.7881
1.5100
2.1478
Slopes
b2
b i + b3
b2 + b3 + b4
0.0237
0.05844
0.00938
126
-------
Table 3-37 THC for Light Light-Duty Trucks (LLDT): Intercept and slope coefficients for the selected spline
model
Coefficients Intercepts
Parameter
Estimate
Std Err
t-value
Pr> H|
Intercept
1.4111
0.0541
26.1010
< 2e-16
Model Year = 1990
3.8269
0.2111
18.1280
< 2e-16
Model Year = 1991
3.6516
0.1703
21.4410
< 2e-16
Model Year = 1992
3.2634
0.1739
18.7610
< 2e-16
Model Year = 1993
3.2756
0.1399
23.4180
< 2e-16
Model Year = 1994
3.2273
0.1413
22.8390
< 2e-16
Model Year = 1995
3.2101
0.1132
28.3630
< 2e-16
Model Year = 1996
2.6421
0.1001
26.3920
< 2e-16
Model Year = 1997
2.2638
0.0929
24.3570
< 2e-16
Model Year = 1998
2.1779
0.0862
25.2690
< 2e-16
Model Year = 1999
1.7882
0.0797
22.4270
< 2e-16
Model Year = 2000
1.6094
0.0769
20.9400
< 2e-16
Model Year = 2001
0.8705
0.0747
11.6480
< 2e-16
Model Year = 2002
0.9834
0.0736
13.3580
< 2e-16
Model Year = 2003
0.5677
0.0705
8.0530
0.0000
Model Year = 2004
0.5374
0.0454
11.8340
< 2e-16
Model Year = 2005
0.4522
0.0445
10.1620
< 2e-16
Model Year = 2006
0.2491
0.0436
5.7120
0.0000
Model Year = 2007
0.0976
0.0428
2.2790
0.0226
Model Year = 2008
0.0599
0.0408
1.4670
0.1425
Model Year = 2009
-0.1038
0.0462
-2.2460
0.0247
Model Year = 2010
0.0000
Age
0.0716
0.0090
7.9090
0.0000
Age (a -k^d^
-0.0285
0.0117
-2.4310
0.0151
Age (a -k2)d2
-0.0330
0.0140
-2.3600
0.0183
Model Year
b„ + b i
btt + bl -b3k1
btl + b1-b3k1-b4k2
1990
5.2380
5.4373
5.8668
1991
5.0627
5.2621
5.6915
1992
4.6746
4.8739
5.3034
1993
4.6867
4.8860
5.3155
1994
4.6384
4.8378
5.2672
1995
4.6212
4.8206
5.2500
1996
4.0533
4.2526
4.6821
1997
3.6749
3.8742
4.3037
1998
3.5890
3.7884
4.2178
1999
3.1993
3.3987
3.8281
2000
3.0206
3.2199
3.6494
2001
2.2816
2.4809
2.9104
2002
2.3945
2.5938
3.0233
2003
1.9788
2.1782
2.6077
2004
1.9485
2.1479
2.5774
2005
1.8633
2.0627
2.4922
2006
1.6602
1.8595
2.2890
2007
1.5087
1.7081
2.1376
2008
1.4710
1.6703
2.0998
2009
1.3073
1.5066
1.9361
2010
1.4111
1.6105
2.0399
b2
b2 + b3
b2 + b} + bi
0.0716
0.04309
0.01005
127
-------
Table 3-38 THC for Heavy Light-Duty Trucks (HLDT): Intercept and slope coefficients for the selected spline
model
Coefficients Intercepts
Parameter
Estimate
Std Err
t-value
Pr> H|
Intercept
1.6683
0.0781
21.3670
< 2e-16
Model Year = 1990
4.6073
0.3464
13.3020
< 2e-16
Model Year = 1991
4.1024
0.3056
13.4230
< 2e-16
Model Year = 1992
4.4423
0.2826
15.7220
< 2e-16
Model Year = 1993
4.7289
0.2260
20.9230
< 2e-16
Model Year = 1994
3.9220
0.2009
19.5250
< 2e-16
Model Year = 1995
3.9106
0.1650
23.6960
< 2e-16
Model Year = 1996
2.7202
0.1749
15.5570
< 2e-16
Model Year = 1997
2.3602
0.1436
16.4380
< 2e-16
Model Year = 1998
2.0586
0.1296
15.8820
< 2e-16
Model Year = 1999
1.8609
0.1202
15.4790
< 2e-16
Model Year = 2000
1.4066
0.1281
10.9800
< 2e-16
Model Year = 2001
1.1281
0.1122
10.0570
< 2e-16
Model Year = 2002
1.1582
0.1218
9.5080
< 2e-16
Model Year = 2003
1.1536
0.1062
10.8620
< 2e-16
Model Year = 2004
0.6204
0.0673
9.2230
< 2e-16
Model Year = 2005
0.8415
0.0672
12.5170
< 2e-16
Model Year = 2006
0.6840
0.0646
10.5850
< 2e-16
Model Year = 2007
0.7443
0.0634
11.7410
< 2e-16
Model Year = 2008
0.4792
0.0596
8.0430
0.0000
Model Year = 2009
0.2349
0.0674
3.4870
0.0005
Model Year = 2010
0.0000
Age
0.0849
0.0134
6.3620
0.0000
Age (a -k^d^
-0.0569
0.0170
-3.3500
0.0008
Age (a -k2)d2
-0.0807
0.0301
-2.6780
0.0074
Model Year
b„ + b i
btt + bl -b3k1
btl + b1-b3k1-b4k2
1990
6.2755
6.6740
7.9650
1991
5.7706
6.1691
7.4601
1992
6.1106
6.5090
7.8001
1993
6.3971
6.7956
8.0866
1994
5.5902
5.9887
7.2797
1995
5.5789
5.9773
7.2684
1996
4.3884
4.7868
6.0779
1997
4.0285
4.4269
5.7179
1998
3.7269
4.1253
5.4163
1999
3.5292
3.9276
5.2187
2000
3.0749
3.4733
4.7644
2001
2.7964
3.1948
4.4859
2002
2.8265
3.2249
4.5160
2003
2.8219
3.2203
4.5113
2004
2.2887
2.6871
3.9781
2005
2.5097
2.9081
4.1992
2006
2.3522
2.7507
4.0417
2007
2.4125
2.8110
4.1020
2008
2.1475
2.5459
3.8369
2009
1.9032
2.3016
3.5927
2010
1.6683
2.0667
3.3577
b2
b2 + b3
b2 + b} + bi
0.0849
0.02800
-0.05269
128
-------
10,000
1,000
10 15
Vehicle Age at Test (years)
~ 1,000 -
10 15
Vehicle Age at Test (years)
10 15
Vehicle Age at Test (years)
-1990-
-2001-
-1991-
-2002-
-1992-
-2003-
-1993-
-2004-
-1994-
-2005-
-1995-
-2006-
-1996-
-2007-
-1997-
-2008-
-1998-
-2009 -
-1999-
-2010
Figure 3-54 THC: Three-piece linear spline deterioration models for three vehicle classes: (a) Passenger cars,
(b) Light Light Duty Trucks, and (c) Heavy Light-Duty Trucks. Note that emissions are expressed on common
logarithmic scale
129
-------
3.6.6 Reverse Transformation
Despite the fact that all parameters in all models are highly significant, the main purpose for
these analyses is not hypothesis testing, but developing emission rates. It is therefore necessary
to reverse transform the logarithmic model results for purposes of prediction.
As the response variable for the models is In j', we exponentiate the results to estimate emissions
in original units (mg/mi).
y = e]ny Equation 3-42
However, under the assumption that the emissions are lognormally distributed, this initial step
returns not the mean emissions level, but rather the "geometric mean" emissions level, which we
can effectively treat as the "median" level, denoted as_yg. This level is of general interest in that
it indicates the emissions level of a "typical" vehicle.
However, for estimation of an emissions inventory, the parameter to be estimated is not the
"geometric mean" but rather the "arithmetic mean," as the arithmetic mean relates directly to
total emissions, e.g., kg, Mg. To estimate the arithmetic mean, which we will denote asya, we
add a second term including the "logarithmic variance" (s2):
ya = einyeo.ss2 _ y^eo.ss2 Equation 3-43
The implication is that underestimating s2 would lead to underestimation of the arithmetic mean
ya. In the models, we estimate the logarithmic variance as the residual error variance. The OLS
models estimate a uniform error variance for residuals throughout the parameter space. The
logarithmic variance is of interpretive interest as it provides an index of the degree of right skew
in the lognormal distribution. In fact, the second term in the equation gives the ratio of the
arithmetic to the geometric mean (ya/yg).
However, as the sample sizes by model year and age are not uniform throughout the dataset,
neither is the variance. The variance is related to sample size, as in random sampling, the
probability of pulling in the extremes of the distribution is proportional to sample size. In
addition, we observed as noted above that the sampling effort was higher for MY since 2004 (see
Table 3-25, Table 3-26 and Table 3-27 above).
To investigate patterns in s2 with model year, we fit a second set of models. Rather than classic
OLS regressions, we used mixed-factor models to take advantage of the capability of these
procedures to estimate heterogeneous error variances by subgroups in the data. We used the
lme() function in the R nlme library.
The resulting variance estimates are in Table 3-39 for NO* and Table 3-40 for THC. The same
results are presented graphically in Figure 3-55 for NO* and Figure 3-56 for THC. While
variances vary from model year to model year, it is clear that they are highest in the model years
with the largest sample sizes, e.g., n > 800.
The task then was to decide how to select values of s2 to use for the reverse transform. We
proceeded on the assumption that the largest samples come closest to capturing the full range of
variability in the population distributions. Conversely, we assume that lower variances in the
smaller samples fail to capture the expected variability.
Another important question concerned whether the error variance might be expected to decline
as vehicles age. In this dataset, the data for older "Tier-1" model years, e.g., prior to 2000, were
130
-------
collected when the vehicles were older than 10 years. As the models would be used to hindcast
emissions for these vehicles when less than five years of age, a key question is whether their
variances when young would be similar to those for the young Tier 2 vehicles (e.g., MY 2004
and later) directly observed in this dataset.
We answered this question in the affirmative, based on remote-sensing data collected by the
University of Denver. There results showed that variances for Tier 1 vehicles measured while
young were as large or larger than any measured for young Tier 2 vehicles.
For the reverse transformation, we assigned a uniform value of s1 for use with each model, which
we applied to all model years. These values were calculated as averages of a subset of model
years for which samples were reasonably large and during which the variances were in a
relatively uniform range. The subsets of model years used for each vehicle class indicated by
gray shading, with the values obtained shown at the bottom of the tables.
For NOx, the values of s2 for LLDT are lower than those for cars, while those for HLDT are
higher. For THC, values of s1 for both truck classes are lower than that for cars and are nearly
equal.
Seeing no obvious reasons based on engine or emissions control technology why the variances
for cars would be highest, we suggest it may reflect the fact that the cars have the largest
samples. Offhand, we would assume that variances would be similar for different vehicle classes,
if all populations were adequately characterized. Nonetheless, for each vehicle class, we applied
the variance estimates obtained from their respective datasets.
131
-------
Table 3-39 NO*: Logarithmic variances by model year for three vehicle classes (NOTE: the gray cells include
those in the 10-yr average below, used for reverse transformation)
Model Year
LDV
LLDT
HLDT
1990
0.8225
0.7676
0.1901
1991
0.9789
0.9840
0.4173
1992
0.8384
0.9304
0.3448
1993
0.6629
0.5316
0.3489
1994
0.9073
0.7264
0.2559
1995
0.8543
0.8819
0.4171
1996
0.7815
1.0679
0.6668
1997
0.7021
1.0168
0.7426
1998
1.1197
0.9893
0.7794
1999
1.1566
1.0418
1.1740
2000
1.4283
1.1173
1.4467
2001
1.5954
1.5210
2.0158
2002
1.5699
2.0570
2.2738
2003
1.7353
1.3071
1.9447
2004
1.6167
1.6073
2.0701
2005
1.6636
1.5316
2.4124
2006
1.8533
1.3472
2.3662
2007
1.8132
1.2774
1.5647
2008
1.4856
1.5210
1.9682
2009
1.4589
1.3820
1.9847
2010
1.3648
1.3353
1.7527
10-yr Average
1.6157
1.4887
2.0353
132
-------
Table 3-40 THC: Logarithmic variances by model year for three vehicle classes (NOTE: the gray cells include
those in the averages below, used for reverse transformation)
Model Year
LDV
LLDT
HLDT
1990
1.7266
1.8780
0.3119
1991
1.5770
1.6869
0.7567
1992
1.4231
1.9454
0.6442
1993
1.7927
1.5504
0.6614
1994
2.3794
1.9607
0.9015
1995
1.8040
1.7387
1.1649
1996
1.9471
1.4782
1.3296
1997
1.7056
1.5644
2.1978
1998
2.4316
1.6758
1.9049
1999
2.3307
1.5439
2.0212
2000
2.5899
1.8922
2.2248
2001
2.6940
2.3209
2.7090
2002
2.7358
2.1830
1.9808
2003
2.7920
2.1026
2.0720
2004
2.6518
2.1217
1.7764
2005
2.2630
1.9287
1.7155
2006
2.2861
1.7526
1.6746
2007
2.2471
1.9959
1.3231
2008
2.3235
2.0149
2.3402
2009
2.1573
2.0544
1.9051
2010
2.1265
2.0754
2.0691
Average
2.4330
2.0550
1.9939
133
-------
3.00
>
o
i_
i_
LU
U
'E
_c
•C 1.00 --
0.00
-
LDV
-
LLDT
-*-HLDT
A A
; l\ k i
! l\w \
! 1 iX
V
1 1 1 1 1
¦k '
A-*~^ /
/
¦ —,—,—,—,—i . . .—.—i—.—. . . i
—¦ ¦ ¦ .—
—¦—¦—
—¦—. . ¦
1985
1990
1995
2000
Model Year
2005
2010
2015
Figure 3-55 NO* Logarithmic variance (s2) by model year for three vehicle classes. Solid horizontal lines
represent values selected for reverse-transformation
3.00
2.50 --
CD
ro
•c 2.00
ro
>
O
li] 1.50
u
'E
_c
•c 1.00
ro
ao
o
0.50 --
0.00
1985
.J, r
T\
i \
—N
u
* AjWH* * \ \ / ; *'
vv' >:/ \ /va / /
"* i
/
\»
A
d
LDV
------ LLDT
#
—A—HLDT
1990
1995 2000 2005
Model Year
2010
2015
Figure 3-56 THC: Logarithmic variance (s2) by model year for three vehicle classes. Solid horizontal lines
represent values selected for reverse-transformation
Deterioration trends as predicted by the models following the reverse transformation are shown
in Figure 3-57 for NO* and Figure 3-58 for THC.
134
-------
2,500
2,000
1,000 ¦¦
(a) Passenger Cars (LDV)
Vehicle Age at Test (years)
2,500 ¦¦
^ 2,000 +
oo
O 1,500 --
1,000 ¦¦
(b) Light Light-Duty Trucks (LLDT)
Vehicle Age at Test (years)
Vehicle Age at Test (years)
-1990—.—1991—*—1992—•—1993—.—1994—•—1995—.—1996—.— 1997—.—1998—.—1999—.—2000
-2001—*—2002—.—2003—.—2004—#—2005—•—2006—.—2007—.—2008—.—2009—•—2010
Figure 3-57 NO*: Trends in emissions vs. age as predicted by reverse-transformed three-piece ln-linear spline
models
135
-------
1,200
1,000 ¦¦
800 +
Q0
£
y eoo 4-
(a) Passenger Cars (LDV)
10 15
Vehicle Age at Test (years)
1,200
1,000
1
00
£ 800
U
JZ
o 600
(b) Light Light-Duty Trucks (LLDT)
10 15
Vehicle Age at Test (years)
< 2,500
QO
£
^ 2,000 --
I-
o
Q! 1,500
(c) Heavy Light-Duty Trucks (HLDT)
10 15
Vehicle Age at Test (years)
-1990-
-2001-
-1991-
-2002-
-1992-
-2003-
-1993-
-2004-
-1994-
-2005-
-1995-
-2006-
-1996-
-2007-
-1997-
-2008-
-1998-
-2009-
-1999-
-2010
Figure 3-58 THC: Trends in emissions vs. age as predicted by reverse-transformed three-piece ln-linear
spline models
136
-------
3.6.7 "Young Vehicle" Adjustments
The adjustments for "young vehicles" were developed by using the spline models to estimate
average IM240 levels at 2 years of age for all model years. The result is a trend in emissions
with model year at age = 2. Two years of age was selected because it represents the midpoint of
the 0-3 year ageGroup, which is actually 4 years in length, i.e., vehicles are three years old until
their "fourth birthday." This rate at age 2 is later used as the basis for applying deterioration.
For comparison, a corresponding trend to represent the MOVES2014 rates for the 0-3 year
ageGroup was constructed by simulating the IM240 cycle using the MOVES2014b rates for the
hot-running emissions process. This step was achieved by calculating sums of rates weighted by
an operating mode distribution for the IM240 cycle. The total (g) is the sum of time-in-mode (hr)
times emission rate (g/hr).
3.6.7.1 Adjustments for NO.v
3.6.7.1.1 Cars
For NOi, estimates from the spline models based on the Denver IM240s are consistently higher
than the simulated MOVES2014 IM240s, as shown in Figure 3-59.
Also, the Denver IM240 results show a steady decline in emissions from 1994 through 2000,
which years include the phase-in and duration of the Tier 1 emissions standards. This pattern
contrasts with that in the MOVES rates, which assume stable emissions during MY 1996-2000,
while the Tier 1 standards were in effect. In other words, the MOVES2014 rates assumed that
emission rates remain stable if the emissions standards are unchanging. However, the evidence
from the Denver data suggest otherwise—that emissions may decline without corresponding
declines in standards. Design features contributing to the declines could include the introduction
of oxygen sensors and on-board diagnostic systems (OBD).
Nonetheless, the chief salient feature is that the Denver IM240 levels are consistently higher than
the simulated MOVES IM240s. This pattern holds over the entire model year range, even during
and after the phase-in of Tier 2 standards, as clearly shown in Figure 3-59.
137
-------
Model Year
400
350
— 300
°j> 250
O 200
Z
° 150
rsi
- 100
50
0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Model Year
Figure 3-59 NO*for Cars: Trends by model year IM240 emissions simulated from MOVES2014 rates and
estimated from Denver IM240 at age = 2 years: (a) Overview for MY 1990-2010; (b) CLOSEUP on National
LEV (2001-2003) and Tier 2 (2004-2010) standards
3.6.7.1.2 Trucks
The picture for trucks is more complicated. As discussed above, we modeled the Denver data for
two truck classes, whereas MOVES treats all trucks as a single class.
Accordingly, we needed to resolve the spline model results into a single truck class. We did this
by weighting them. We used fractions derived in development of rates for MOVES2010 and
MOVES2014. Originally, the fractions were applied to individual truck classes LDT1-LDT4. In
the current analysis, fractions for LLDT were calculated by summing fractions for LDT1 and
LDT2, and fractions for HLDT by summing fractions for LDT3 and LDT4 (Table 3-41).
138
-------
The summed fractions were used to construct a combined single trend for all trucks, designated
at "LDT," as shown in Figure 3-60.
Table 3-41 Truck Class fractions in the light-duty fleet, by model year
Mo del Year
LDT1
LDT2
LDT3
LDT4
LLDT
HLDT
1990
0.100
0.595
0.185
0.120
0.695
0.305
1991
0.100
0.595
0.185
0.120
0.695
0.305
1992
0.100
0.595
0.185
0.120
0.695
0.305
1993
0.100
0.595
0.185
0.120
0.695
0.305
1994
0.100
0.595
0.185
0.120
0.695
0.305
1995
0.100
0.595
0.185
0.120
0.695
0.305
1996
0.100
0.595
0.185
0.120
0.695
0.305
1997
0.100
0.595
0.185
0.120
0.695
0.305
1998
0.100
0.595
0.185
0.120
0.695
0.305
1999
0.100
0.595
0.185
0.120
0.695
0.305
2000
0.100
0.595
0.185
0.120
0.695
0.305
2001
0.098
0.598
0.187
0.117
0.696
0.304
2002
0.085
0.634
0.172
0.109
0.719
0.281
2003
0.093
0.585
0.183
0.140
0.677
0.323
2004
0.085
0.558
0.316
0.040
0.644
0.356
2005
0.078
0.748
0.147
0.027
0.826
0.174
2006
0.097
0.610
0.247
0.046
0.707
0.293
2007
0.089
0.554
0.340
0.017
0.644
0.356
2008
0.085
0.550
0.350
0.015
0.635
0.365
2009
0.085
0.550
0.350
0.015
0.635
0.365
2010
0.085
0.550
0.350
0.015
0.635
0.365
139
-------
3500.00
3000.00
~ 2500.00 -¦
1500.00 -¦
- 1000.00 -¦
500.00 ¦¦
0.00
: (a) Model Years 1990-2010
-~-LLDT
-¦—HLDT
-A-lDT
¦—i
1
1 1 1
1
1
—-r—'
1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
Model Year
500.00
400.00
1
^ 300.00
x
O
-Z.
o 200.00
100.00 -¦
0.00
(b) Model Years 2000-2010
-~-LLDT
-¦-HLDT
-A-LDT
1
2000 2001 2002 2003 2004 2005 2006
Model Year
2007
2008
2009
2010
Figure 3-60 NO* Estimated IM240 emissions vs. model year at age 2, for individual and weighted truck
classes: (a) full model-year range (1990-2010); (b) Closeup on model-years (2000-2010)
Trends in predicted and simulated IM240 emissions for combined trucks (LLDT + HLDT =
LDT) are shown in Figure 3-61. Like the trends for cars, the predicted Denver results show a
steady decline in truck emissions from 1993 to 2001. During this period, however, the
differences between the Denver and MOVES2014 values are not as prominent as those for cars.
In addition, the Denver results show a gradual decline from 2001-2004, whereas the
MOVES2014 values remain stable. In this interval, the MOVES rates reflect the assumption that
the heavier trucks (HLDT) remain at elevated Tier-1 levels, while the lighter trucks (LLDT) have
come under reduced National LEV standards. During the adoption of Tier 2 standards (2004-
2010) the Denver trend is consistently higher than the MOVES trend, as it is for cars, although
differences are relatively small.
140
-------
Model Year
Model Year
Figure 3-61 NO*for Trucks (LDT): Trends by model year IM240 emissions simulated from MOVES2014 rates
and estimated from Denver IM240 at age = 2 years: (a) Overview for MY 1990-2010; (b) CLOSEUP on
National LEV (2001-2003) and Tier 2 (2004-2010) standards
3.6.7.2 Calculating NO.v Adjustments
Based on these trends, as shown in Figure 3-60 for cars and Figure 3-61 for trucks, the "young-
vehicle" adjustments for each model year were calculated as the ratio
predicted Denver IM2 40 Equation 3-44
Ay°un0 ~ simulated MOVES IM240
The adjustments for cars and trucks are shown in Figure 3-62. The adjustments for cars are
generally larger ( > 2.0) for model years prior to 1997 and between about 1.5 and 2.0 for model
years after 2005. The adjustments for cars are consistently > 1.0, except for model year 2000,
where the value is very close to 1.0.
141
-------
The adjustments for trucks are generally smaller than those for cars, always < 2.0 for model
years prior to 2000, and <1.5 for all model years after 2006. In the intervening years, 2001-2005,
the adjustments are < 1.0, ranging as low as 0.50.
c
01
£
"O
<
3.00
2.50
2.00
1.50
1.00
0.50
0.00
» • • 9 \
(a) Cars
1985 1990 1995
2000 2005 2010 2015 2020
3.00
2.50
4—1
(—
2.00
CD
£
4-»
1/1
1.50
3
<
1.00
0.50
0.00
_ V
A
J |
n i r
\
/\s
V
(
(b) Trucks
1985 1990 1995
2000 2005
Model Year
2010
2015
2020
Figure 3-62 NO*: 'Young-vehicle" adjustments for (a) Cars and (b) Trucks
3.6.7.3 Adjustments for THC
3.6.7.3.1 Cars at Age 2
As with NOi, estimates from the spline models based on the Denver IM240s are consistently
higher than the simulated MOVES IM240s, as shown in Figure 3-63 below.
Like the trends for NO*, the Denver IM240 results show a steady decline in emissions from 1994
through 2000 (Figure 3-63 (a)), which years include the phase-in and duration of the Tier 1
emissions standards. The MOVES rates for THC also assume stable emissions during MY 1996-
2000. During the second decade (2000-2010) the striking feature is that the predicted Denver
142
-------
IM240s exceed the MOVES rates by a larger margin than for NO* (~5-fold rather than ~2-fold)
as shown in Figure 3-63 (b).
o
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Model Year
(b) Model Years 2000-2010 ---MOVES2014: Simulated IM240
-e-MOVES3: Deterioration Model
Figure 3-63 THC for Cars: Trends in IM240 emissions simulated from MOVES2014 rates and estimated from
Denver IM240 vs. model year at age = 2 years: (a) full model year range (1990-2010); (b) CLOSEUP on
model year range (2000-2010)
3.6.7.3.2 Trucks at Age 2
A single trend for trucks was constructed by taking a weighted average of the trends for LLDT
and HLDT, using the weights shown in Table 3-41 above. The summed fractions were used to
construct a combined single trend for all trucks, designated at "LDT," as shown in Figure 3-64.
Hydrocarbon emissions for both classes show a marked drop at the outset of the Tier 1 standards
(1995-1996) (Figure 3-63 (a)). During this period, the emissions for HLDT are noticeably higher
than those for LLDT. In addition, and contrast to NO*, the trend for HLDT remains higher than
that for LLDT throughout the Tier-2 phase-in (2004-2010) (Figure 3-64 (b)).
143
-------
2,500
2,000
j| 1,500
U
O 1,000
(N
1 1 1 1 1
(a) Model Years 1990-201C
)
—•—HLDT
-¦-11 DT
1\
-LDT
t\ i
\ /
\ /
\/
\
] 1—
1—
1—
i 1—
—i—
!?—5—i
a—1
1990
1992 1994 1996 1998 2000 2002
Model Year
2004 2006 2008 2010
90
~ 70
£
o
Csl
0
(b) Model Years 2000-2010
—•—HLDT
-¦-LLDT
-A-LDT
V\
_
2000 2001 2002 2003 2004 2005 2006
Model Year
2007
2008
2009
2010
Figure 3-64 THC: Estimated IM240 emissions vs. model year at age 2, for individual and weighted truck
classes: (a) full model-year range (1990-2010); (b) Closeup on model-years (2000-2010)
Trends in predicted and simulated IM240 emissions for combined trucks (LLDT + HLDT =
LDT) are shown in Figure 3-65. Like the trends for cars, the predicted Denver results show a
steep decline in truck emissions from 1993 to 2001. During this period, however, the differences
between the Denver and MOVES values are greater than a factor of 2. In 2004, at the outset of
the Tier 2 phase-in, the MOVES2014 rates drop by a factor of 3, whereas the estimates based on
Denver data continue a gradual but steady decline. During this period, the Denver estimates
remain -3-5 times higher than the MOVES2014 rates.
144
-------
900
800
700
600
— 500
U
fE 400
I 300
- 200
I 1
0—e—e—®
(a) Model Years 1990-2010
-•-MOVES 2014
u uer
ver iivizf
u
—
—\
\
t
\
J
q—,—
\
—v
—i—
N
1
—i
(
(
(
1990 1992 1994 1996 1998 2000 2002
Model Year
2004
2006
2008
2010
OD
E,
u
o
CM
100
90
60
50
(b) Model Years 2000-201C
)
-•-MOVES2014
-©—Denver IM240
\
,
,
r
2000
2002
2004 2006
Model Year
2008
2010
Figure 3-65 THC for Trucks (LDT): Trends in estimated and simulated IM240 emissions vs. model year at age
= 2 years: (a) Overview for MY 1990-2010; (b) CLOSEUP on National LEV (2001-2003) and Tier 2 (2004-
2010) standards
3.6.7.4 Calculating THC Adjustments
Based on these trends, as shown in for cars and for trucks, the "young-vehicle" adjustments for
each model year were calculated as for NO*, using Equation 3-44.
The adjustments are cars are shown in (a). For model years prior to 2000, the adjustments are
always > 1.0, and below a maximum of 4.5. In 2000, the adjustment peaks at 10.0 as the two
trends diverge at the outset of the National LEV standards. In successive model years, the
adjustment declines, stabilizing at -6.0 in 2004 and thereafter.
The adjustments for trucks are smaller than those for cars, always < 4.0 for model years prior to
2000, and < 5.0 for all model years after 2005. In the intervening years, 2001-2005, the
adjustments are smaller, ranging as low as 1.0 in 2003.
145
-------
11.00
10.00
9.00
8.00
c 7.00
CJ
E 6.00
"I 5.00
^ 4.00
3.00
2.00
1.00
0.00
1985 1990 1995 2000 2005 2010 2015 2020
Model Year
11.00
10.00
9.00
8.00
c 7.00
CJ
E 6.00
"I 5.00
^ 4.00
3.00
2.00
1.00
0.00
1985 1990 1995 2000 2005 2010 2015 2020
Model Year
Figure 3-66 THC: 'Young-vehicle" adjustments for (a) Cars and (b) Trucks
3.6.8 Deterioration Adjustments
3.6.8.1 Running Process for NOx
We also used the spline models to project emissions trends vs. age for model years 1990-and-
later.
For trucks it is necessary to construct a single trend, as MOVES treats light-duty trucks as a
single class. We achieved this goal by calculating a weighted average of the trends for LLDT
and HLDT. For this purpose, we used the trends for MY 2000, as the fractions for this model
year (0.695, 0.305) are close to the averages for the entire model year range 1990-2010 (0.689,
0.311). For cars, LLDT, HLDT and LDT, deterioration trends for MY 2000 are shown in Figure
3-67.
i i /¦ r\i i\
(a) Cars (LDV)
\
V
i
v A
V/1
'\
1
q 1 1
(b) Trucks (LDT)
3 , ,
3
j
1
X
q
r~
r
V
/
j !
M
x
/
s
- - 1 1 1 1
1 1 1 1
1 1 1 1 1 1 1 1
i i i i
i i i i
i i i i
146
-------
1,200
1,100
1,000
900
800
CtD
700
X
O
600
z:
o
500
'vf
r\i
400
300
200
100
0
10 15
Age (years)
Figure 3-67 NO* Trends in emissions vs. age for trucks in model year 2000, by class
We also assembled mean simulated MOVES IM240s by Age Group for model years 1990-and-
later, which we plotted against the midpoints of the ageGroups. Alongside the MOVES rates, we
plot the Denver results against ages coinciding with or close to the midpoints of the ageGroups,
i.e., 2, 5, 7, 9, 12.5, 17.5 and 23 years, respectively.
In Figure 3-68, we've plotted examples for a "Tier 1" model year (1998) and a "Tier 2" model
year (2008) for both cars and trucks. The differences at age = 2 reflect the effects of the "young
vehicle" adjustments as described above. For the remaining ages, the Denver trends reflect the
age slopes for the spline models and those for the MOVES2014 values reflect the deterioration
assumptions in the MOVES2014 rates. For ages after 5 years (ageGroup 4-5 years), the patterns
vary with the Denver predictions exceeding the MOVES rates in some cases and the reverse in
others.
147
-------
MOVES2014: Simulated IM240
M0VES3: Deterioration Model
(a) Cars, MY 1998
(c) Cars, MY 2008
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
1,600
1,400
Age (years)
Age (years)
Figure 3-68 NO*: Predicted trends in IM240 emissions vs age for cars and trucks in two model years
To express the deterioration shown in in proportional or relative terms, we can normalize the
trends in Figure 3-68 to the first age group (age = 2). The values at age = 2 are converted to 1.0
and those for the remaining age groups to ratios relative to the first group, as shown in Figure
3-69. Note that the relative trends in the Denver-based values are identical in both model years
for cars and trucks as this outcome is an implication of the premises of the spline models.
The trends show clearly that proportional deterioration in the MOVES2014 rates is substantially
higher than in the Denver IM240 dataset. This point is conspicuous in the case of cars. The
MOVES rates increase by factors of 2.5 and 3.0 in 1998 and 2008, respectively. In contrast, the
Denver-based values increase by only a factor of 1.5. For trucks the result is somewhat less
marked, with MOVES2014 rates increasing by about a factor of 3.0 and the Denver-based values
by about 2.25.
5 10 15 20 25
Age (years)
(d) Trucks, MY 2008
(b) Trucks, MY 1998
—MOVES2014: Simulated IM240
-©-MOVES3: Deterioration Model
5 10 15 20 25
Model Year
MOVES2014: Simulated IM240
-©-MOVES3: Deterioration Model
148
-------
4.0 t-
3.5 ±
.2 3-° T
* 2.5 t
c
o
'ro 2-° q-
o
ai 1-5 T
<=> 1.0 +-
0.5 i-
0.0 +-
0
4.0 q-
3.5 +
.2 3.0 §-
en
c 2.5 4-
o
ro 2.0 i-
g
aj 1.5 4-
* 3
Q 1.0 4-
0.5 4-
0.0 +-
0
Figure 3-69 NO*: Deterioration ratios for cars and trucks in two model years
3.6.8.2 Running Process for THC
For trucks, we constructed a single trend for trucks (LDT) as a weighted average of the trends for
LLDT and HLDT, using the MY2000 LLDT & HLDT weighting factors, as described for NO*,
above. The individual and combined trends are shown in Figure 3-70.
160
140
120
I
"So 100
^ 80
I-
o
a 60
40
20
0
0 5 10 15 20 25
Age (years)
Figure 3-70 THC: Trends in emissions vs. age for trucks in model year 2000, by class
(a) Cars, MY 1998
—MOVES2014: Simulated IM240
MOVES3: Deterioration Model
(b) Trucks, MY 1998
Age (years)
—•—MOVES2014: Simulated IM240
-e-MOVES3: Deterioration Model
Age (years)
(c) Cars, MY 2008
—MOVES2014: Simulated IM240
MOVES3: Deterioration Model
Age (years)
(d) Trucks, MY 2008
—•—MOVES2014: Simulated IM240
-©-MOVES3: Deterioration Model
Age (years)
^4
—•—HLDT
-¦—LLDT
—A—LDT
149
-------
As with NOi, we plotted mean simulated MOVES IM240s and Denver-based values against ages
coinciding with or close to the midpoints of the ageGroups, i.e., 2, 5, 7, 9, 12.5, 17.5 and 23
years, respectively.
In Figure 3-71, we show results for MY 1998 and 2008 for both cars and trucks. The differences
at age = 2 reflect the effects of the "young vehicle" adjustments as described above. In 1998
(Figure a and b), the deterioration trends in the MOVES2014 rates are aggressive enough that the
MOVES2014 rates are higher than the Denver-based values after 9 years of age (the 8-9 year
ageGroup). In 2008 (Figure c and d), the Denver rates are higher than the MOVES2014 rates at
all ages.
(a) Cars: Model Years 1998
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
Age (years)
(b) Trucks: Model Years 1998
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
Age (years)
40.00
35.00
3o.oo ;
25.00
20.00 ;
i5.oo ;
io.oo ;
5.00 ;
o.oo
(c) Cars: Model Years 2008
-MOVES2014: Simulated IM240
-MOVES3: Deterioration Model
40.00
35.00
r 30.oo
P 25.00
I 20.00
? 15.00
: 10.00
5.00
0.00
Age (years)
(d) Trucks: Model Years 2008
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
Age (years)
Figure 3-71 THC: Predicted trends in IM240 emissions vs age for cars and trucks in two model years
Deterioration trends normalized to 2 years are shown in Figure 3-72. Like NO*, the trends show
clearly that proportional deterioration in the current MOVES rates is substantially higher than in
the Denver IM240 dataset. However, for THC, the differences are more pronounced than for
NO.,-. The MOVES rates increase by maximum factors of 5-7 in both model years. In contrast,
the Denver-based values increase by factors between 1.7-2.0 for both cars and trucks.
150
-------
(a) Cars: Model Years 1998
(c) Cars: Model Years 2008
-MOVES2014: Simulated IM240
-MOVES3: Deterioration Model
-MOVES2014: Simulated IM240
-MOVES3: Deterioration Model
Age (years)
Age (years)
(b) Trucks: Model Years 1998
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
(d) Trucks: Model Years 2008
-•-MOVES2014: Simulated IM240
-©-MOVES3: Deterioration Model
Age (years)
Figure 3-72 THC: Deterioration ratios for cars and trucks in two model years
3.7 Running Exhaust Emission Rates (CO for MY 1990 and Later)
3.7.1 Data Source
For CO, we did not use the Denver IM240 dataset as we did for HC and NO*. When reviewing
the Denver IM240 data, we saw that emission trends with model year were contrary to
expectations. The averages for model years 2007-2010 were higher than those for MY 2004-6,
despite advances in technology over that time. For the current update, time was not available to
adequately evaluate the issue and rule out CO measurement issues.
Instead, we used a large set of remote-sensing data compiled by the Colorado Department of
Public Health and Environment (CDPHE). This dataset was collected to serve the Clean-screen
program for the Denver Metropolitan Inspection and Maintenance Program, described above.
The data is collected through the deployment of remote-sensing equipment around the city on an
ongoing basis. The scope of the dataset is similar to the Denver IM240 dataset used for HC and
NO,,
We elected to use a subset of the data that included CY2009-2014, a six-year period for which
the instruments and data processing were consistent. We excluded the two most recent calendar
years because the remote sensing contractor adopted a newer instrument and modified data
processing procedures which may have affected the observed trends over time. For modeling
purposes, we used a data subset including model years 1990-2010.
We are continuing to evaluate the RSD data and measurement methods to inform our analysis for
future versions of MOVES.
151
-------
3.7.2 Vehicle Classes
We relied on vehicle classes as defined by CDPHE. Vehicles are classified simply as "Cars"
(LDV) or "Trucks" (LDT). We did not attempt to distinguish "Light" from "Heavy" light trucks
as we did with the IM240 dataset.
The samples are very large, containing millions, rather than thousands of data points (Table
3-42). However, it is important to bear in mind that the sample sizes reflect individual
measurements, approximately 1 sec in duration, rather than the I/M test cycles which are 240
sees in duration.
In addition, a feature in remote sensing data is that some fraction of the measurements take
values that are zero or negative. Because negative emission rates are physically impossible, we
interpret the negative values as "missing." The numbers of negative values increase with model
year, as vehicles become cleaner and more difficult for the remote-sensing instrument to
quantify.
Table 3-42 Definitions for Vehicle Classes in the Denver Evaluation Sample
Category
Vehicle Class
Description
No. Meas.
(incl. negatives)
No. Meas.
(excl. negatives)
Cars
LDV
Light-Duty Vehicles
14,965,000
13,385,000
Trucks
LDT
Light-Duty Trucks
19,860,000
17,608,000
Total
34,825,000
30,993,000
The table shows total numbers of measurements, including and excluding negative. For the entire
sample, the prevalence of negatives is approximately 11% for both cars and trucks. However, the
numbers of negatives vary throughout the sample, as shown in Table 3-43 For the oldest vehicles
(model year ca. 1990), the fractions of zero/negative values are < 5%, whereas in the newest
model years (ca. 2010) the fractions exceed 25%.
Table 3-43 CO: Samples of Passenger Cars (P) and Light Trucks (T) in the Denver Evaluation Sample
CDPHE RSD | CO | Percentage of negative values from all gasoline measurements by MY and vehicle type
152
-------
3.7.3 Data Review
For the remote-sensing data, the datasets were so large that plotting all points proved impractical.
However, we did average the data to obtain trends by model year and age and plotted these
trends on linear and logarithmic scales. Note that the data shown in the plots were averaged after
excluding negative values. While biasing the results, this approach ensures that the means on
linear scale would match those on logarithmic scale, as the negative values cannot be included
when the logarithmic transforms are performed.
As neglecting the negative values is incorrect and leads to positive bias in the results, the
impression given by these plots must be discounted. Nonetheless, they are helpful in giving an
impression that guides the modeling of the dataset.
The trends in means are broadly similar to those viewed above for HC and NOx. Due to the
extremely large samples, the trends look less erratic and better behaved. The linear trends show
the characteristic fan behavior and the logarithmic trends show the parallelism evident in the
IM240 results. Between 1995 and 1996, the trucks show a larger gap at the outset of the Tier 1
standards than evident in the cars.
153
-------
Cars, Linear scale
Trucks, Linear scale
model year
o
1990
o
1991
o
1992 O 1993 O
1994
o
1995
o
1996
°
1997
o
1998
o
1999 o 2000
2001
+
2002
+
2003
+
2004
+
2005
+
2006 + 2007 +
2008
+
2009
+
2010
model year
o
1990
o
1991
o
1992 O 1993 O
1994
o
1995
o
1996
o
1997
o
1998
o
1999 o 2000
2001
+
2002
+
2003
+
2004
+
2005
+
2006 + 2007 +
2008
+
2009
2010
model year
°
1990
o
1991
o
1992 o 1993 o
1994
o
1995
o
1996
°
1997
o
1998
o
1999 o 2000
2001
+
2002
+
2003
+
2004
+
2005
+
2006 + 2007 +
2008
+
2009
2010
model year
o
1990
o
1991
o
1992 o 1993 o
1994
o
1995
o
1996
o
1997
o
1998
o
1999 o 2000
2001
+
2002
+
2003
+
2004
+
2005
+
2006 + 2007 +
2008
+
2009
2010
Figure 3-73 CO for Passenger Cars (LDV) and Light Trucks (LDT): Fuel-specific remote-sensing emissions
(g/kg), means by model year and age: (a) Cars, linear scale; (b) Cars, natural logarithmic scale; (c) Trucks,
linear scale; (d) Trucks, natural logarithmic scale
3.7.4 Model Structure
Despite the differences in data sources, the model structure for CO is identical to that used for
HC and NO*. We fit 3-piece linear splines, as previously shown in Figure 3-52, Equation 3-40
and Table 3-29. It was, however, necessary to modify the approach to assignment of knots, as
described below.
3.7.4.1 Optimizing Assignment of Knots
A surprising outcome in fitting models to such large datasets is that statistical tests could not be
used in the usual way to select among parameters and models. The reason is that all tests were
highly significant in all models. This finding necessitated a different approach to assign the knots
in the CO model.
The first step was to fit what we called "overlapping regressions." Each of these regressions is a
non-spline regression to a subset of data for five model years. The model was fit with a single
slope term and a separate intercept for each model year, as shown in Equation 3-45
154
-------
lny = b0 + bi^m + b2a + s
Equation 3-45
where:
lny = natural logarithm of fuel-specific remote-sensing emissions (g/kg),
bo = grand intercept for a reference model year,
b\ = intercept coefficient for model year, as difference from bo,
m = model year as a class or categorical variable,
62 = coefficient for age at measurement (a) as a continuous predictor (yr),
If the earliest model year was m, a model would be fit including intercepts for the set of model
years m+m+2, m+3, m+4} \ye call the models "overlapping" because the second model
would include intercepts for the set of model years +2, m +3, m +4, m +5} ^ so on^ for
successive models, through {w+16, m+17, m+18, m+19, m+20} mentioned, in our dataset, m
= 1990 and m+20 = 2010.
To account for the presence of the zero and negative values, we employed "left-censored" Tobit
regressions. These models were fit, not by OLS, but rather by maximum likelihood, with the
likelihood function modified to incorporate the negative values, treated as "censored." Each
censored value is assumed to represent some unknown positive value between 0 and an effective
"limit of quantitation." For each model, the effective limit of quantitation was assumed to be the
minimum positive measured value of InCO in the current subset of data. We fit the models using
the Lifereg procedure in SAS9.4, assuming normal distributions.®
In the compiled results, the parameters of interest are the slope terms for age and the "scale"
parameters, summarized in Table 3-44. When the Tobit model assumes a normal distribution, the
"scale" parameter represents the standard deviation of the residual errors, or the logarithmic
standard deviation of the ln-transformed CO data. Squaring this parameter gives the logarithmic
variance needed for the reverse transformation, further discussed below. Note that the scale
parameters are more uniform for the series of regressions than the corresponding variance
estimates based on the IM240 data. See Table 3-39 (page 132) and Table 3-40 (page 133).
e The options on the model statement are set to "nolog d=normal."
155
-------
Table 3-44 CO: Slope terms and logarithmic standard deviations for overlapping regressons, for cars and
trucks
Model-year Range
1990 -1994
1991 -1995
1992 -1996
1993 -1997
1994 -1998
1995 -1999
1996 -2000
1997 -2001
1998 -2002
1999 -2003
2000 -2004
2001 -2005
2002 -2006
2003 -2007
2004 -2008
2005 -2009
2006 -2010
Slope Terms
Cars
Trucks
0.03258
0.02064
0.03668
0.02766
0.03776
0.03590
0.04388
0.04336
0.04979
0.05106
0.05664
0.05699
0.06166
0.06319
0.06983
0.06454
0.07606
0.06655
0.08017
0.06789
0.08097
0.06756
0.08378
0.06695
0.08233
0.06673
0.07861
0.06542
0.07529
0.06381
0.07564
0.06541
0.07399
0.06519
Logarithmic Std. Dev.
Cars
Trucks
1.628
1.726
1.622
1.730
1.625
1.699
1.635
1.666
1.652
1.635
1.665
1.622
1.683
1.612
1.692
1.610
1.703
1.604
1.704
1.604
1.696
1.586
1.681
1.564
1.662
1.544
1.635
1.529
1.611
1.531
1.593
1.531
1.576
1.535
To lay the basis for assignment of knots, the next step was to assign the slope terms for each
model year range to the "parallelogram" shaped MY x Age blocks to which they applied, as
shown for a limited set of examples in Table 3-45. With all blocks thus vertically arranged, the
slope terms for each age level were averaged across all model years, including all repetition
within and across blocks, to produce a single slope composite slope trend by age, one for cars
and a second for trucks as shown in Figure 3-74(a) and (c).
Within the trends for cars and trucks we calculated the first differential of the composite slopes at
age a (Amca) as shown in Equation 3-46.
Amca = mca - mcaEquation 3-46
The slope differentials for cars and trucks are presented graphically in Figure 3-74 (b) and (d).
The plots for trends show the slopes for the youngest vehicles start relatively low, then increase
to very broad peaks at 8-9 years, and decline thereafter. The differential plots identify points of
inflection in the composite trends. The plots (Figure 3-74, (b) and (d)) show broad inflections at
5-7 years for cars and 7-9 years for trucks. Both cars and trucks have sharp inflections at 15
years. Based on the differentials, knots were assigned for the spline models as shown in Table
3-46.
156
-------
Table 3-45 CO for cars: Averaging blocks of slope terms for three model-year ranges: 1990-1994,1999-2003
and 2006-2010
Model Year
Ag
e (years)
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
1990
0.033
0.033
0.033
0.033
0.033
0.033
1991
0.033
0.033
0.033
0.033
0.033
0.033
1992
0.033
0.033
0.033
0.033
0.033
0.033
1993
0.033
0.033
0.033
0.033
0.033
0.033
1994
0.033
0.033
0.033
0.033
0.033
0.033
1995
1996
1997
1998
1999
0.080
0.080
0.080
0.080
0.080
0.080
2000
0.080
0.080
0.080
0.080
0.080
0.080
2001
0.080
0.080
0.080
0.080
0.080
0.080
2002
0.080
0.080
0.080
0.080
0.080
0.080
2003
0.080
0.080
0.080
0.080
0.080
0.080
2004
2005
2006
0.074
0.074
0.074
0.074
0.074
0.074
2007
0.074
0.074
0.074
0.074
0.074
0.074
2008
0.074
0.074
0.074
0.074
0.074
0.074
2009
0.074
0.074
0.074
0.074
0.074
0.074
2010
0.074
0.074
0.074
0.074
0.074
Age (years) Age (years)
Age (years) Age
Figure 3-74 CO: composite trends in age slopes and 1st differential of slope from overlapping regresssions for
cars and trucks
157
-------
Table 3-46 CO: Assignment of knots for three-piece linear spline models
Vehicle Class
ki
fi2
Passenger Cars (LDV)
1
15
Light Trucks (LDT)
8
15
3.7.5 Model Results
Model fitting results for CO for cars and trucks are shown in Table 3-47 and Table 3-48 above,
respectively. The application of the models to predict logarithmic trends is shown graphically in
Figure 3-75.
The figures are depicted in logarithmic scale. However, for clarity of presentation, they are
presented as common logarithms, i.e., base 10, despite the fact that the models were fit to natural
logarithms. On logarithmic scale, the parallelism of trends by model year, within the three
segments is easy to see. However, just as with the IM240 data, the sequencing of trends by MY
is not always monotonic.
The patterns in slopes for both CO models are similar to those for cars with NO* and THC,
although less pronounced. The slope terms for CO in the first segment are steeper than those for
HC and NO* cars, e.g., -0.07 as opposed to -0.025. The slope terms in the center segment are
slightly steeper than in the first, but the increase is smaller than for HC and NO*, e.g., -0.005 as
opposed to -0.03. In the right-hand segment, the slopes are steeper than those for HC and NO*,
and do not decline. The outcome is that based on the remote-sensing data, mean CO emission
levels continue to increase at moderate rates, even after 20 years of age.
158
-------
= 1990
= 1991
= 1992
= 1993
= 1994
= 1995
= 1996
= 1997
= 1998
= 1999
= 2000
= 2001
= 2002
= 2003
= 2004
= 2005
= 2006
= 2007
= 2008
= 2009
= 2010
\d 2
\d 3
Table 3-47 CO for Passenger Cars (LDV): Intercept and slope coefficients for the selected spline model
Intercepts
Estimate
0.8595
1.1215
1.1539
1.0126
1.0056
0.8715
0.7923
0.6199
0.6134
0.5525
0.3628
0.2244
-0.0051
-0.0624
-0.1409
-0.1132
-0.2093
-0.1695
-0.1782
-0.1243
-0.0792
0
0.07418
0.004263
-0.04531
1.656
2.743
Std Error
0.002507
0.008563
0.007702
0.006910
0.006352
0.005875
0.005419
0.005192
0.004865
0.004620
0.004361
0.004120
0.003982
0.003875
0.003786
0.003639
0.003475
0.003274
0.003109
0.002979
0.003142
0
0.000487561
0.00067944
0.001056919
0.000350825
X
117,578.27
17,152.52
22,444.15
21,475.93
25,060.45
22,004.53
21,372.50
14,258.17
15,896.99
14,303.17
6,920.33
2,965.97
1.66
259.05
1,385.64
966.83
3,627.59
2,679.87
3,285.36
1,740.48
635.32
23,145.96
39.36
1,837.51
Pr{> }
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.19694683
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
Model Year
b0 + b i
O
+
1
b q ~t~ b i - b 3 k \ - b 2
1990
1.9810
1.9511
2.6307
1991
2.0134
1.9836
2.6631
1992
1.8721
1.8423
2.5218
1993
1.8651
1.8353
2.5149
1994
1.7310
1.7011
2.3807
1995
1.6518
1.6219
2.3015
1996
1.4794
1.4496
2.1292
1997
1.4729
1.4430
2.1226
1998
1.4120
1.3821
2.0617
1999
1.2223
1.1924
1.8720
2000
1.0839
1.0540
1.7336
2001
0.8544
0.8245
1.5041
2002
0.7971
0.7673
1.4469
2003
0.7186
0.6887
1.3683
2004
0.7463
0.7165
1.3961
2005
0.6502
0.6204
1.3000
2006
0.6900
0.6602
1.3397
2007
0.6813
0.6514
1.3310
2008
0.7352
0.7054
1.3850
2009
0.7803
0.7505
1.4301
2010
0.8595
0.8297
1.5092
Slopes
b 2
b2 + b3
b2 + b3 + b4
0.0742
0.0784
0.0331
159
-------
= 1990
= 1991
= 1992
= 1993
= 1994
= 1995
= 1996
= 1997
= 1998
= 1999
= 2000
= 2001
= 2002
= 2003
= 2004
= 2005
= 2006
= 2007
= 2008
= 2009
= 2010
't/ 2
1d 3
Table 3-48 CO for Light-duty Trucks (LDT): Intercept and slope coefficients for the selected spline model
Intercepts
Estimate
0.6743
1.6983
1.5661
1.5099
1.5173
1.3885
1.3307
0.9477
0.8857
0.7590
0.5106
0.4085
0.2809
0.2192
0.1404
-0.0685
-0.0772
-0.0490
-0.0022
0.2270
0.0469
0
0.06487
0.00464
-0.04186
1.58545
2.514
Std Error
0.0018
0.0089
0.0077
0.0069
0.0059
0.0052
0.0047
0.0044
0.0040
0.0037
0.0035
0.0032
0.0031
0.0029
0.0028
0.0027
0.0025
0.0024
0.0022
0.0021
0.0025
0
0.00033
0.00056
0.00107
0.00030
X
141,025.61
36,698.98
41,844.13
48,461.88
66,848.94
72,623.62
80,753.59
46,428.13
47,826.48
41,551.27
21,795.84
16,000.66
8,241.98
5,537.81
2,450.39
666.46
952.89
430.76
1.02
11,262.74
365.85
0
38,076.22
69.36
1,543.73
Pr{>x2}
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.31367070
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
0.00000000
Model Year
"Q
+
o
¦Q
O
+
1
-r
¦Q
i
¦Q
i
"Q
+
o
¦Q
1990
2.3726
2.3355
2.9633
1991
2.2404
2.2033
2.8311
1992
2.1842
2.1471
2.7749
1993
2.1916
2.1544
2.7823
1994
2.0629
2.0257
2.6536
1995
2.0050
1.9679
2.5957
1996
1.6221
1.5849
2.2128
1997
1.5600
1.5228
2.1507
1998
1.4333
1.3962
2.0240
1999
1.1850
1.1478
1.7757
2000
1.0828
1.0457
1.6735
2001
0.9552
0.9180
1.5459
2002
0.8935
0.8564
1.4842
2003
0.8147
0.7775
1.4054
2004
0.6058
0.5686
1.1965
2005
0.5971
0.5600
1.1878
2006
0.6253
0.5882
1.2160
2007
0.6721
0.6349
1.2628
2008
0.9013
0.8641
1.4920
2009
0.7212
0.6841
1.3119
2010
0.6743
0.6372
1.2650
Slopes
b2
b2 + b 3
-r
¦Q
+
¦Q
+
¦Q
0.06487
0.06951
0.02765
160
-------
Vehicle Age at Test (years)
Vehicle Age at Test (years)
—•—1990—•—1991—•—1992—•—1993—*—1994—•—1995—•—1996—•—1997—•—1998—*—1999—•—2000
—•—2001 —*—2002 —*—2003 —•— 2004—*—2005 —•—2006 —*—2007 —•—2008—*—2009 —*—2010
Figure 3-75 CO: Three-piece linear spline deterioration models for two vehicle classes: (a) Passenger cars
and, (b) Light Duty Trucks. Note that emissions are expressed on common logarithmic scale
3.7.6 Reverse Transformation
The Tobit regression procedure cannot fit multiple variance terms as the mixed-factor model
used with HC and NO* can. For the reverse transformation with CO, we used the uniform scale
parameters fit by the spline models, shown in Table 3-47 and Table 3-48, at bottom. Previously
in Table 3-44, we saw that the multiple scale parameters fit in the sets of "overlapping"
regressions are fairly uniform. We concluded that using a uniform scale parameter is a
161
-------
After transformation, the gentle positive increases in slope at the first knot (7-8 years) gives the
CO trends an appearance of gentle upwards curvature. The decline in slope at the second knot
(15 years) is more abrupt and pronounced.
10 15
Vehicle Age at Test (years)
reasonable assumption. As the scale parameters represent standard deviations, we squared them
to represent logarithmic variances. As with the HC and NO* models, we performed the reverse
transformation using Equation 3-43.
Figure 3-76 CO: Trends in emissions vs. age as predicted by reverse-transformed three-piece ln-linear spline
models
5 10 15 20
Vehicle Age at Test (years)
—^1990—_1991—_1992 —*-1993—•—1994—*—1995—•—1996 —•—1997 —•—1998—•—1999—•—2000
—^2001 —•—2002 —*—2003 —•—2004 —2005 —•—2006 —•—2007 —•—2008 —#—2009—•—2010
3.7.6.1 Translation from Fuel to Distance Bases
Following the reverse transformation, the results still represent fuel-specific emissions, i.e., g/kg.
For use in developing MOVES emission rates, it was necessary to express the fuel-specific
emissions as mean IM240 results in mg/mi. To achieve this step, is was necessary to multiply the
fuel-specific means by corresponding fuel-consumption estimates.
162
-------
We assumed that the appropriate estimates would represent fuel consumption on the IM240
cycle. To obtain such estimates, we extracted energy-consumption rates for running operation
from the MOVES emissionRate table. After translating the energy rates to fuel-consumption,
using an appropriate heating value (41.762 kJ/g), we estimated total fuel consumed on the cycle
as a weighted sum of fuel consumption rate (kg/hr) by time-in-mode (hr) over the cycle, based
on an operating-mode distribution for the cycle. Finally, we divided the total fuel by the total
distance of the IM240 cycle (1.96 miles) to get a final result in kg fuel/mile. As MOVES does
not represent an age effect for energy or fuel consumption, we simulated IM240 fuel
consumption rates by model year, as shown for cars and trucks in Figure 3-77.
This final result was multiplied by fuel-specific CO rates (g/kg) to estimate mg CO/mi on the
IM240 cycle.
Cars
-Trucks
0.0000 -I T 1 T 1 T 1 T 1 T 1 T 1
1990 1992 1994 1996 1998 2000 2002
Model Year
Figure 3-77 Fuel consumption on the IM240 cycle, as estimated from MOVES energy-consumption rates
3.7.7 "Young Vehicle Adjustments"
Estimates from the spline models based on the remote-sensing data are consistently higher than
the simulated MOVES2014 IM240s. For cars, the spline-model values are higher than the
simulated MOVES2014 results except for the first several model years (Figure 3-78). Between
1994 and 2000, the spline predictions are higher, but the differences are smaller than those for
HC. After 2000, both trends are similar in that they settle to stable levels, but with the RSD-
based spline predictions slightly more than twice as high.
The trends for trucks are similar to those for cars, with the exception that the trends for trucks in
the final eight models years show gentle declines not evident in the trends for cars (Figure 3-79).
163
-------
-MOVES2014: Simulated IM240
-M0VES3: Deterioration Model
1994 1996
1998 2000 2002
Model Year
5,000
4,500
4,000
3,500
3,000
2,500
2,000
1,500
1,000
500
0
1990
1992
2004 2006 2008 2010
Figure 3-78 CO for Cars: Trends in estimated and simulated IM240 emissions vs. model year at age = 2 years
Model Year
Figure 3-79 CO for Trucks: Trends in estimated and simulated IM240 emissions vs. model year at age = 2
years
3.7.7.1 Calculating Adjustments
Based on these trends, as shown in for cars and for trucks, the "young-vehicle" adjustments for
each model year were calculated as for NO*, using Equation 3-44. Generally, the adjustments for
CO are larger than those for NO*, but less than half of those for HC.
The adjustments are cars are shown in Figure 3-80(a). For model years prior to 2000, the
adjustments are variable, ranging from slightly < 1.0 to -1.6. After model year 2000, the
adjustments are larger, between 2.0 and 2.6.
Prior to 2000, the adjustments for trucks follow a similar trend, but with wider variability,
ranging from -0.8 to -1.8 Figure 3-80(b). After 2000, the truck adjustments also follow a trend
similar to cars, but are slightly smaller, reaching maximum values of -2.5.
164
-------
3.00
2.50
£ 2.00
-------
8,000 ;
7,000
6,000
5,000
4,000
3,000
2,000
1,000
(a) Cars, MY 1998
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
(c) Cars, MY 2008
-•-MOVES2014: Simulated IM240
-©-MOVES3: Deterioration Model
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
(b) Trucks, MY 1998
-MOVES2014: Simulated IM240
-MOVES3: Deterioration Model
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
(d) Trucks, MY 2008
—MOVES2014: Simulated IM240
-©-MOVES3: Deterioration Model
Age (years)
Age (years)
Figure 3-81 CO: Predicted trends in IM240 emissions vs age for cars and trucks in two model years
As shown in Figure 3-82, when viewing deterioration in relative terms (that is, with emissions at
all ages normalized to emissions at age 2), the overall picture is similar to that for NO* and HC.
Generally, the spline-based trends have notably lower relative deterioration than the
MOVES2014 rates, reaching maximum ratios of 3.5 for cars and 3.0 for trucks. In contrast, for
cars in 2008 and trucks in both years, the relative deterioration in the MOVES2014 rates reaches
maxima of 4.0 to 4.5. The single exception is MY2008 cars as shown in Figure 3-82(b), in
which the relative deterioration is similar for both MOVES and the proposed update.
(a) Cars, MY 1998
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
(c) Cars, MY 2008
—MOVES2014: Simulated IM240
-©-MOVES3: Deterioration Model
(b) Trucks, MY 1998
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
(d) Trucks, MY 2008
MOVES2014: Simulated IM240
MOVES3: Deterioration Model
Figure 3-82 CO: Predicted deterioration ratios vs. age for cars and trucks in two model years
166
-------
3.8 Estimation of Emission Rates for Cold Starts
Within the MOVES modal structure, operating modes for start emissions are defined in terms of
soak time (preceding the engine start), as described above in 2.4 (page 21). This section
discusses the development of base rates for "cold starts" (operating mode 108).
Activity for start emissions are defined in terms of numbers of start events per day, combined
with distributions of soak time, both described in a separate report.2
Note that the data sources described in previous sections to estimate rates for running operation
do not include results for start emissions. Datasets available for analysis of start emissions are
more limited in size and scope.
3.8.1 Subgroup 1: Vehicles manufactured in model year 1995 and earlier
Base start emissions for passenger cars and light-duty trucks, are dependent upon two factors:
1. the (base) emissions level at approximately 75 degrees Fahrenheit,41
2. an adjustment based on the length of soak time42
These emissions were derived for MOVES2010 and have not been updated.
3.8.1.1 Data Sources
Data used in these analyses were acquired from the following four sources:
1. EPA's Mobile Source Observation Database (MSOD) as of April 27, 2005. Over
the past decades, EPA has performed emission tests (usually the Federal Test
Procedure) on large numbers of vehicles under various conditions.
We identified (in the MSOD) 549 gasoline-fueled vehicles (494 passenger cars and
55 light-duty trucks) that had FTPs performed at temperatures both within the
normal FTP range (68° to 86° Fahrenheit) as well as outside that range (i.e., either
below 68° or above 86°). Aside from the differences in ambient temperature, the
test parameters for the paired FTPs on each vehicle were identical. The FTPs were
performed at temperatures from 16 through 111° F.
2. EPA's Office of Research and Development (ORD) contracted (through the Clean
Air Vehicle Technology Center, Inc.) the testing of five cars (model years 1987
through 2001). Those vehicles were tested using both the UDDS and the IM240
cycle at temperatures of: 75, 40, 20, 0 and -20 °F.43
3. Southwest Research Institute (SwRI) tested four Tier 2 vehicles (2005 model year
car and light-duty trucks) over the UDDS at temperatures of: 75, 20, and 0 °F.44
4. During 2004-05, USEPA Office of Transportation and Air Quality (OTAQ) and
Office of Research Development (ORD), in conjunction with the Departments of
Energy and Transportation, conducted a program in the Kansas-City Metropolitan
Area. During this study, designed to measure particulate emissions, gaseous
emissions were also measured on the LA92 cycle.50
167
-------
3.8.1.2 Defining Start Emissions
Using the FTP data described above, we estimated cold-start emissions as the difference in mass
between Bag 1 and Bag 3 (g). However, because Bag 1 follows a 12-hour (720 minute) soak and
Bag 3 follows a 10-minute soak, it is possible to use soak/time relationships to modify the Bagl-
Bag3 difference so as to account for the respective soak periods. The start/soak relationships we
applied were adapted from a study performed by the California Air Resources Board.45 Based on
these data, we derived a correction factor "A" as shown in Equation 3-47 and Table 3-49.
Cold Start Emissions =
(Bag 1-Bag 3)
I-A
Equation 3-47
Table 3-49 Correction factor A for application in Equation 3-47 (MY 1995 and earlier)
Vehicle Type
THC
CO
NO,
No Catalyst
0.37101
0.34524
1.57562
Catalyst Equipped
0.12090
0.11474
0.39366
Heated Catalyst
0.05559
0.06937
1.05017
Model-year groups used to calculate start rates for vehicles in model year 1995 and earlier are
shown in Table 3-50. In some cases, model-year groups were adjusted to compensate for sparsity
of data in narrower groups. For example, the average NO* start emissions for MY 1983-1985
trucks are slightly negative. This result is possible if emissions are truly higher in FTP phase 3
than phase 1, but is likely due to erratically behaving means from small samples. Thus, these
model years were grouped with the 1981-1982 model years, which for trucks had similar
emission standards. In addition, the MY1994-1995 gasoline truck sample includes a very high-
emitting vehicle, which strongly influences the results for CO. To compensate, these vehicles
were grouped with the 1990-1993 model years. The values in the table represent the difference
of Bag-1 minus Bag-3, adjusted, as described above, to estimate cold-start emissions.
168
-------
Table 3-50 Cold-start emissions (Bag 1 - Bag 3, ad
usted) for gasoline-powered cars and trucks
Model-year
Group
II
Mean (g)
Standard deviation (g)
CV-of-the-Mean (RSE)
Years
THC
CO
NOx
THC
CO
NOx
THC
CO
NOx
Cars
1960-1980
1,488
5.172
75.832
0.608
6.948
83.812
2.088
0.035
0.029
0.089
1981-1982
2,735
3.584
52.217
1.118
7.830
60.707
1.682
0.042
0.022
0.029
1983-1985
2,958
2.912
34.286
0.922
5.216
44.785
1.321
0.033
0.024
0.026
1986-1989
6,837
2.306
21.451
1.082
2.740
32.382
1.034
0.014
0.018
0.012
1990-1993
3,778
1.910
17.550
1.149
1.728
13.953
1.034
0.015
0.013
0.015
1994-1995
333
1.788
16.233
1.027
1.203
31.648
0.742
0.037
0.107
0.040
Trucks
1960-1980
111
9.008
115.849
0.155
9.179
113.269
2.682
0.097
0.093
1.641
1981-1985
910
4.864
94.608
0.0412
4.992
67.871
1.797
0.034
0.024
1.445
1986-1989
1,192
3.804
45.918
2.107
2.298
36.356
2.152
0.017
0.023
0.030
1990-1995
1,755
3.288
40.927
2.192
4.211
42.478
2.158
0.031
0.025
0.024
3.8.2 Subgroup 2: Vehicles manufactured in MY1996 and later
Start rates for vehicles manufactured in model year 1996 and later were estimated using data
from the In-use Verification Program (IUVP), as with running rates for MY2001 and later (see
Section 3.3, page 58).
For model years 1996-2000, rates for vehicles at 0-3 years of age (ageGroup=0003) are shown
above in Table 3-16, in the row for MY2000.
For MY 2001 and later, cold-start rates (opModeID=108) were estimated as described in 3.3
above, using the data and approaches described in steps 1-4 and step 6. As with running
emissions, Figure 3-22 (page 70) and Figure 3-35 (page 87) illustrate the calculation of weighted
average FTP results for NO* by model year.
169
-------
3.9 Estimation of Emission Rates for Hot to Warm Starts
Within the MOVES modal structure, operating modes for start emissions are defined in terms of
soak time (preceding an engine start). The following section discusses the development of base
rates for "warm" or "hot" starts following seven soak periods of varying length defined in
MOVES (operating modes 101-107).
3.9.1 Subgroup 1: Model Years 2003 and earlier
3.9.1.1 Relationship between Soak Time and Start Emissions
The "cold-start," as defined and calculated above, is represented as opModeID=108. An
additional seven modes are defined in terms of soak times ranging from 3 min up to 540 min
(opModelD = 101-107). To estimate start rates for the additional seven modes, we applied soak-
time/start relationships described below. The specific values used are adapted from the
MOBILE6 soak-effect curves for catalyst-equipped vehicles.15 To adapt these relationships to the
MOVES operating modes, the soak time was divided into eight intervals, each of which was
assigned a "nominal" soak time.
For model years 1995 and earlier, we adapted and applied the soak-time adjustments used in
MOBILE6.2 for gasoline-fueled vehicles, as shown in Table 3-51. Additionally, all pre-1981
model year passenger cars and trucks use the same catalyst-equipped soak curve adjustments,
although some of these vehicles were not catalyst-equipped.
Table 3-51 Calculated soak-time adjustments, derived from MOBILE6 soak-time coefficients for catalyst-
opModelD
Soak
period
midpoint
(min)
THC
Adjustment
CO
NO,
101
3
0.051
0.034
0.093
102
18
0.269
0.194
0.347
103
45
0.525
0.433
0.872
104
75
0.634
0.622
1.130
105
105
0.645
0.728
1.129
106
240
0.734
0.791
1.118
107
540
0.909
0.914
1.053
108
720
1.000
1.000
1.000
For model years 1996-2003, soak fractions were also adapted from the approach applied in the
MOBILE model.20 Specifically, the piece-wise regression equations used in MOBILE6 for
"conventional catalyst" engines were evaluated at the midpoint of the soak period for each
operating mode (Table 3-51). For each mode, the start rate is the product of the cold-start rate
170
-------
and the corresponding soak fraction. Figure 3-83 shows the soak fractions for THC, CO and
NOi, with each value plotted at the midpoint of the respective soak period.
Soak Time (minutes)
Figure 3-83 Soak fractions applied to cold-start emissions (opModelD = 108) to estimate emissions for
shorter soak periods (operating modes 101-107, applied to MY 1996-2003)
3.9.2 Subgroup 2: Model Years 2004 and Later
The soak fractions adapted from MOBILE6 are based on data collected in the early 1990's. More
recently, the question arose as to whether they could be considered applicable to vehicles
designed to comply with Tier 2 (or LEV-II) and Tier 3 exhaust emissions standards. To address
this question, we initiated a research program during the summers of 2016 and 2017, with the
goal of examining the relationships between soak time and start emissions for a set of light-duty
vehicles certified to Tier 2 or Tier 3 standards.
Data collected by the California Air Resources Board (CARB) was also included in order to
increase the number of vehicles influencing the new soak curves.
3.9.2.1 Measuring Start Emissions using PEMS
This work differed from previous efforts in that it represents a first attempt for EPA to estimate
start emissions using portable emissions measurement systems (PEMS), rather than by using the
FTP cycle on a chassis dynamometer. During July-September, 2016, the test vehicles, outfitted
with Sensors SEMTECH-D instruments, were repeatedly driven over a 2.7-mile route in Ann
Arbor, MI, starting and ending at the National Vehicle and Fuel Emissions Laboratory (NVFEL).
The route and drive times were designed to minimize variability in trip time and idling due to
traffic conditions.
171
-------
Measurements were collected on six vehicles, one to seven years old at the time of measurement
(Table 3-52). A typical speed trace of the route is shown in Figure 3-84.
Table 3-52 EPA-Tested Light-Duty Vehicles for the Start/Soak Project
Make and Model
Model Year
Engine Displacement
Standard
Number of Trips
Ford Explorer
2009
4.0 L
Bin 4
42
Ford F150
2011
3.5 L
Bin 4
20
Saturn Outlook
2009
3.6 L
Bin 5 (ULEV)
47
Toyota Camry
2009
2.4 L
Bin 5 (ULEV)
19
Ford F150
2017
3.5 L
Bin 5 (ULEV)
13
Toyota Camry
2017
2.5 L
Tier 3 Bin 125
20
Vehicles were soaked indoors at 72° F prior to driving each repeat trip on the route. For purposes
of this analysis, only trips when the outdoor ambient outdoor temperature was above 50°F were
used. Repeat trips were performed for soak periods targeted to the midpoint times of each
MOVES operating mode (Table 3-51, page 170).
During each repeat route, the PEMS measured continuous CO2, CO, THC and NO* emissions at
a time-interval of approximately 1.0 Hz. For purposes of quality assurance, time series were
viewed to identify irregularities and measurement issues.
Soak Route Speed Profile
0 200 400 600
Trip Counter (seconds)
Figure 3-84 An Example Speed Trace for the Drive Route
In analysis of the data, it was important to verify that the route was long enough for engines to
warm up fully. To examine this question, we summarized and viewed results for catalyst and
coolant temperatures. Trends in catalyst temperatures for the measured soak periods for one
vehicle (the Explorer) are shown in Figure 3-85. These results for selected individual drives
suggest that the catalyst temperature stabilizes at 300°C or higher between 300 to 400 seconds
after engine start, depending on the duration of engine soak prior to the start. Similar results for
coolant temperatures are shown for the Toyota Camry in Figure 3-86.
An interesting result is that the catalyst takes more time to come to operating temperature for
intermediate soaks (45-240 min, operating modes 103-106) than for the longest soak period (720
min, operating mode 108). However, the coolant temperature shows the opposite pattern, with
172
-------
coolant reaching operating temperature more quickly for the intermediate soaks than for the
longer soaks.
173
-------
2009 Ford Explorer Catalyst Temperature During Soak Start Drives
350-
250-
150-
50-
350
250
150-
50-
350-
250-
O 150-
o) 50"
2 350-
-------
2009 Toyota Camry Coolant Temperature During Soak Start Drives
200-
150
100
200-
150-
100-
200-
150-
' 100~ MOVES Operating Mode
101
o>
k_
•§200- ^ _ . —102
1-150- — " a ~103
I — 104
-100" -105
® —106
8 " —108
150 ' s
tn
100
200
150- — K
100
200- -
150- X
Cj
100-
6 200 400 600 800
Trip Counter (seconds)
Figure 3-86 Mean trends in coolant temperature for the Toyota Camry, by soak period
175
-------
3.9.2.2 Measuring Soak-time Relationships on the Dynamometer
The data collected by EPA using PEMS was supplemented by a dataset collected by the
California Air Resources Board and used to update start emission rates for EMFAC2017.46
These data were measured as cycle aggregates on the California Unified Cycle. We made use of
data from Phase 1 of the cycle for 32 vehicles certified to LEV-II standards. The start phase of
the Unified cycle is approximately 300 sec in duration.
To make use of the CARB data, we assigned the soak periods used in its collection to soak
periods corresponding to MOVES start operating modes.
3.9.2.3 Comparing Dynamometer and PEMS Measurements
To obtain a broad overview of the data from both sources, we first averaged all sets of results by
vehicle and soak period.
Emissions trends by vehicle and method are shown in for THC, CO and NO* in Figure 3-87,
Figure 3-88, and Figure 3-89, respectively.
As is typical with emissions data, the trends in start emissions with soak period are highly
variable across individual vehicles in both datasets. The CARB dataset is much larger and hence
the range of variability is wider, capturing more vehicles with emissions at the low end of the
range, as well as small numbers of vehicles with unusually high emissions.
With these considerations in mind, it appears the CARB and EPA datasets are broadly similar,
both in terms of emissions levels and in the shapes of trends by pollutant. However, we can also
conclude that results derived solely from the smaller PEMS dataset would be biased high. We
also note that the PEMS dataset is limited in that only one vehicle was measured at the nine-hour
soak period (540 min). The CARB data is also valuable in covering this period, which represents
operating mode 107.
176
-------
600 0 200
Soak Time (minutes)
Vehicle
-+¦
AUDI 2010 A4
HOND 2004 ACCORD DX ~
HOND 2009 FIT
-+¦
TOTA 2007 SCION TC
2009
CAMRY
BMW 2004 325i
HOND 2004 ACCORD LX
HYND 2006 SONATA
TOTA 2007 YARIS
2009
EXPL
BMW 2005 325 i
~
HOND 2004 CRV
HYND 2007 ELANTRA
TOTA 2008 CAMRY XLE
2009
OUTL
CHRY 2007 CALIBER
-%¦
HOND 2005 ACURA TL
MB 2005 C230
-0-
TOTA 2008 ES350
2011
F150
FORD 2004 FOCUS
HOND 2006 CIVIC
MB 2006 S500
~
TOTA 2009 CAMRY
2017
CAMRY
GM 2007 CTS
HOND 2006 ELEMENT
NISS 2009 SENTRA
TOTA 2009 COROLLA LE
2017
F150
GM 2007 ION
HOND 2007 CIVIC
TOTA 2004 CAMRY LE
-*¦
VOLK 2007 JETTA
GM 2008 MALIBU
HOND 2008 ACCORD
TOTA 2005 COROLLA
VOLV 2004 S60
Figure 3-87 THC: Start emissions by soak period and vehicle for dynamometer and PEMS measurement methods
177
-------
Dynamometer (CARB)
15-
3
CO
c
o
"tn
CO
£
uj
t
(0
CO
"!5
-4->
o
10-
5-
0-
PEMS (EPA)
200
400
600 0 200
Soak Time (minutes)
400
600
Vehicle
~
AUDI 2010 A4
HOND 2004 ACCORD DX
HOND 2009 FIT
TOTA 2007
SCION TC
2009
CAMRY
BMW 2004 325i
HOND 2004 ACCORD LX
HYND 2006 SONATA
TOTA 2007
YARIS
2009
EXPL
BMW 2005 325 i
HOND 2004 CRV
HYND 2007 ELANTRA
TOTA 2008
CAMRY XLE
2009
OUTL
CHRY 2007 CALIBER
HOND 2005 ACURA TL
MB 2005 C230
TOTA 2008
ES350
2011
F150
FORD 2004 FOCUS
HOND 2006 CIVIC
MB 2006 S500
TOTA 2009
CAMRY
2017
CAMRY
¦#-
GM 2007 CTS
«#¦
HOND 2006 ELEMENT
-#¦
NtSS 2009 SENTRA
TOTA 2009
COROLLA LE
2017
F150
GM 2007 ION
HOND 2007 CIVIC
TOTA 2004 CAMRY LE VOLK 2007
JETTA
GM 2008 MALIBU
HOND 2008 ACCORD
TOTA 2005 COROLLA
VOLV 2004
S60
Figure 3-88 CO: Start emissions by soak period and vehicle for dynamometer and PEMS measurement methods
178
-------
2.0-
1.5-
3
CO
c
o
"tn
to
m 1.0-
t
CO
*-»
C/>
75 0.5-
•*->
o
0.0-
Dynamometer (CARB)
PEMS (EPA)
200
400
600 0 200
Soak Time (minutes)
400
600
Vehicle
AUDI 2010 A4
HOND 2004 ACCORD DX ~ HOND 2009 FIT
TOTA 2007
SCION TC
2009
CAMRY
BMW 2004 3251
HOND 2004 ACCORD LX HYND 2006 SONATA
TOTA 2007
YARIS
2009
EXPL
BMW 2005 325 i
HOND 2004 CRV
HYND 2007 ELANTRA
TOTA 2008
CAMRY XLE
2009
OLITL
~
CHRY 2007 CALIBER
~
HOND 2005 ACURA TL
MB 2005 C230
TOTA 2008
ES350
~
2011
F150
FORD 2004 FOCUS
HOND 2006 CIVIC
MB 2006 S500
TOTA 2009
CAMRY
2017
CAMRY
GM 2007 CTS
~
HOND 2006 ELEMENT
NISS 2009 SENTRA
TOTA 2009
COROLLA LE
2017
F150
GM 2007 ION
~
HOND 2007 CIVIC
TOTA 2004 CAMRY LE ~ VOLK 2007
J ETTA
-»•
GM 2008 MALIBU
HOND 2008 ACCORD
~ TOTA 2005 COROLLA
VOLV 2004
S60
Figure 3-89 NO*; Start emissions by soak period and vehicle for dynamometer and PEMS measurement methods
179
-------
After averaging the data by vehicle and soak period, mean soak-time trends were constructed by
following several additional steps.
Step 1: Correct for running-exhaust emissions
In addition to the emissions attributed to the excess fuel injected into cylinders during an engine
start period, we assume that typical "running emissions" and "hot-start emissions" are included
in the total. To isolate the excess emissions attributable to the start condition, we subtracted the
results for the 0-6 minute soak period from the measurements for the remaining soak periods.
This calculation was performed separately for each vehicle. This step is analogous to subtracting
Bag 3 from Bag 1 when estimating FTP start emissions.
Step 2: Average results across vehicles
Next, we averaged the means for individual vehicles across vehicle to obtain average trends. We
performed this step separately for the dynamometer and PEMS datasets.
Step 3: Calculate program-specific soak ratios
As in initial step in developing soak-time relationships, we normalized the mean emissions (in
grams) at all soak periods to those for the 12-hr soak period, i.e., cold start. We called this step
"program-specific" because we performed the normalization separately for the dynamometer and
PEMS datasets.
These intermediate ratios are shown for THC, CO and NOx in Figure 3-90, Figure 3-91 and
Figure 3-92 below, respectively. The ratios for the PEMS and dynamometer datasets are labeled
"EPA" and "CARB," respectively.
Step 4: Calculate final ratios
In this final step, we averaged the program-specific ratios for the two datasets to obtain a single
set of soak-time ratios. For each soak period, the final ratio was calculated as an average of two
intermediate ratios, weighted by numbers of vehicles in each data source for that period. The
final ratios are also shown in the figures, labelled as "EPA + CARB weighted average."
Due to the subtractions performed in step 1, the ratios for the first operating mode, opModelD
101, could not be directly estimated from the means. After correcting for running and hot-start
emissions, operating mode 101 would have had a mass of 0.0 g. To impute the ratios for this
mode, the soak ratios for the opModelD 101 was extrapolated. This fraction was estimated by
multiplying the fraction at operating mode 102 (soak time =18 minutes) by 3/18, the
proportional difference between the midpoints of the soak periods for these two operating modes.
For comparison, the figures also include the "older" soak curves, previously shown in Figure
3-83, page 171. The comparisons show the largest differences in soak curves for THC and NOx,
especially for soak times less than 240 minutes. Both the THC and NOx ratios surpass 1.0 before
the 720-minute soak mark, indicating that THC and NOx emissions from starts after less than 240
minutes soaking are greater than after 720 minutes or more.
180
-------
1.2-
Source
CARB
EPA
MOVES
< EPA + CARB Weighted Average
(with respect to number of vehicles)
o.o-
200 400
Soak Time (minutes)
600
Figure 3-90 THC: Program-specific and final soak-time ratios for Tier-2/LEV-II vehicles. The "MOVES" line
refers to values used in MOVES2014 and retained in M0VES3 for MY 2003 and earlier
181
-------
Source
EPA
- MOVES
. EPA + CARB Weighted Average
{with respect to number of vehicles)
200 400
Soak Time (minutes)
Figure 3-91 CO: Program-specific and final soak-time ratios for Tier-2/LEV-II vehicles. The "MOVES" line
refers to values used in MOVES2014 and retained in MOVES3 for MY 2003 and earlier
182
-------
o-
200 400
Soak Time (minutes)
600
Figure 3-92 NO*: Program-specific and final soak-time ratios for Tier-2/LEV-II vehicles. The "MOVES" line
refers to values used in MOVES2014 and retained in MOVES3 for MY 2003 and earlier
The final results for use in MOVES3 are shown in Table 3-53. As mentioned, these fractions will
be applied to model years 2004 and later.
183
-------
Table 3-53 Revised Soak Fractions for Light-duty Start Emissions, for MY 2004 and later
opModelD
Midpoint
Soak time
(min)
S
THC
>oak Fraction
CO
s
NQv
101
3
0.0193
0.0167
0.0509
102
18
0.1159
0.1003
0.3053
103
45
0.4974
0.3649
1.4425
104
75
0.7149
0.5732
2.0743
105
105
0.7646
0.5931
2.2659
106
240
0.8039
0.6303
2.0355
107
540
1.160
0.8719
1.8055
108
720
1.000
1.000
1.000
3.9.3 Applying Deterioration to Starts
3.9.3.1 Assessing Start Deterioration in Relation to Running Deterioration
The large datasets used to develop rates for running emissions provided much information about
deterioration for hot-running emissions, but no direct information on deterioration for start
emissions. Our best data source for start deterioration was data from the IUVP program, used to
develop running rates for NLEV and Tier 2 vehicles (see Section 3.3). However, because the
IUVP data is a relatively small data set, and restricted to vehicles in good repair, we were
concerned that it would not capture the true variation in emissions. We considered whether it
would be better to simply apply the running deterioration rates described in Sections 3.2, 3.6 and
3.7, to start emissions. To investigate this, we compared start and running deterioration in the
IUVP data. As described below, we eventually applied adjusted running deterioration rates that
accounted for the differences in start and running deterioration as seen in the IUVP data.
A valuable aspect of the IUVP data is that they provide FTP results with the measurement phases
separated. As before, we focused on cold-start emissions, calculated as Bagl - Bag3 (g), and hot-
running emissions, represented by Bag2 (g/mi). For this purpose, these data are also valuable
because they provide emissions measured over a wide range of mileage, up to 100,000 mi,
although the corresponding range of vehicle age is relatively narrow (0-5 years). Thus, we
elected to first evaluate trends in emissions vs. mileage and only later convert to the age-based
rates needed for MOVES.
Starting with the National LEV standards in MY 2001, the hydrocarbon species used for
certification is non-methane organic gases (NMOG), rather than total hydrocarbons (THC). At
the outset, we plotted the data for NMOG and NO* vs. odometer reading, on linear and
184
-------
logarithmic scales. Scatterplots of start and running NMOG emissions are shown in Figure 3-93
and Figure 3-94; corresponding plots for InNMOG are shown in Figure 3-95 and Figure 3-96.
Similarly, scatterplots of start and running NOx emissions are shown in Figure 3-97 and Figure
3-98; corresponding plots for InNOx are shown in Figure 3-99 and Figure 3-100.
In viewing the data, some observations are apparent. The data are grouped, with one group
representing vehicles measured at less than 50,000 miles, centered around 10,000-20,000 miles,
and a second group representing vehicles measured at 50,000 to 100,000 miles. Given that the
purpose of the IUVP program is compliance assessment, the two groups are designed to assess
compliance with certification (< 50,000 mi) and useful-life (>50,000 mi) standards, respectively.
As expected, distributions of emissions are skewed, but with running emissions more skewed
than start emissions. On a logarithmic scale, the degree of skew is shown by the variability of
the transformed data, with the ln(start) spanning 3-3.5 factors of e, and the ln(running) spanning
6-7 factors of e. Finally, and of most relevance to this analysis, deterioration trends are visible in
the In plots, with the masses of points at >50,000 miles centered higher than those for < 50,000
miles.
To assess the presence of trends in emissions and mileage more rigorously, we ran linear
statistical models on the ln-transformed data. To illustrate, we will focus on models run on
vehicles certified to LEV standards, as shown in Table 3-56 and Table 3-57. The model structure
includes a grand intercept for all vehicle classes (LDV, LDT1-4), and separate intercepts for each
vehicle class. All parameters are highly significant, both for InNMOG and InNOx. A more
complex model structure was attempted, which included individual mileage slopes for different
vehicle classes. However, this model was not retained, as it did not improve the fit, nor were the
interaction terms themselves significant. The covariance structure applied was simple, in that a
single residual error variance was fit for all vehicle classes.
Models were fit to vehicles certified to other standards, such as ULEV and Tier 2/Bin-5, the
results for which are not shown here. The models for ULEV show very similar patterns to those
for LEV, whereas the models fit to Bin-5 data were not considered useful as the range of mileage
covered for these more recent vehicles was not wide enough to demonstrate deterioration trends
(i.e., < 25,000 mi).
The models confirm the visual impression given by the plots of InNMOG and InNOx. Positive
trends in emissions do appear evident in these data, but the increase in emissions with mileage is
very gradual. The trends in InNOx are steeper than those for InNMOG, and the trends for
running emissions are steeper than those for start emissions. However, the differences between
the slopes for start and running are less pronounced for InNOx than for InNMOG. For InNOx, the
running slope is 1.25 times that for starts, and for InNMOG, the running slope is 1.65 times that
for starts.
185
-------
Y~i—i—ii—i—r
10000 20000
—i—1—1—i-i—r
30000 40000
50000 60300 70000 30000 90000 100000 110000 120000 130000
T 1 1 1 1
140000 150000
odometer
vehdass a D a LET2 ° Q ° LDVT1 * * a MDV2
Figure 3-93 Cold-start FTP emissions for NMOG (g) vs. odometer (mi), for LEV vehicles, from the IUVP
program
186
-------
hot running
D
O
a
Q
a
a
A
o
C
s
~
a
o
B
1
o
o
8
1
8
^ §
~
ft _
*
So
e £
*
D ^
£
0
d°
o
%
n o
s
? 0
o
k n 0 M
W
jstf -tt
cP
**
O O
i i i i
°
*4*|
<3
sCa
s * 4
0 8: Q
*
ftfia U °
1 1 1 1
O
1 1 1 1 1
1 1 1 1
o ° ^
i i i i
1 1 1 1
1 1 1 1
i i i i
fit ff nrs4-
JyLcfig
i—i—i—i—
Osf,
—i—i—i—i—
A
A
i i i i
A
i i i i
i i i i
i i i i i
1000G 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000
odometer
vehclass n n D LDT2 ° ° o LDVH ^ i & MDV2 * *= =* MDV3 + + +¦ MDV4
Figure 3-94 Hot-running (Bag 2) FTP emissions for NMOG (g/mi) vs. odometer (mi), for LEV vehicles, from
the IUVP program
187
-------
In cs
10000 20000 30000 40000 50000 60000 70000 3CODO 90000 100000 110000 120000 130000 14COCO 150000
odometer
vehclass °°° LDT2 ° o o LDVT! ^ * *= * MDV3 + + +¦ MDV4
Figure 3-95 Cold-start FTP emissions for ln(NMOG) vs. odometer (mi), for LEV vehicles, from the IUVP
program (LOGARITHMIC SCALE)
188
-------
In hr
^ —i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—|—i—i—i—i—[
10000 20000 3QOOO 40000 50000 60000 70000 30030 90000 100000 110000 120000 130000 14C0C0 150000
odometer
vehclass LDT2 o o o LDVT! ^ * & MDV2 * * * MDV3 + + +* MDV4
Figure 3-96 Hot-running (Bag 2) FTP emissions for ln(NMOG) vs. odometer (mi), for LEV vehicles, from the
IUVP program (LOGARITHMIC SCALE)
189
-------
Figure 3-97 Cold-start FTP emissions for NO.v (g) vs. odometer (mi), for LEV and ULEV vehicles, from the
IUVP program
190
-------
hot running
~ a
0
it
~
~
¦+
0
0 0
O a
0
«: °
A
o
n ~
f 0
0 6 *
D Clfl Q„ *
X
A
4 4 °C
~
a
a a B* *
0 *0
A
~
A
a
~
n A
+
a
o
~
3^
dfj
o a
B ~ B
-l. c£P
MjD
O S:
6 D Q°.
n R)
&
Q d"
~ °
O u
o
*
o
~
A
:
O tM
* 4
+ ~
o ~
* +
~
D S HO
O
B:0*®
O *
+
+¦
~
D
1000G 20000 30000 40000 50000 60000 70000 80000 90000 100000 110000 120000 130000 140000 150000
odometer
vehdass aDD LDT2 o a o LEVT1 * a a MDV2 * * * MEf\r3 + + + MDV4 + + +* MDV5
Figure 3-98 Hot-running (Bag 2) FTP emissions for NO.v (g/mi) vs. odometer (mi), for LEV and ULEV
vehicles, from the IUVP program
191
-------
*
* °
*
+
D
*
0
=t
~
*
&
d%
*
°a*
A -F*
a£*«te^
_jt . n n + HjJv
+
° **
** *
+
it
+
>
<*>=
~
-t 0 0
~
IT * n D
^ CL,"^
a ***+
S-i a,™ ^0°
A
o
4
§'°<$ s
313 O ~+ i°0il
ojn 0f>Q+o
O +
o
rfc 0
Q
Io V
ao
-------
da\+
5 0
*BD f
o° %
° * a 0,1
o * 0^,0
s
° 0 1?
°n"£ "9"*
A*
^ °° °n 4,1
+ He
p™ £aso a If
iChX^ *fl?,D • So
»0 I. o
lMW ™ ®8 „
oJLa"%
=" a & n °
° J5 a0 JB 0 ~
D ~ Odd ~ 1= ~ tf° ul
Ch^ _ *= * ^ r
u u ^
f°i Do*
So
-1—I—I—I—
—' 1—1—1—1—1—f
140000 150GOQ
vehclass ~ ~ ~ LDT2
odometer
000 LDVT1 £. a £1 MDV2
Figure 3-100 Hot-running (Bag 2) FTP emissions for In (NO*) vs. odometer (mi), for LEV vehicles from the
IUVP program
193
-------
Table 3-54 Model fit parameters for InNMOG, for LEV vehicles
Parameter
Predictor
Estimate
Standard error
Denom. D.F.
f-value
Pr > t
Cold-Start i
Bag 1 - Bag 3) i
residual error = 0.1942)
Slope
Odometer (mi)
0.000004982
0.0
2,404
CO
<0.0001
intercept
LDV-T1
-1.9603
0.02224
2,404
-88.14
<0.0001
intercept
LDT2
-1.7353
0.02429
2,404
-71.43
<0.0001
intercept
LDT3 (MDV2)
-1.5735
0.03520
2,404
-44.70
<0.0001
intercept
LDT4 (MDV3)
-1.2937
0.03233
2,404
-40.01
<0.0001
Hot-Running (Bag 2) (residual error = 1.3018)
Slope
Odometer (mi)
0.000008237
0.0
2,225
CO
<0.0001
intercept
LDV-T1
-6.1604
0.05961
2,225
-103.34
<0.0001
intercept
LDT2
-6.2554
0.06577
2,225
-95.11
<0.0001
intercept
LDT3 (MDV2)
-5.9018
0.09239
2,225
-63.88
<0.0001
intercept
LDT4 (MDV3)
-5.5949
0.08766
2,225
-63.83
<0.0001
Table 3-55 Model fit parameters for InNO y, LEV+ULEV vehicles
Parameter
Predictor
Estimate
Standard error
Denom. D.F.
f-value
Pr > t
Cold-Start (Bag 1 - Bag 3) (residual error = 0.68)
Slope
Odometer (mi)
0.000009541
0.0
1,657
GO
<0.0001
intercept
LDV-T1
-2.6039
0.05231
1,657
-50.74
<0.0001
intercept
LDT2
-2.4538
0.06056
1,657
-40.52
<0.0001
intercept
LDT3 (MDV2)
-2.0769
0.08173
1,657
-25.41
<0.0001
intercept
LDT4 (MDV3)
-1.645
0.08882
1,657
-18.52
<0.0001
Hot-Running (Bag 2) (residual error = 2.9643)
Slope
Odometer (mi)
0.000012
0.00000165
1,622
7.13
<0.0001
intercept
LDV-T1
-4.7396
0.1092
1,622
-43.40
<0.0001
intercept
LDT2
-4.9527
0.1304
1,622
-37.98
<0.0001
intercept
LDT3 (MDV2)
-4.3144
0.1740
1,622
-24.80
<0.0001
intercept
LDT4 (MDV3)
-4.1214
0.1835
1,622
-22.47
<0.0001
Having drawn these conclusions, we developed an approach to apply them to emission rate
development. To begin, we applied the statistical models by calculating predicted values of
InNMOG and InNOx at mileages from 0 (the intercept) to 155,000 miles. We reverse-
transformed the models using Equation 3-28 (page 41) to obtain predicted geometric and
arithmetic means with increasing mileage, as shown in Table 3-56 for NMOG and Table 3-57 for
NO*.
We normalized the predicted means at each mileage to the value at 0 miles to obtain a
"deterioration ratio" Rdst, by dividing each predicted value at a given mileage by the predicted
value at 0 miles (i.e., the intercept); Rdst for the intercept =1.0 (Equation 3-48).
194
-------
7) ^a,miles
'\ict — ~~3 Equation 3-48
*a,0
We took this step to express start and running trends on a comparable relative multiplicative
basis, as trends in absolute running and start emissions are clearly not comparable.
Finally, to relate start and running trends, we calculated the ratio in Rdst for start to that for
running, designated as Rrsi
D
r> det, start
-^rei ~~ ~7) Equation 3-49
det, running
Values or Rdst and R,c\ for NMOG and NOx are shown in Table 3-56 and Table 3-57,
respectively, with corresponding results shown graphically in Figure 3-101 and Figure 3-102,
respectively.
Table 3-56 Application of models for NMOG, representing emissions trends for LDV-T1 vehicles certified to
LEV standards
Parameter
Odometer (mi, xl0,000)
0
1.5
2.5
3.5
4.5
5.5
6.5
7.5
8.5
Cold Start
InNMOG
-1.960
-1.886
-1.836
-1.786
-1.736
-1.686
-1.636
-1.587
-1.537
Geometric mean
0.141
0.152
0.159
0.168
0.176
0.185
0.195
0.205
0.215
Arithmetic
mean
0.156
0.168
0.176
0.185
0.195
0.205
0.215
0.226
0.238
Deterioration
ratio (Ilk,)
1.000
1.078
1.133
1.190
1.251
1.315
1.382
1.453
1.527
Hot Running
InNMOG
-6.160
-6.037
-5.954
-5.872
-5.790
-5.707
-5.625
-5.543
-5.460
Geometric mean
0.00211
0.00239
0.00259
0.00282
0.00306
0.00332
0.00361
0.00392
0.00425
Arithmetic
mean
0.00404
0.00458
0.00497
0.00540
0.00586
0.00636
0.00691
0.00750
0.00815
Deterioration
ratio (Ilk,)
1.000
1.132
1.229
1.334
1.449
1.573
1.708
1.855
2.014
Relative Ratio
(i?rel)
1.000
0.9952
0.922
0.892
0.864
0.836
0.809
0.783
0.758
195
-------
Table 3-57 Application of models for NO*, representing emissions trends for LDV-T1 vehicles certified to LEV
standards
Parameter
Odometer (mi, xl0,000)
0
1.5
2.5
3.5
-/.5
5.5
6.5
7.5
8.5
Cold Start
InNO,
-2.604
-2.461
-2.365
-2.270
-2.175
-2.079
-1.984
-1.888
-1.793
Geometric mean
0.0740
0.0854
0.0939
0.1033
0.1137
0.1250
0.1376
0.1513
0.1665
Arithmetic mean
0.1039
0.1199
0.1319
0.1452
0.1597
0.1757
0.1933
0.2126
0.2339
Deterioration
ratio (Ilk,)
1.000
1.154
1.269
1.396
1.536
1.690
1.859
2.045
2.250
Hot Running
InNO,
-4.740
-4.560
-4.440
-4.320
-4.200
-4.080
-3.960
-3.840
-3.720
Geometric mean
0.0087
0.0105
0.0118
0.0133
0.0150
0.0169
0.0191
0.0215
0.0242
Arithmetic mean
0.0385
0.0461
0.0520
0.0586
0.0660
0.0745
0.0840
0.0947
0.1067
Deterioration
ratio (Ilk,)
1.000
1.097
1.350
1.522
1.716
1.935
2.181
2.460
2.773
Relative Ratio (R,c\)
1.000
0.964
0.940
0.918
0.895
0.874
0.852
0.832
0.811
2.5
2.0
1.5
1 10
0.5
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0
Mileage (mi, x 10,000)
Figure 3-101 LEV deterioration ratios for cold-start and hot-running NMOG emissions, plus the ratio of the
two ratios (Start: Running)
1 1
Gold-start
Hot-running
Start:Runnina
196
-------
Figure 3-102 LEV deterioration ratios for cold-start and hot-running NO.v emissions, plus the ratio of the two
ratios (Start:Running)
For both NMOG and NO*, the difference between running and start deterioration was large
enough that we decided that it was not appropriate to assume that starts deteriorate at the exactly
the same rate as running emissions. Instead we elected to use the IUVP data to estimate distinct
start deterioration assumptions.
3.9.3.2 Translation from Mileage to Age Basis (MY 1989 and earlier)
The question remained, as to how the results derived from the IUVP data and presented above
could be applied during the generation of emission rates. At the outset, a question arises from the
fact that the results shown above were generated on the basis of mileage, whereas MOVES
assigns deterioration on the basis of age. It was therefore necessary to translate the R,d from a
mileage basis to an age basis. We achieved the translation through a series of steps.
First, we assumed a rate of mileage accumulation of about 10,000 miles per yearf-47 from which
it follows that the R,d at 125,000 miles would occur at about 12.5 years of age, or would be
represented by the 10-14 year ageGroup. Accordingly, we assigned midpoints to the 0-3 and 10-
14 year ageGroups of 2 and 12.5 years, respectively, and assume that R,d declines linearly with
age. These assumptions allow calculation of a declining trend in the ratio with respect to age.
The slope of the trend is the change in ratio (AR,d) over the corresponding change in time
(Atime). Equation 3-50 shows an example of this calculation for NMOG, which is used to
represent THC in the emission rates.
f The FHWA reports light-duty vehicles traveled on average 11,576 miles per year in201847. We believe an
approximation is sufficient because the average miles traveled per year on reduce as vehicles age, and the use of age
groups already requires some approximation.
197
-------
0.675-1.0 -0.325
m,„., =
A^e
Atime 12.5 — 2
10.5
= -0.30952
Equation 3-50
The calculation of the slope lets us estimate a value of Rrel for each ageGroup.
Pelage =1-000 ~
Equation 3-51
The results, as applied for hydrocarbons and NOx, are shown in Table 3-58 and Figure 3-103.
The net result is a 15-40 percent reduction in multiplicative start deterioration, relative to running
deterioration. The ratios for hydrocarbons were also applied for CO, as the results of analyses
with CO were similar.
Table 3-58 Relative deterioration ratios (Aei), for THC and NOx, assigned to each ageGroup (Note: ratios for
AgeGroup
Age (years)
Relative Ratio (/?,ei)
THC
NOx
0-3
2
1.000
1.000
4-5
5
0.845
0.892
6-7
7
0.783
0.848
8-9
9
0.721
0.805
10-14
12.5
0.613
0.729
15-19
17.5
0.613
0.729
20 +
23
0.613
0.729
198
-------
1.000
0.900
0.800
d 0.700
^ 0.600
^ 0.500
^ 0.400
(1)
£ 0.300
¦f 0.200
02 0.100
0.000
- M/"\»
-*-lML
ana ou
10
Age (years)
15
20
25
Figure 3-103 Relative deterioration ratios (/?rei), for THC and NO, assigned to each ageGroup
3.9.3.3 Translation from Mileage to Age Basis (MY 1990 and later)
3.9.3.3.1 Start Process for NOx
As we have shown in 3.6.8 and 3.7.8, the revised analysis has yielded meaningful reductions in
proportional deterioration compared to the levels in MOVES2014. As in MOVES2014, we
propose to model deterioration for start emissions as less than but proportional to that for running
emissions. Then, we need to develop emission rates.
As in MOVES2014, the relation between start and running emissions is based on regression
analyses of data measured on the FTP cycle through the In-Use Verification Program, described
above in 3.9.3.1. As the regressions were performed on the basis of mileage, and MOVES
assesses deterioration on the basis of age, it was necessary to relate mileage to age, assuming
mileage accumulation of 12,500 mi/year, i.e., at age 1 mileage is 12,500 mi, and at age 2 mileage
= 25,000 mi, etc. (Table 3-59).
Based on the regression results, the deterioration ratio for starts (i?start)is calculated in terms of
the ratio for running (R run) &S
Rstart — 1 "I" RrunSstart,run Equation 3-52
Where -SWartmn is the relative sensitivity of start to running emissions, calculated as the ratio of
fractional differences in predicted emissions E in each ageGroup a to that at age 2, as shown in
Equation 3-53.
199
-------
-'a,st art
^2,start
- 1
Equation 3-53
The calculation of the relative sensitivity is illustrated in Table 3-59. Deterioration ratios for
running and start emissions are shown graphically in Figure 3-104.
Table 3-59 NO*: Calculation of relative sensitivity of cold-start to hot-running emissions
Age
Mileage
0
0
2
25.000
5
62.500
7
87.500
9
112.500
12.5
156.250
17.5
218.750
23
287.500
Cold-Start
InNOx
NOx (g/mi)
Norm. 2 yr
frac. diff.
-2.6039
0.1039
-2.3654
0.1319
1.0000
0.0000
-2.0076
0.1887
1.4302
0.4302
-1.7691
0.2395
1.8154
0.8154
-1.5305
0.3041
2.3044
1.3044
-1.1131
0.4616
3.4982
2.4982
-0.5168
0.8379
6.3507
5.3507
0.1391
1.6147
12.2376
11.2376
Hot-Running
InNOx
NOx (g/mi)
Norm. 2 yr
frac. diff.
-4.7396
0.0385
-4.4396
0.0520
1.0000
0.0000
-3.9896
0.0815
1.5683
0.5683
-3.6896
0.1100
2.1170
1.1170
-3.3896
0.1485
2.8577
1.8577
-2.8646
0.2510
4.8307
3.8307
-2.1146
0.5313
10.2267
9.2267
-1.2896
1.2123
23.3361
22.3361
Sensitivity
0.0000
0.7569
0.7300
0.7022
0.6522
0.5799
0.5031
rcj
cc
c
o
2.5
2.0
1.5
o 1.0
-Cars: Starts
¦Trucks: Running—~—Trucks: Starts
Figure 3-104 NO*: Deterioration ratios for running and start emissions
3.9.3.3.2 Start Process for THC
For THC, proportional deterioration for starts was calculated in relation to running emissions as
for NOi, using Equation 3-52 and Equation 3-53.
The calculation of the relative sensitivity is illustrated in Table 3-60. Deterioration ratios for
running and start emissions are shown graphically in Figure 3-105.
200
-------
Table 3-60 THC: Calculation of relative sensitivity of cold-start to hot-running emissions
Age
Mileage
0
0
2
25.000
5
62.500
7
87.500
9
112.500
12.5
156.250
17.5
218.750
23
287.500
Cold-Start
InTHC
THC (g/mi)
Norm. 2 yr
frac. diff.
-1.9603
0.1556
-1.8358
0.1763
1.0000
0.0000
-1.6489
0.2125
1.2054
0.2054
-1.5244
0.2407
1.3653
0.3653
-1.3998
0.2726
1.5464
0.5464
-1.1819
0.3390
1.9230
0.9230
-0.8705
0.4628
2.6255
1.6255
-0.5280
0.6518
3.6979
2.6979
Hot-Running
InTHC
THC (g/mi)
Norm. 2 yr
frac. diff.
-6.1604
0.0093
-5.9545
0.0114
1.0000
0.0000
-5.6456
0.0156
1.3619
0.3619
-5.4397
0.0191
1.6733
0.6733
-5.2337
0.0235
2.0559
1.0559
-4.8734
0.0337
2.9479
1.9479
-4.3586
0.0563
4.9329
3.9329
-3.7923
0.0993
8.6903
7.6903
Sensitivity
0.0000
0.5676
0.5425
0.5174
0.4738
0.4133
0.3508
2.0
ro
cc
c
o
4-»
rcj
o
(U
4-»
(U
Q
1.5
1.0
0.5
0.0
—
10
15
20
25
• Cars: Running -^>-Cars: Starts
• Trucks: Running—~—Trucks: Starts
Figure 3-105 THC: Deterioration ratios for running and start emissions
3.9.3.3.3 Start Process for CO
For CO, proportional deterioration for starts was calculated in relation to running emissions as
for NOi, using Equation 3-52 and Equation 3-53 .
The calculation of the relative sensitivity is illustrated in Table 3-61. Deterioration ratios for
running and start emissions are shown graphically in Figure 3-106.
Table 3-61 CO: Calculating the relative sensitivity of start to running deterioration
Age
Mileage
0
0
2
25.000
5
62.500
7
87.500
9
112.500
12.5
156.250
17.5
218.750
23
287.500
Cold-Start
InCO
CO (g/mi)
Norm. 2 yr
frac. diff.
-0.2186
0.9604
-0.0954
1.0863
1.0000
0.0000
0.0895
1.3068
1.2030
0.2030
0.2127
1.4782
1.3608
0.3608
0.3359
1.6721
1.5392
0.5392
0.5516
2.0745
1.9097
0.9097
0.8596
2.8229
2.5987
1.5987
1.1985
3.9616
3.6468
2.6468
Hot-Running
InCO
CO (g/mi)
Norm. 2 yr
frac. diff.
-2.7594
0.1828
-2.5333
0.2292
1.0000
0.0000
-2.1941
0.3217
1.4038
0.4038
-1.9680
0.4033
1.7600
0.7600
-1.7418
0.5057
2.2066
1.2066
-1.3461
0.7512
3.2777
2.2777
-0.7808
1.3221
5.7688
4.7688
-0.1590
2.4622
10.7436
9.7436
Sensitivity
0.0000
0.5028
0.4747
0.4469
0.3994
0.3352
0.2716
201
-------
ro
cc
c
o
4-»
ro
<_
O
(U
4-»
(U
Q
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
====&===:
33
aifl iii^M
i
i
i
s
N
II
¦
10
15
20
25
¦Cars: Running —~—Cars: Starts
¦Trucks: Running—~—Trucks: Starts
Figure 3-106 CO: Deterioration ratios for running and start emissions
3.10 Constructing Updated Rates (Model Years 1990 and Later)
Having completed the analyses described in 3.6, 3.7, 3.9.2 and 3.9.3.3, we constructed the
updated MOVES3 running and start gaseous exhaust rates for light-duty cars and trucks by
adjusting the MOVES2014 rates in the emissionRateByAge table. Note that these updates apply
only to rates for MY 1990 and later. The rates for MY 1989 and earlier are unchanged.
We did this in several steps, described below.
3.10.1 Step 1: Extract LD gasoline rates from the Input database
We extracted a subset of rates from the emissionRateByAge table in the previous MOVES
database. The scope of rates extracted is described below:
Database: MOVESDB20200123.
Pollutant/Process: Running and start exhaust for HC, CO and NO* (polprocessid = 101,
201, 301 and 102, 202, 302),
Age Group: Ages 0-3 years (ageGroupID = 3),
Operating Modes: 23 Modes for running coast/cruise/acceleration (0, 1, 11-16, 21-30,
33-40), g eight modes for start operation (101-108),
Fuel type: Gasoline (fuelTypelD) = 1,
Regulatory Class: light-duty cars (LDV) and trucks (LDT) (regClassID = 20, 30),
Model year: Model year groups from 1990-93 through 2051-2060.
g Note that operating modes 26 and 36 do not exist.
202
-------
3.10.2 Step 2: Apply Young-vehicle Adjustments to Running Rates
We applied the "young vehicle adjustments" described in 3.6.7 and 3.7.7 to calculate revised
I/M reference rates (meanBaseRatelM) in the first ageGroup (0-3 years). The adjustments were
merged into the emissionRateByAge segment on basis of regulatory class and model year. We
applied these adjustments to running rates but not to start rates. The adjustments for model year
2010 were applied to all future model years through 2060.
3.10.3 Step 3: Apply Deterioration Adjustments
We calculated revised I/M reference rates for the remaining six ageGroups, based on results of
analyses described in 3.6.8 and 3.7.8 for running emissions and 3.9.3.3 for start emissions. We
merged the deterioration adjustments into the rates segment on the basis of pollutant process,
regulatory class and ageGroup. The deterioration adjustments were applied multiplicatively and
uniformly to both running and start rates in all model years 1990-2060.
3.10.4 Step 4: Apply Non-IM Ratios
We calculated the non-I/M reference rates (meanBaseRate) from the I/M reference rates
(meanBaseRatelM) by applying non-I/M ratios. These ratios increase by ageGroup and were
merged into the rates segment on basis of pollutant process and ageGroup. These ratios are the
same values for all model years (see 3.5) and are applied multiplicatively and uniformly to both
running and start rates for both regulatory classes in all model years.
3.10.5 Step 5: Replicate Rates for Additional Fuel Types
After completing Step 4, we replicated the subset of rates for gasoline (fuelTypelD = 1) to
generate corresponding subsets for diesel (fuelTypelD = 2) and E85 (fuelTypelD = 5). Because
data on E-85 and diesel-fueled LD vehicles is lacking and at least since the introduction of Tier-2
standards, they are required to meet the same emission standards as gasoline vehicles, we found
it appropriate to use the same rates in modelling their emissions.
As we do not represent an "I/M difference" for light-duty diesel vehicles, for this fuel only, we
reset the meanBaseRatelM to equal the meanBaseRate.
For E85 and diesel, we assigned the dataSourcelD as 4900 and 4910, respectively.
3.11 Final Results for Update for MOVES3
Having completed the steps described in 3.10, we have generated a complete set of updated rates
for model years 1990 and later, encompassing the Tier 0, Tier 1, National LEV, Tier 2 and Tier 3
emissions standards.
In this section, we present and review the resulting emission rates, including comparison to rates
developed for MOVES3 in comparison to the rates used for the previous public release,
MOVES2014b. We note trends in the rates from the perspective of key variables in the table
structure. These include vehicle-specific power (for running rates), soak time (for start rates),
age (for both running and start rates) and I/M status.
Because the rates are generated by applying multiplicative factors, the patterns and trends are
generally proportional, so only a few representative examples need be shown.
203
-------
3.11.1 Trends with Vehicle-Specific Power
The operating modes for most of the rates for the running-exhaust process, with the exception of
the idle and deceleration/braking modes, are defined in term of vehicle-specific power (VSP,
kW/Mg).
We present rates for a subset of the operating modes, 21-30, which show a complete VSP trend
at moderate speed (25-50 mph), from < 0 kW/Mg (coasting) to > 30 kW/Mg (hard acceleration).
To give proper scaling, the midpoint values of VSP for each mode are used for plotting, as
shown in Table 3-62.
Table 3-62 Midpoint VSP values assigned to selected operating modes for plotting purposes
Operating Mode
Vehicle Specific Power (VSP, kW/Mg)
21
-2
22
2.5
23
4.5
24
7.5
25
10.5
27
15.0
28
21.0
29
27.0
30
34.0
The plots present the "I/M reference rate" (meanBaseRatelM) for cars (regClassID = 20, on left)
and trucks (regClassID = 30, on right). The plot shows four model years, taken as cross sections
across the long-term trend of improving technology and declining standards. The model years
1998, 2004, 2010 and 2017 represent the "Tier 1", "Onset of Tier 2", "mature Tier 2" and "onset
of Tier 3," respectively.
The appearance of all plots is generally similar, because the scaling in the rates is proportional
throughout, and because each row in the plots is scaled independently of the others. In viewing
the plots, it is important to note the differences in scales by model year.
Plots for THC, CO and NO* are presented in Figure 3-107, Figure 3-108 and Figure 3-109
below. These figures present rates for "young vehicles" in the 0-3 year ageGroup.
In all cases, the updated MOVES3 rates are higher than the previous rates in all cases at VSP <
15 kW/Mg, and in many but not all cases at VSP >15 kW/Mg. This difference is largely due to
the application of the "young-vehicle" adjustments described above, although it is not always
conspicuous at low VSP where the rates are smaller. The difference is the most marked for THC,
as the "young-vehicle" adjustments for this pollutant were often more than twice as large as for
CO and NO*. See for 3.6.7.1 for NO,, 3.6.7.3 for THC, and 3.7.7 for CO.
For CO, the updated rates are higher than the previous rates for both cars and trucks in all model
years. Of the three pollutants, CO shows the most marked increase in the steepness of the trend
at higher VSP, which may reflect the tendency towards increased CO production as the engine
shifts towards rich operation.
For THC and NO*, however, the updated rates are lower than the previous rates at higher power,
with this tendency more pronounced for trucks than cars, and becoming more pronounced for
204
-------
model years after 2004. Note that in the MOVES2014 rates, the trends for THC and NOx have
sharp "elbows" in the trends at 15 kW/Mg. These sharp increases in the trends reflect the
assumption that emissions control systems would be less effective at higher VSP, resulting in
sharper VSP trends for the "high power" modes. In this update, this assumption has been
revised, as review of more recently acquired data did not support it as described in Section
3.3.2.4. Accordingly, the MOVES3 trends in the three more recent model years appear
qualitatively similar to that for 1998, although scaled down to represent the more recent
technologies and emission standards.
205
-------
20
30
i
Ha
o
jjj
2010
§
Vehicle-Specific Power (kW/Mg)
Figure 3-107 THC; Emission rate (meanBaseRatelM in g/hr) vs. VSP for operating modes 21-30, for cars (20)
and trucks (30) in four model years (1998,2004,2010,2017), at ages 0-3 (Note that rows are scaled
independently)
206
-------
Version
MOVES3
MOVES2014
30 0
Vehjcle-Specific Power (kWVMg)
Figure 3-108 CO: Emission rate (meanBaseRatelM in g/hr) vs. VSP for operating modes 21-30, for cars (20)
and trucks (30) in four model years (1998,2004,2010,2017), at ages 0-3 (Note that rows are scaled
independently)
207
-------
Figure 3-109 NO*: Emission rate (meanBaseRatelM in g/hr) vs. VSP for operating modes 21-30, for cars (20)
and trucks (30) in four model years (1998,2004,2010,2017), at ages 0-3 (Note that rows are scaled
independently)
3.11.2 Trends with Soak Time
The operating modes for the rates for the start-exhaust process, are defined in term of soak time,
i.e., the time since the engine was last turned off, as described in 3.8.1.2 on page 168.
We present rates for the eight start operating modes, 101-108, which reflect a range in soak time
from several minutes to 12 hours (720 min), at which point we assume that the engine is
completely "cold". To give proper scaling, the midpoint values of soak time for each mode are
used for plotting, as shown in Table 3-63 below.
208
-------
Table 3-63 Midpoint soak-time values assigned to operating modes for plotting purposes
Operating Mode
Soak time (hr)
101
0.05
102
0.30
103
0.75
104
1.25
105
1.75
106
4.0
107
9.0
108
12.0
The plots present the "I/M reference rate" (meanBaseRatelM) for cars (regClassID = 20, on left)
and trucks (regClassID = 30, on right). The plot shows the same four model years used for the
VSP trends above.
As with the VSP trends, the appearance of all plots is generally similar, because the scaling in
the rates is proportional throughout, and because each row in the plots is scaled independently of
the others.
Plots for THC, CO and NOx are presented in Figure 3-110, Figure 3-111 and Figure 3-112
below, respectively. These figures present rates for "young vehicles" in the 0-3 year ageGroup.
In all three figures, note that the updated and previous trends are identical in MY1998. This
pattern follows from the fact that the "young-vehicle" adjustments were not applied to start
emissions, and also that the "older" soak-time relationships apply to this model year (see Figure
3-83, page 171). In addition, note that the rates for the "cold starts" (soak time = 12 hr,
opModeID=108) are also identical, as the "young-vehicle" adjustments were not applied to start
rates. The differences shown for the remaining seven operating modes, i.e., "warm" and "hot"
starts, reflect the differences between the "older" and "updated" soak-time relationships (see
Figure 3-90 to Figure 3-92, page 181).
For THC, the updated soak-time trends are generally similar to the older trends, but the updated
start rates are higher than before for soak times between 1.25 and 9.0 hours. For times < 1 hr, the
updated rates are lower, as the updated trend shows a less steep curvature for hot starts.
For CO, the updated trends are also generally similar in shape to the older trends, but the updated
rates are lower at all times except 12 hours.
For NOx, the updated trends differ markedly from the older trends. Rather than increasing gently
from the 12-hr soak to a broad peak at the 1.25-hr soak, the updated rates increase more steeply
from the 12-hr soak to a sharper peak at the 1.75-hr soak, then declining steeply to the 0.05-hr
soak. The updated rates for the two shortest soak times are lower than before.
209
-------
Soak time {hours)
Figure 3-110 THC: Emission rate (meanBaseRatelM, g/start) vs. soak time for operating modes 101-108, for
cars (20) and trucks (30) in four model years (1998,2004,2010,2017), at ages 0-3 (Note that rows are
scaled independently)
210
-------
Soak time (hours)
Figure 3-111 CO; Emission rate (meanBaseRatelM, g/start) vs. soak time for operating modes 101-108, for
cars (20) and trucks (30) in four model years (1998,2004,2010,2017), at ages 0-3 (Note that rows are
scaled independently)
211
-------
Figure 3-112, NO*: Emission rate (meanBaseRatelM, g/start) vs. soak time for operating modes 101-108, for
cars (20) and trucks (30) in four model years (1998,2004,2010,2017), at ages 0-3 (Note that rows are
scaled independently).
3.11.3 Trends with Age
Trends with age display the deterioration assumptions projected through the rates, reflecting a
variety of data sources and analysis methods throughout the complete set. Comparing age trends
is of particular interest because the reevaluation and revision of deterioration assumptions was
one of the chief motivations in initiating the current update.
We present subsets of rates for the MOVES ageGroups, which show complete deterioration
trends from 0-3 years through 20+ years. To give proper scaling, the midpoint values of age
ranges for each ageGroup are used for plotting, as shown in Table 3-64.
212
-------
Table 3-64 Midpoint ages for the MOVES ageGroups used for plotting
ageGroupID
Age range (yr)
Midpoint Age (yr)
3
0-3
2
405
4-5
5
607
6-7
7
809
8-9
9
1014
10-14
12.5
1519
15-19
17.5
2099
20+
23
The plots present the "I/M reference rate" (meanBaseRatelM) for cars (regClassID = 20, on left)
or trucks (regClassID = 30, on right). As with previous plots, these plot shows four model years,
although not always the same in all plots.
Unlike the previous two sets of plots, this set includes both rates for running and start operating
modes. Each plot includes two running and two start modes, but with the specific modes varying
by plot.
As before, each row in the plots is scaled independently of the others. In this set, however, the
model years are arranged in rows, so that the decline in the rates with model year is clearly
evident. In fact, for more recent model years, the age trends are difficult to see due to scaling
effects.
Plots for THC, CO and NOx are presented Figure 3-113, Figure 3-114 and Figure 3-115 below.
The figure for THC presents rates for cars, whereas those for CO and NOx present rates for
trucks.
For the running rates the updated rate at age=2 is consistently higher than the previous rates, due
to application of the "young-vehicle" adjustments. This point is particularly conspicuous for the
THC rates.
For the start rates the updated rate at age 2 is always identical to that in the previous rates in
model year 1998 and for operating mode 108 (cold start). In these cases, the lack of difference
follows from not applying the "young-vehicle" adjustments. In model year 1998, the rates at age
2 are identical because the updated soak-time relationships were not applied. For model years
following 1998, however, and for operating modes other than 108, the rates differ at all ages
because the updated soak-time relationships apply, combined with updated deterioration.
For the updated rates, the shape of the age trends is always qualitatively the same, because these
trends reflect the characteristic trends in the underlying three-piece spline deterioration models
applied in the update. While these similarities always apply, they are not always obvious in the
plots due to scaling effects.
In the MOVES2014 trends, however, the trends for MY1998 differ from those in the later model
years, due to differences in methods applied in the development of the rates for MOVES2010.
That the deterioration in the update is substantially reduced is particularly evident in the start
rates for THC and CO, and also to some degree in the start rates for NOx. While not always as
clear in the running rates, due to vertical offsets between the trends, Figure 3-69 (NOx, page
213
-------
149), Figure 3-72 (THC, page 151) and Figure 3-82 (CO, page 166) show clearly that relative or
proportional deterioration is much lower in the updated rates.
5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20
Age (years)
214
-------
Figure 3-113 THC for Cars: Emission rate (meanBaseRatelM) vs. age for two running operating modes (13,
25, g/hr) and two start modes (101,106. g/start), in four model years (1998,2004,2010,2017), (Note that
rows are scaled independently)
215
-------
Figure 3-114 CO for Trucks: Emission rate (meanBaseRatelM) vs. age for two running operating modes (15,
27, g/hr) and two start modes (102,108, g/start), in four model years (1998,2004,2010,2017). (Note that
rows are scaled independently)
5 JO 15 20 5 10 15 20 5 10 15 20 5 10 15 20
Age (years)
Figure 3-115 NO* for Trucks: Emission rate (meanBaseRatelM) vs. age for two running operating modes (21,
28, g/hr) and two start modes (103,108, g/start), in four model years (1998,2004,2010,2017). (Note that
rows are scaled independently)
3.11.4 Trends with I/M Status
The emissionRateByAge table contains two sets of rates, one representing a default "I/M
reference" condition (meanBaseRatelM), and a second representing a default "non-I/M
reference" condition (meanBaseRate).
216
-------
In the current update, as well as in MOVES2010 and MOVES2014, the meanBaseRatelM was
estimated first, as the datasets available to estimate deterioration are collected in I/M areas and in
association with I/M programs. These datasets include the Phoenix I/M evaluation sample in
MOVES2010 and MOVES2014. This dataset is still applicable in MOVES3 for model years
prior to 1990. For MOVES3, newly available datasets include the Denver Evaluation Sample
and the CDPHE remote-sensing data.
The non-I/M reference rates are estimated from the I/M references by applying ratios that vary
by age (see 3.5, page 96). Thus, in the figures below, the I/M and non-I/M rates are presented as
age trends. It is important to emphasize that the differences between the non-I/M and I/M
defaults assume complete program compliance. This difference is discounted somewhat during
model runs, based on the parameters that estimate compliance effectiveness
(IMcompli anceF actor).
Examples are presented below for Figure 3-116, Figure 3-117 and Figure 3-118 for THC, CO
and NOx, respectively. In the plots, the rates represent cars or trucks in an individual model year,
with panels for MOVES2014 and MOVES3. As with the trends with age, the plots include two
operating modes for running operation, and two for start operation.
In the MOVES2014 trends, the non-I/M trend resembles the I/M trend, as it is derived from it by
application of the ratios. Because the ratios are both multiplicative and increase with age, the
implication is that deterioration emission rates are higher and deterioration somewhat steeper in
non-I/M areas.
In the MOVES3 trends, as with the previous age trends, the characteristic shapes of the
underlying deterioration models are evident in both the I/M and non-I/M rates. In the update, the
two sets of rates are exactly proportional.
In the MOVES2014 rates, a difference between the two sets of rates is that the I/M rates tend to
stabilize in the two oldest age groups whereas the non-I/M rates continue to increase, with the
increase more marked in the start rates. These differences are based on assumptions regarding
behavior of emissions trends in non-I/M areas (see 3.2.2.3.1, page 57).
In the updated rates, by contrast, aside from application of the ratios to estimate the non-I/M
default rates, no additional assumptions were made regarding whether deterioration trends in
non-I/M areas would differ from those in I/M areas. This differs from the approach in previous
versions of MOVES as documented in Section 3.2.2.3.1. Deterioration in non-I/M areas is an
important area of uncertainty, due to the lack of large datasets outside of I/M areas. Thus, this
question remains difficult to evaluate.
217
-------
Age (years)
Figure 3-116 THC for Cars in MY1998: Emission rate (meanBaseRatelM, meanBaseRate) vs. age for two
running operating modes (14,28, g/hr) and two start modes (102,108, g/start) (Note that rows are scaled
independently)
218
-------
Age (yeans)
Figure 3-117 CO for Trucks in MY2008; Emission rate (meanBaseRatelM, meanBaseRate) vs. age for two
running operating modes (14,28, g/hr) and two start modes (102,108, g/start) (Note that rows are scaled
independently)
219
-------
Age (years)
Figure 3-118 NO*for Trucks in MY2008: Emission rate (meanBaseRatelM, meanBaseRate) vs. age for two
running operating modes (14,28, g/hr) and two start modes (102,108, g/start) (Note that rows are scaled
independently)
3.12 Development of Emission Rates representing California Standards
In general, the principle of pre-emption does not allow the states to promulgate or enact their
own vehicle emission standards. However, due to the unique severity of the air pollution issues
in Southern California, the Clean Air Act allows the state of California to seek waivers of
preemption. When granted by EPA, such waivers allow California to enact and enforce its own
emissions standards, under the condition that such standards are at least as stringent as applicable
Federal standards.
220
-------
California has enacted several such programs, beginning with Tier 0 (c. 1977-1992) and Tier 1 in
1993. These were followed by the "Low Emission Vehicle" programs, beginning with "LEV-I"
in 1994h and continuing with "LEV-IT' and "LEV-III" in 2001 and 2015, respectively. Under
the LEV programs, multiple standard levels were assigned, designated as "Transitional Low
Emission Vehicle" (TLEV), "Low Emission Vehicle" (LEV), "Ultra Low Emission Vehicle"
(ULEV) and "Super Ultra Low Emission Vehicle" (SULEV).
Although assigned the same labels, each standard level can be assigned different numeric values
for each vehicle class, i.e., LDV, LDT1, LDT2, LDT3 and LDT4. Tor simplicity, we have
assumed that the California "Medium-Duty" classes, MDV2 and MDV3, can be treated as
equivalent to Tederal LDT3 and LDT4 classes, despite differences in loaded vehicle weights.
In addition, Section 177 of the Clean Air Act allows other states to adopt California emission
standards, with the proviso that adopted standards are identical to standards for which waivers
have been granted. States do not need approval from EPA to adopt California standards. As of
2019, 13 states had elected to adopt California LEV-II standards for emissions of criteria
pollutants from varying classes of light-duty motor vehicles.48 Collectively, these states will be
called the "CA/S177" states.1 In addition, these states have adopted the LEV-III standards.49
Effectively, then, two sets of emission standards are in place throughout the United States. One
outcome of this situation is that many vehicles coming to market over the past 20 years have
been certified to both CA and Tederal standards. The analysis described in this section
incorporates this reality by applying an assumption that the emissions behavior of vehicles with
multiple certifications would be governed by the "most stringent" certification. Tor example, a
vehicle certified to Tier 2/Bin-5 in the Tederal sales regions but certified to LEV-II/SULEV in
California, is assigned to "Bin-2" or "SULEV" for purposes of developing emission rates, rather
than to Bin 5.
This section describes the process used to develop a set of emission rates representing the LEV
programs, covering model years 1994-2031. The methods used are similar to those used to
develop rates representing vehicles under the Tederal standards (NLEV, Tier 2 and Tier 3) as
described in 3.4 (page 80). In general, as the implementation of LEV standards involved higher
fractions of vehicles at lower standard levels than under the corresponding Tederal standards;
rates for a LEV program in a given model year are equal to or lower than corresponding
'Tederal" rates.
To apply this assumption, we developed the CA/S177 rates by scaling down the Tederal rates by
appropriate margins. The calculations were performed in a series of steps, with the first three
steps identical to those used to develop the Tederal rates. The following discussion assumes that
the reader is familiar with the relevant sections of this report (See 3.4.1 (page 81) to 3.4.3)).
However, the final steps differ from that used to generate the default rates, as described below in
3.12.4 and 3.12.5.
h The "National LEV" (NLEV) program was a voluntary program modeled on the LEV-I program, and applicable to
LDV, LDT1 and LDT2 vehicles.
1 These states include Colorado, Connecticut, Delaware, Maryland, Maine, Massachusetts, New Jersey, New York,
Oregon, Pennsylvania, Rhode Island, Washington and Vermont.
221
-------
3.12.1 Averaging IUVP Results
The calculation of CA/S177 rates uses the same set of average IUVP results as the default rates.
Equivalencies between Federal and corresponding LEV standards is shown in Table 3-65. Note
that the equivalences listed in the table are not exhaustive; they are limited to the subset that
were applied in developing emission rates.
Table 3-65 Selected equivalencies between Federal and corresponding CA/S177 standards
Pro
gram
Fed.
CA/S177
Tier l1
Tier l1
NLEV
LEV-I
Tier 22
LEV-II2
Vehicle Class
Fed.
CA/S177
LDV-T1
LDV-T1
LDT2
LDT2
LDT3
MDV2
LDT4
MDV3
LDV, LDT1
PC, LDT1
LDT2
LDT2
LDV, LDT1,
LDT2,3,4
PC, LDT1,
LDT2,3,4
Standard Level
Fed.
CA/S177
LDV-T1
LDV-T1
LDT2
LDT2
LDT3
MDV2
LDT4
MDV3
TLEV
TLEV
LEV
LEV
ULEV
ULEV
TLEV
TLEV
LEV
LEV
ULEV
ULEV
Bin 5
LEV
Bin 33
ULEV3
Bin 2
SULEV
1 Under Tier 1, each vehicle class was assigned a specific standard.
2 Under this program, there was no assigned correspondence between vehicle class
and standard level for the FTP standards, however, such an assignment remains in
effect for the SFTP standards.
3 This equivalence is exact for THC and CO only, for NOr, LEV-II/ULEV is
equivalent to Bin 5 (LEV-II/LEV).
3.12.2 Develop Phase-In assumptions
Differences between the CA/S177 and Federal programs are expressed primarily through the
phase-in assumptions. For this step we developed phase-in assumptions representing the phase-in
of California Tier-1, LEV-I and LEV-II programs. These assumptions cover model-years from
1994 through 2016. Starting in model year 2017 for cars, and 2018 for trucks, Federal rates are
harmonized with CA rates during the Tier 3/LEV-III phase-in and thereafter.
The CA/S177 phase-in was based on fractions of sales, grouped by standard level and model
year. The LEV phase-in, however, is simplified in that, as in the LEV-II standards, the three
largest truck classes, LDT2, 3 and 4, were consolidated into a single class, which we will refer to
as LDT234.
222
-------
Phase-in assumptions for passenger cars (PC) and light trucks (LDT1) are shown in Figure
3-119. In model year 2009 and later, the CA/S177 fleet is dominated by ULEV, SULEV and
LEV vehicles, in that order. The phase-in for trucks (LDT234) is shown in Figure 3-120
As a final step, a distinct "simplified" Federal phase-in was also developed. In this version, the
truck classes LDT2, LDT3 and LDT4 were also pooled, to facilitate comparison to the CA/S177
version.
40% --
30% :-
20% :-
10% y
0% :
I I I I I I I I
¦ Tier 1
ILEV-I/TLEV
I LEV-I/LEV
I LEV-I/ULEV
ILEV-II/LEV
I LEV-II/ULEV
ILEV-II/SULEV
Model Year
Figure 3-119 Phase-In assumptions for CA Tier-1, LEV-I and LEV-II standards for passenger cars and light-
trucks (PC, LDV, LDT1)
100% j
90% :-
80% :-
70% :-
60% :-
c
aj
u 50% --
QJ
o.
40% :-
30% y
20% :-
10% y
0% :
m
I I I I I I I
¦ Tier 1
I LEV-I/TLEV
I LEV-I/LEV
I LEV-I/ULEV
I LEV-II/LEV
I LEV-II/ULEV
I LEV-II/SULEV
Model Year
Figure 3-120 Phase-In assumptions for CA Tier-1, LEV-I and LEV-II standards for light trucks (LDT2, LDT3,
LDT4)
223
-------
3.12.3 Merge FTP .Results and Phase-In Assumptions
In this step the FTP results and phase-in assumptions were merged so as to calculate weighted
average results for composites, cold-start and hot-running emissions, as described in 3.3.2.3
(page 69). However, as the truck classes for the CA/S177 phase-in were pooled and assigned a
uniform phase-in, calculating weighted averages by truck class did not play a role in these
calculations as in the default calculations.
This step was repeated for the CA phase-in and for the Federal phase-in.j
Sets of weighted averages by model year are shown for FTP Composite Emissions (Figure 3-1,
Figure 3-121), FTP cold-start emissions (Bag 1 - Bag 3) (Figure 3-122), and FTP hot-running
emissions (Bag 2) (Figure 3-123).
J Note that the 'Federal' phase-in is identical to that used to develop the default rates.
224
-------
THC, Trucks
THC, Cars
—B— Fed TO, Tl, NLEV, T2
—CA/S177 LEV 1, LEV
II, LEV III
2005
Model Year
0.35
. 0.30
j 0.25
0.20
¦ 0.15
0.10
0.05
0.00
1995 2000 2005 2010 2015
Model Year
CO, Trucks
CO, Cars
reu iu, i x, inlnv, i
—»-CA/Sl77 LEVI, LEV
1, LEVIII
2005
Model Year
8.00
___ 7.00
4. 6.00
QO
5.00
0 4.00
Q.
1 3.00
U
o_ 2.00
"" 1.00
0.00
1995 2000 2005 2010
Model Year
NOx, Trucks
NOx, Cars
\
\
TO T1 Ml F\/ J)
\
—•— CA/S177 LEV I, LEV
, LEVIII
\
\
V
X
B D-H
. . TT3
T T I
IIIIIM.. ,l
2005
Model Year
1.00
0.90
F
0.80
QO
0.70
a;
0.60
0 SO
Q.
F
0.40
o
u
0.30
Q.
0.20
0.10
0.00
2000 2005
Model Year
Figure 3-121 Weighted average FTP composite emissions for cars and trucks, for Federal and CA/S177
standards
225
-------
THC, Trucks
THC, Cars
5.00
4.50
4.00
-22 3.50
re 3.00
2 2.50
3 2.00
£ 1.50
"" 1.00
0.50
0.00
n PpH t
T1 Ml F\/ T'
—CA/S177 LEVI, LEV
, LEVIN
1990
2005
Model Year
5.00
4.50
< 4.00
- 3.50
3.00
2.50
2.00
1.50
1.00
0.50
0.00
2000 2005 2010
Model Year
CO, Trucks
CO, Cars
40.00
35.00
3 30.00
I 25X10
!£ 20.00
u 15.00
Q.
£ io.oo
5.00
0.00
\
—Fed TO, Tl, NLEV,T2
\
—^CA/S177 LEV I, LEV
II, LEV III
J
L
~ ~
\
\
iU— n n
2000 2005
Model Year
40.00
35.00
^ 30.00
re 25-00
£ 20.00
u 15.00
Q.
£ io.oo
5.00
0.00
2000 2005 2010
Model Year
NOx, Trucks
NOx, Cars
3.00
2.50
"3
- 2.00
re
2 1.50
o
u
Q_ 1.00
i—
0.50
0.00
—B—FedT0,Tl, NLEV, T2
\
>
>
I
II, LEV III
\
^ h h g
\
3.00
2.50
>
' 2.00
1.50
1.00
0.50
0.00
%
•4
Nnjd ~
2005
Model Year
2005
Model Year
Figure 3-122 Weighted average FTP cold-start emissions, for Federal and CA/S177 standards
226
-------
THC, Trucks
THC, Cars
0.45
_ 0.40
0.35
00
™ °-30
00
•| 0.25
| 0-20
S 0.15
X
O- 0.10
^ 0.05
0.00
1990
1995
2000 2005 2010
Model Year
2015
2020
0.45
___ 0.40
4. 0.35
00
™ °-30
00
•| 0.25
| 0-20
S 0.15
X
O- 0.10
^ 0.05
0.00
1990
1995
2000 2005 2010
Model Year
2015
2020
Figure 3-123 Weighted average FTP hot-running emissions (Bag 2), for trucks and cars, under Federal and
CA/S177 standards
1995
1995
CO, Cars
2000 2005 2010
Model Year
NOx, Cars
2000 2005 2010
Model Year
2020
2020
— 3.50
E
3.00
00
« 2.50
1 2.00
5 1.50
o
J L0°
^ 0.50
0.00
0.80
— 0.70
E
0-60
00
« 0.50
1 0.40
5 0.30
o
J °-20
k 0.10
0.00
2015
2015
4.00
^ 3.50
4s 3-00
oo
W> 2.50
c 2.00
5 1.50
0
1 1.00
^ 0.50
0.00
0.80
sr- 0.70
-S. 0.60
w> 0.50
c 0.40
5 0.30
o
J 0.20
^ 0.10
0.00
1995
1995
CO, Trucks
2000 2005 2010
Model Year
NOx, Trucks
2000 2005 2010
Model Year
2015 2020
III
2015 2020
3.12.4 Scaling CA/177 Rates to Federal Rates
At this point the next step in the calculation differs from the approach used to generate the
default Federal rates. As in the calculation of the default rates, we normalized hot-running
emissions for both FTP and US06 to Federal T1 levels, represented by MY1998. However, in
this calculation, we also performed this normalization for cold-start rates. The results were sets
of ratios relative to Tier 1 for both running and start emissions.
Next, we calculated ratios of the weighted CA ratio to its Federal counterpart, by model year, as
shown in Equation 3-54,
227
-------
^CAFed - ~— Equation 3-54
Fed
where Rca-.v^a = the ratio of the CA/S177 weighted average to that for the Federal phase-in, and
Rvsd and Rc \ are ratios of respective weighted averages to that for MY 1998, in the CA/S177 and
Federal phase-ins, respectively. Note that if raw values of /levi ed were > 1.0, they were
adjusted to 1.0, under than assumption that fleet averages under the LEV program(s) would be <
corresponding averages under the Federal program(s).
Values of /levi ed are presented below. Note that ratios were calculated and applied separately for
each of the three gaseous pollutants (THC,CO,NOx) and for start emissions (opmodeid =101-
108), "FTP Bag-2" running emissions (opmodeid = 0,1, 11-16, 21-27, 33-37) and "US06"
running emissions (opmodeid = 28-30, 38-40).
In MY2017 and later, following the onset of the Tier 3/LEV-III phase-in, all ratios are set to 1.0,
to reflect an assumption that under T3, the Federal program is targeting the same NMOG+NOx
fleet average requirements as LEV-III. See Section 3.4 for more information on these rates.
228
-------
E 0.60
3
tr
o 0.40
X
Gl
t 0-20
O.OO
THC, Trucks
/
.... /
/ V-V
w
Running
¦ ¦ ¦ ¦ i . * . . i . ¦ ¦ ¦ i . ¦ ¦ .
.... 1 ... .
THC, Trucks
? '
E 0.60
3
cc
O 0.40
X
Ql
t 0.20
0.00
-0-3 Start
* Running
2000 2005 2010
Model Year
2D 00 2005 2010 2015
Model Year
CO, Trucks
CO, Trucks
E
^6
1.20
" 1.Q0
S
' Q.BO
s
E 0.50
~
cc
jj 0.40
Q.
t 0.20
O.OO
/
/
V
-i
V
B Start
• 1 Running
2000 2005 2010
Model Year
E
1*
120
10D
' O.BQ
g
E O€0
~
cc
j 0-40
Cl
t O20
OOO
t\
'X
f
J
/
*-*>£2
\
P Start
Running
2D 05
Model Year
N Ox, Trucks
NOx, Cars
1.20
1.0D
M
y1B
E 0.60
3
C£
O 0.40
X
Q.
t 0.20
O.OO
"7
*-4
1
O - Start
• Running
2005
Model Year
120
100
H
^ 0B0
s
E 0.50
3
CC
o 0.40
X
Cl
t O20
GOO
O 1 Start
¦ ¦ ¦ ¦ 1
1 4 . 1 |
¦ ¦ill
¦ i t ¦ 1
• Running
2D 05
Model Year
Figure 3-124 Ratios of relative emission levels by model year under CA/S177 and Federal standards, both
individually normalized to "Tier-1" levels (See Equation 3-54)
The LEV rates derived by application of the ratios, as described above, are shown in the plots
below. Each plot shows two panels, for cars and trucks, so that each are present in each
comparison. Note that the rates developed in this step are "I/M reference rates"
(meanBaseRatelM). The "non-I/M reference" rates were subsequently generated in relation to
the reference rates.
For each pollutant, one operating mode is shown for running emissions, and one for start
emissions. Due to the proportional scaling in the rates, single modes are sufficient to illustrate
trends and patterns.
The plots show the default Federal rates (in blue), the initial LEV rates derived by ratio as
previously described (in red). Plots are presented for THC, CO and NO.., in that order, with the
same colors used in all plots.
229
-------
Trends for THC and CO, shown in Figures Figure 3-125 to Figure 3-128, are considered first as
the patterns are very similar for these two pollutants. In addition, the qualitative patterns are
similar for running process, represented by opMode 27, and for the start process, represented by
opMode 108.
The plots show trends in rates vs MY in the first age group (0-3 years). As mentioned, the
default Federal rates are shown in blue and the initial LEV rates in red. Note that the LEV trends
for cars drop to a consistent level between MY -2010 and 2016 but then increase from 2016 to
2017, at the beginning of the LEV-III phase-in. For trucks, this behavior is more pronounced,
showing an actual "spike" between 2016 and 2018.
For NOt, shown in Figure 3-129 and Figure 3-130, the pattern differs. The LEV rates, like the
Federal rates, begin to decline at the onset of the Tier3/LEV-III phase-in, without showing any
short-term increases.
Note that the plots also show an additional green trend, labelled 'extrap.' The derivation and
significance of these trends is explained in 3.12.5 below.
the, options before final assignment
opModelD=27
regclass= Cars regclass = Trucks
modelYearlD
meanBaseRate mbrjev mbr_extrap
Figure 3-125 THC: Trends in Emissions for Federal and Estimated CA/S.177 rates, for cars and trucks at age
0-3 years, for the running emissions process (opModelD = 27)
230
-------
the, options before final assignment
opModelD=108
regclass = Cars regdass = Trucks
v
2010 2020 2010 2020
modelYearlD
meanBaseRate mbrjev mbr_extrap
Figure 3-126 THC: Trends in Emissions for Federal and Estimated CA/S.177 rates, for cars and trucks at age
0-3 years, for the start emissions process (opModelD = 108)
co, options before final assignment
opModelD=27
regclass= Cars regclass = Trucks
modelYearlD
meanBaseRate mbrjev mbr_extrap
Figure 3-127 CO: Trends in Emissions for Federal and Estimated CA/S.177 rates, for cars and trucks at age 0-
3 years, for the running emissions process (opModelD = 27)
231
-------
co, options before final assignment
opModelD=108
regclass= Cars regclass = Trucks
modelYearlD
meanBaseRate mbrjev mbr_extrap
Figure 3-128 CO: Trends in Emissions for Federal and Estimated CA/S.177 rates, for cars and trucks at age 0-
3 years, for the start emissions process (opModelD = 108)
nox, options before final assignment
opModelD=27
regclass = Cars regclass = Trucks
modelYearlD
meanBaseRate mbrjev mbr_extrap
Figure 3-129 NO* Trends in Emissions for Federal and Estimated CA/S.177 rates, for cars and trucks at age
0-3 years, for the running emissions process (opModelD = 27)
232
-------
nox, options before final assignment
opModelD=108
regclass = Cars regclass = Trucks
meanBaseRate mbrjev mbr_extrap
Figure 3-130 NO* Trends in Emissions for Federal and Estimated CA/S.177 rates, for cars and trucks at age
0-3 years, for the running emissions process (opModelD = 108)
3.12.5 Extrapolating Phase-in Trends
The charts above show that based simply on the phase-ins, disjuncts appear at the beginning of
the Tier-3 phase-in (MY 2017-2018), in which the rates increase briefly before declining again.
This behavior gives the impression that the rates during the phase-in would be higher than during
Tier 2/LEV-II, e.g., 2010-2016.
In any case, the simple application of the ratios, as described above, led to the counterintuitive
results shown in the charts above. We developed an approach to adjust and correct these rates.
In projecting the phase-in of the Tier 3 standards, we made specific assumptions. See 3.4.1, page
81 and 3.4.2, page 83. The foundational assumptions can be restated as follows:
- the Tier 3 rates would meet the same NMOG+NO.T fleet-average requirements projected
for LEV-III,
following the onset of the phase-in, the trends in emission rates in Tier 3 and LEV-III
would follow declining linear trends, and
Tier-3 rates would converge with the LEV-III rates starting in 2017 for cars, and 2018 for
trucks. The LEV-III phase-in begins earlier, in 2015, giving LEV-III a "head start." The
Federal rates start later but immediately 'catch up' at the onset of the Tier-3 phase-in.
As mentioned, the initial estimates assume that the LEV rates are meeting LEV-III fleet averages
prior to the onset of the phase-in (2015), and then actually increase before starting to decline
again.
To rectify the situation, we extrapolated the linear phase-in trends backwards to reconstruct their
behavior between 2015 and 2018. Using subsets of rates at age = 0-3 years for MY 2017, 2018,
2020 and 2021, we calculated slopes in the phase-in trends. These slopes were calculated for
each pollutant on the basis of process (running and start) and operating mode. The calculations
were performed separately for cars and trucks.
233
-------
For cars, we calculated the slopes from between 2020 and 2017 (mcar), the latter of which is the
year when the Tier-3 phase-in began for cars.
Rim, 2017 — Rim, 2020
mrar =
2020 - 2017
where R\umy is the emission rate (meanBaseRatelM) the given model year.
Similarly for trucks, we calculated the slopes between MY 2021 and 2018 (wtmck), the latter of
which is the year when the Tier-3 phase-in began for trucks.
^truck —
Rim, 2018 Rim, 2021
2021 - 2018
Then for cars, we extrapolated this slope backwards from 2017 to earlier model years
Rim,my = Rim, 2017 + (2017 — MY)mcar
where MY = 2016, 2015 and 2014, to obtain projected rates R*immy lying on the linear phase-in
trend.
And for trucks, we extrapolated the slope backwards from 2018 backwards to earlier model years
Rim,my = Rim, 2018
+ (2018 - MY)mtmcks
where MY = 2017, 2016, 2015 and 2014.
For both cars and trucks, the extrapolated value for 2014 was projected backwards for MY to
MY 2005. As mentioned, the extrapolated trends are shown in green for HC, CO and NOx start
and running emissions in Figure 3-125 to Figure 3-130 in 3.12.4 above.
Having performed the extrapolation, modified rates were assigned by applying the following
logic:
For cars:
IF MY > 2005 AND <2016, THEN
IF the initial rate (Rim,my) < the extrapolated rate (R*im,my), THEN
Reassign the rate to the extrapolated value (R*im,my),
ELSE retain the initial rate.
For trucks, the logic is identical except for the applicable model-year range:
IF MY > 2005 AND <2017, THEN
IF the initial rate (Rim,my) < the extrapolated rate (R*im,my), THEN
Reassign the rate to the extrapolated value (R*im,my),
ELSE retain the initial rate.
The plots with the final results are shown below, for the same set of operating modes, for THC,
CO and NOx. The plots show that the extrapolated trends are selected for THC and CO, both for
start and running. For NOx, the initial trends are retained.
234
-------
the, final assignments
opModelD=27
regclass= Cars regclass = Trucks
modelYearlD
meanBaseRate mbr_lev_adj
Figure 3-131 THC: Final assignments for Federal and Estimated CA/S.177 emission rates, for cars and trucks
at age 0-3 years, for the running emissions process (opModelD = 27)
the, final assignments
opModelD=108
regclass = Cars regclass = Trucks
modelYearlD
meanBaseRate mbr_lev_adj
Figure 3-132 THC: Final assignments for Federal and Estimated CA/S.177 emission rates, for cars and trucks
at age 0-3 years, for the start emissions process (opModelD = 108)
235
-------
co, final assignments
opModelD=27
regclass = Cars
regclass = Trucks
2000 2005 2010 2015 2020 2000 2005 2010 2015 2020
modelYearlD
meanBaseRate mbr_lev_adj
Figure 3-133 CO: Final assignments for Federal and Estimated CA/S.177 emission rates, for cars and trucks
at age 0-3 years, for the running emissions process (opModelD = 27)
co, final assignments
opModelD=108
regclass = Cars
regclass = Trucks
2000 2005 2010 2015 2020 2000 2005 2010 2015 2020
modelYearlD
meanBaseRate mbr_lev_adj
Figure 3-134 CO: Final assignments for Federal and Estimated CA/S.177 emission rates, for cars and trucks
at age 0-3 years, for the start emissions process (opModelD = 108)
236
-------
nox, final assignments
opModelD=27
regclass = Cars
regclass = Trucks
60 -
¦§ 40"
\
o:
\
g
\
m
i \
ra
\ \
E
\
20 -
0 -
^
2000 2005 2010 2015 2020
2000 2005 2010 2015 2020
modelYearlD
meanBaseRate —
mbr_lev_adj
Figure 3-135 NO* Final assignments for Federal and Estimated CA/S.177 emission rates, for cars and trucks
at age 0-3 years, for the running emissions process (opModelD = 27)
nox, final assignments
opModelD=108
regclass = Cars regclass = Trucks
modelYearlD
meanBaseRate mbr_lev_adj
Figure 3-136 NO*: Final assignments for Federal and Estimated CA/S.177 emission rates, for cars and trucks
at age 0-3 years, for the start emissions process (opModelD = 108)
3.12.6 Additional Steps
As mentioned, the rates developed as described represent "I/M reference rates" at age = 0-3
years. Following completion of the steps described in 3.12.1 to 3.12.5, the following three steps
were completed.
237
-------
3.12.6.1 Apply Deterioration Adjustments
To project emission rates for the remaining six ageGroups, deterioration was projected by ratio
as described for the Federal default rates in 3.10.3, page 203.
3.12.6.2 Apply Non-I/M ratios
Having projected deterioration for the "I/M reference rates" (meanBaseRatelM), we projected
the "non-I/M reference rates" (meanBaseRate) representing default emission rates in non-I/M
areas, as described for the Federal default rates in 3.10.4, page 203.
3.12.6.3 Replicate Rates for additional Fuels
Having generated I/M and non-I/M reference rates for gasoline (fuelTypelD = 1), we replicated
the gasoline rates in their entirety to represent diesel (fuelTypelD = 2) and E85 (fuelTypelD = 5),
as described in 3.10.5, page 203.
3.12.7 Availability
The emissionRateByAgeLEV table contain the subsets of CA/S177 rates and is incorporated into the
default MOVES database. Instructions for using it are available in the MOVES graphical user interface.
3.12.8 Early Adoption of National LEV Standards
The National Low Emission Vehicle Standards program was adopted in 2001. However, a group of states
in the "Northeast Trading Region" (NTR) adopted the standards early, in 1999. Using an approach
identical to that used to develop the CA/S177 rates, we developed a supplemental table for the
emissionRateByAge values representing the adoption of NLEV rates in model years 1999 and 2000. As
with the national program, "early" NLEV applied only to the LDV, LDT1 and LDT2 vehicle classes.
As with the CA/S 177 rates, we developed phase-in assumptions specific to "early" NLEV. Figure 3-137
shows that fractions of Tier-1 vehicles start declining markedly in MY 1999, whereas in the default phase-
in, the fractions for Tier 1 are 100 percent until MY2001 for LDV-T1 and LDT2. The fractions shown
apply to LDT2, as well as to LDV-T1. Vehicle classes LDT3 and LDT4 remain in Tier 1 until the onset
of Tier 2, in MY2004.
The NTR rates were developed by scaling default rates for start and running emissions down
appropriately as implied by the differences in phase-in assumptions, as performed for the LEV rates and
described in 3.12.1 through 3.12.4.
The supplemental table for early NLEV rates is stored in the MOVES default database. Instructions for
using it are available in the MOVES graphical user interface.
238
-------
c
0)
u
0)
a.
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Tier 1
ITLEV
I LEV
ULEV
1998
1999
2000
Model Year
2001
2002
Figure 3-137 Phase-in assumptions for early NLEV adoption, for LDV, LDT1 and LDT2
3.13 Rates for E-85 Vehicles
The rates developed as described in Section 3 represent gasoline-fueled conventional-technology
engines.
Because data on E-85 LD vehicles is lacking and they are required to meet the same emission
standards as gasoline vehicles, we use the start and running rates developed for gasoline vehicles
in modelling other fuels and technologies.
We replicated the entire set of gasoline rates for high-level ethanol blends, i.e., "E77" through
"E85." However, for lower-level ethanol blends (i.e., 0-20 vol. percent), the effect of ethanol
(and other effects related to blending) is represented through fuel adjustments, rather than
through the base rates, as described in this document. The development and application of fuel
adjustments is described in a separate report.22
4 Particulate-Matter Emissions from Light-Duty Gasoline
Vehicles
The emission rates for particulate matter described in this chapter are developed in two parts.
The first part (Section 4.1) derives modal emission factors and deterioration rates for vehicles
manufactured before 2004. The second part (Section 4.2) presents the updated rates in MOVES3
for vehicles manufactured since 2004, by scaling the base modal emission rates in MOVES2014
according to newer test data and applies emission rate modifications for the phase-in of future
standards.
239
-------
4.1 Particulate-Matter Emission Rates for Model Year 2004 and Earlier
Vehicles
The primary study that this chapter relies on is the Kansas City Light-duty Vehicle Emissions
Study (KCVES) conducted in 2004-2005.50 The Environmental Protection Agency and several
research partners conducted this study to quantify tailpipe particulate-matter emissions from
gasoline-fueled light duty vehicles in the Kansas City Metropolitan Area. During the summer
and winter phases, 261 and 278 vehicles were measured, respectively, with some overlap
between the phases. The measurements were conducted on a portable dynamometer using the
LA92 driving cycle under ambient temperature conditions.
Analyses of some of the data from this program are presented in the report: "Analysis of
Particulate Matter Emissions from Light-Duty Gasoline Vehicles in Kansas City,"51 This
"analysis report" (which is the partner to this chapter) presented preliminary emission rates for
PM, elemental carbon fraction (EC) and organic carbon fraction (OC), as well as temperature
adjustment factors for start and hot-running emissions processes. These preliminary results form
the basis for the emission rates developed in this chapter. The rates in the analysis report are
based on aggregate or "bag" emissions measured on the filters, and are thus, presented as
grams/start for start emissions and grams/mile for hot running operation.
The dataset included vehicles manufactured over several decades, measured at various ages
during CY2004-05. Thus, the program taken alone did not enable us to forecast emissions for
current vehicles as they age, or to backcast emissions of older vehicles when they were young.
This chapter describes the development of a deterioration model based on a comparison of
former PM studies with the KCVES. The rates from this deterioration model allow both
forecasting and backcasting as required by MOVES.
In addition, the preliminary analyses51 did not attempt to translate results measured on the LA92
cycle (used in KCVES) into terms of other cycles (such as the FTP) or to "real-world" driving.
As with the gaseous pollutants, MOVES has the capability to represent hot running "modal"
emission rates so that emissions vary depending on the driving pattern represented. The
operating modes defined for PM are the same as for the gaseous emissions (see Table 2-5). This
chapter describes how the continuous PM measurements collected in the study were used to
populate the modal rates for MOVES. Because of the reliance on continuous PM measurement,
it is worth describing the measurement procedures used in this program.
4.1.1 Particulate Measurement in the Kansas City Study
For measurements conducted on the dynamometer, vehicles were operated over the LA92
Unified Driving Cycle (see Figure 4-1). The LA92 cycle consists of three phases or "bags."
Phase 1 ("bag 1") is a "cold start" that lasts the first 310 seconds (1.18 miles). "Cold start" is
technically defined as an engine start after the vehicle has been "soaking" in a temperature
controlled facility (typically ~72°F) with the engine off. In the Kansas City study, the vehicles
were soaked overnight under ambient conditions. Phase 1 is followed by a stabilized Phase 2 or
"hot running" (311 - 1427 seconds or 8.63 miles). At the end of Phase 2, the engine is turned off
and the vehicle is allowed to "soak" in the test facility for ten minutes. At the end of the soak
period, the vehicle is started again, and is driven on the same driving schedule as Phase 1. This
Phase 3 is called a "hot start" because the vehicle is started when the engine and after-treatment
240
-------
systems are still hot. Criteria pollutants were measured both in continuous and aggregate modes.
Particulate was collected during each of the three phases on 47 mm Teflon filters at 47°C ± 2°C.
time
Figure 4-1 Phases 1 and 2 of the LA92 Cycle, representing "cold-start" and "hot-running" operation,
respectively
In addition to the gaseous pollutants measured via the constant-volume sampler (CVS),
continuous measurements of total PM mass were taken using two instruments. The first was a
Booker Systems Model RPM-101 Quartz-crystal microbalance (QCM) manufactured by Sensors,
Inc.; the second was a Thermo-MIE Inc. DataRam 4000 Nephelometer. In addition to total
mass, estimated black carbon was measured continuously with a DRI photoacoustic instrument.
In addition, integrated samples were collected and analyzed by DRI for PM gravimetric mass,
elements, elemental and organic carbon, ions, particulate and semi-volatile organic compounds,
and volatile organic air toxics. All sampling lines were heated and maintained at 47°C ± 2°C.
The samples were extracted from the dilution tunnel through a low particulate loss 2.5 |j,m
cutpoint pre-classifier. Further details and a schematic of the sampling instrumentation are
shown in Figure 4-2 and Figure 4-3.
241
-------
Diluted exhaust
at 46 C
from vehicle
tailpipe
T
to aldehyde sample aldehyde
flow comtroller A cartridge
to particle sample
flow controller
particle filter-
background
sample line
S
Backgrd HC analyzer
Air Conditioner
water
trap -
pump ¦
filter —
flow
measurement
and control
rTTTTTrn
T high CO 4
^"alyzer I
sr cartridge
¦le heated 4 vent
1 sample I
T line | |
ma
Aheated
¦>A
pump^
heated ^
sample Heated
X
Positive
Displacement
Pump (PDP)
line
flow
measurement
and control
HC analyzer
Dilute exhaust
collection bags
low CO
analyzer
NOx analyzer
C02 analyzer
Figure 4-2 Schematic of the constant-volume sampling system used in the Kansas-City Study
<-Dyno
CVS 10cm, 5
PM2.5 IMFACTORS
PHOTO ACOUSTIC
(black carbon)
QCM CART SYSTEM
(m ass)
Data RAM
(light
scattering)
DusTrak
(light
scattering)
Figure 4-3 Continuous PM analyzers and their locations in the sample line
It is worth briefly describing the apparatus used to measure PM on a continuous basis. A more
thorough description may be found in the contractor's report?0 As of the date of this program,
242
-------
measuring continuous particulate was a daunting technical challenge. Each technique has
specific advantages and disadvantages. For this study, the cumulative mass as measured on the
Teflon filters was treated as a benchmark. Thus, prior to using the continuous measurements to
estimate modal emissions, the sums of the time series for the continuous measurements were
normalized to their corresponding filter masses to compensate for systematic instrument errors.
The Quartz Crystal Microbalance measures the cumulative mass of the PM deposited on a crystal
face by measuring the change in its oscillating frequency. It is highly sensitive to many artifacts
such as water vapor and desorption of lighter organic constituents. Due to the high degree of
noise in the continuous time series, the measurements were averaged over 10 seconds, thus
damping the temporal effects of transients. The QCM can accurately capture cumulative PM
over time, however, measurement uncertainties increase for successive points in time because the
values depend on a calculated difference between two sequential, and similar, measurements.
Due to the resulting high variability, including large and rapid fluctuations from positive to
negative emissions at any given instant, and vice versa, use of the QCM measurements was not
viewed as a practical option for use in emission rate development for MOVES, except as a check
on the other instruments.
The Dustrak and Dataram both work on light-scattering principles. As such, they have very
rapid response times and can measure larger PM volumes with reasonable accuracy. However,
their accuracy degrades when measuring low PM volumes. Since most PM mass lies within the
larger particles, the instruments should be able to capture most of the continuous mass
concentrations though it may miss a substantial portion of the smaller (nano) particles. To
provide a qualitative check on this supposition, the time-series for the QCM and optical
instruments were aligned and checked to ensure that significant mass was not missed. Based on
this analysis, the Dustrak instrument was observed to be the most reliable of the 3 instruments,
and mass correction at low loads was not judged to be worth the effort given the uncertainties
involved. This time-consuming analysis was done by eye for each test and the results are not
presented in this chapter.
The photoacoustic analyzer (PA) is unique among the continuous instruments in its ability to
capture only the soot or elemental carbon components of PM. The fast analyzer detects the
resonances coming off the carbon-carbon bonds in soot. Unfortunately, there were insufficient
Thermal Optical Reflectance (TOR) elemental carbon (EC) measurements from quartz filters to
normalize the PA data, but some comparisons are shown in the contractor's report.50 In this
study, the PA data were compared qualitatively with the Dustrak and Dataram and found to be
consistent with expected ratios of elemental to total carbon during transient events, leading to the
conclusion that these instruments were largely consistent. These results are also not presented in
this chapter as every single trace was compared by eye. The data is used to determine the modal
relationship of elemental to total PM.
Due to the uncertainty of experimental measurement techniques for continuous PM at the time of
the Kansas City study, these instruments were employed only as a semi-qualitative/quantitative
means of determining modal emission rates, and the use of such data does not qualify them as
EPA recommended or approved devices or processes.
243
-------
4.1.2 New Vehicle or Zero Mile Level (ZML) Emission Rates
In this section, we develop an approach to extend the PM results from the KCVES to estimate
average emissions across the fleet. The section also compares the new vehicle results from many
different studies in order to estimate "zero mile" level (ZML) emission rates for all model years.
Before modeling deterioration, it is first necessary to capture ZML emission rates.
In constructing a model of emissions from the Kansas City data (Figure 4-4), the greatest
challenge is distinguishing between model-year and age effects. As with most datasets, this issue
arises because the program was conducted over a two-year period. As a result, it is very difficult
to distinguish the reduction in emissions with model year from the increase in emission with age.
Emissions tend to decrease as technologies are introduced on vehicles (with later model years) in
order to comply with more stringent emissions standards. However, these technologies and
vehicles tend to deteriorate over time, thus for the same model year vehicle, older vehicles
(greater age) will have higher emissions (on average) than newer vehicles.
100
90
80
70
60
1
"5) 50
30
20 :
: I J- i " " II -
10 ; {
1 1 1 1 ¦ "
o 1 ¦ 1 , ¦ ¦ M II t L , , I U , ; i .
1975 1980 1985 1990 1995 2000
Model Year
Figure 4-4 Average particulate emission rates from the Kansas City study, by model year, shown as cycle
aggregates on the LA92 The five year averages (e.g. 1988-1991,1993-1997,1998-2002) are also shown
without error bar
In concept, the most accurate means of quantifying emissions from vehicles over time is to
conduct a longitudinal study, where emissions are measured for the same vehicles over several
(or many) years. However, implementing such a study would be costly. Moreover, it is
impossible to obtain recent model year vehicles that have been significantly aged. In the
following sections, we will describe some limited longitudinal studies conducted in the past.
~ KC measured
¦ KC 5 yr measured avg
|
|
,
1
¦
T
1 1 1 1
i 1
<• _L
¦
.. i
—•—i—
h . h
244
-------
Then, we will present our modeling methodology to isolate model year (technology) in this
chapter from age (deterioration) in the next.
4.1.2.1 Longitudinal Studies
There have been a few longitudinal studies conducted in the past that are relevant for PM
emissions. Unfortunately, they are all limited in their ability to conclusively distinguish model-
year effects from age effects.
Gibbs et al. (1979) measured emissions from 56 vehicles with mileage ranging from 0 to 55,000
miles (odometer) on 3 different cycles.52 Hydrocarbon emissions were analyzed, but
unfortunately, PM results were not reported as a function of mileage. The authors state that
"emission rates of measured pollutants were not found to be a consistent function of vehicle
mileage," however, the following figure shows that some increasing trend seems to exist for
THC (Figure 4-5).
3.5
3 -
2.5 -
E
2 -
u>
o
1.5 -
I
1 -
0.5 -
0 -
0 10 20 30 40 50
mileage (*1000)
60
Figure 4-5 Hydrocarbon emissions as a function of mileage (Gibbs et al., 1979)
Hammerle et al. (1992) measured PM from two vehicles over 100,000 miles.53 However, their
results for PM deterioration are somewhat inconclusive, as the following figure shows, since the
deterioration seems to occur mainly in the beginning of life, with very little occurring after
20,000 miles. Also, the study is limited to two specific vehicle models.
245
-------
Figure 4-6 Particulate emissions as a function of odometer for two Ford vehicles (Hammerle et al., 1992)
Both of these studies assume that odometer is a surrogate for age. While there are some
deterioration mechanisms that worsen with mileage accumulation, there are others that
deteriorate with effects that occur over time, such as corrosion due to the elements, deposits and
impurities collecting in the gas tank and fuel system, etc. Therefore, we believe that any study
that describes deterioration as a function of odometer (alone) may not account for all causes of
deterioration.
Whitney (2000) re-recruited 5 vehicles that had been measured in previous study 2 years prior
(CRC-E24).54 There are two significant limitations of this follow-up study: (1) the interval
between studies was only 2 years, though the odometers had increased 22,200 miles (on average)
and (2) these vehicles were tested on a different drive cycle, the LA92 compared to the previous
study, which used the FTP. We will explore the potential cycle differences on PM later, but
assuming the cycles give similar PM results, the PM emissions were only 8 percent higher (on
average). This increase is due to a single vehicle, which had significantly increased PM
emissions (the rest were the same or slightly lower). Unfortunately, this is not a large enough
sample and time period on which to resolve age effects, but it may be sufficient to conclude that
the differences between PM from the FTP and LA92 drive cycles are minimal for PM.
The three longitudinal studies described above are inconclusive, though they do hint that
deterioration does occur.
4.1.2.2 New Vehicle, or ZML Emission Rates and Cycle Effects
In order to isolate the effect of model year (technology) from age (deterioration), it is useful to
look at the model-year effect independently. This goal can be achieved by analyzing emissions
from new vehicles from historical studies. New vehicle emission rates tend to have lower
variability than older vehicles (in absolute terms) since they have lower emissions that comply
with more stringent THC standards. These standards, which decrease over time, tend to affect
PM emissions as well since many of the mechanisms for HC formation also form PM.
246
-------
Several independent studies have measured PM emissions from nearly new vehicles. For our
purposes, we will define "new" as a vehicle less than 3 years old, i.e., vehicles within the 0-3
year age Group. Table 4-1 lists the 15 studies employed for this analysis.
Table 4-1 Historical gasoline PM studies including new vehicles at time of study
Program
Year of study
No.
vehicles
Drive cycle
Gibbs et al.52
1979
27
FTP
Cadle et al.55
1979
3
FTP
Urban & Garbe56,57
1979, 1980
8
FTP
Lang etal.5g
1981
8
FTP
Volkswagen59
1991
7
FTP
CARB60
1986
5
FTP
Hammerle et al., 199253
1992
2
FTP
CRC E24-1 (Denver)61
1996
11
FTP
CRC E24-2 (Riverside)62
1997
20
FTP
CRC E24-3 (San
Antonio)63
1998
12
FTP
Chase et al64
2000
19
FTP
Whitney (SwRI)54
1999
LA92
KC (summer)50,51
2004
13
LA92
EPA (MSAT)65
2006
4
FTP
Li et al., 200666
2006
3
FTP, LA92
Before we examine these emissions, we should convince ourselves that the LA92 driving cycle
will not give substantially different PM emissions than the FTP so that we can compare these test
programs directly. As described above, the results from Whitney (2000) seem to indicate little
difference between the two cycles. Even though the tests were conducted 2 years apart, one
would expect that the aging effects in combination with the slightly more aggressive LA92 cycle
(used later) would have given higher PM emissions. However, this was not the case, and only
one of the 5 vehicles showed significantly increased emissions.
Li et al., (2006) measured three vehicles on both cycles at the University of California,
Riverside.66 The PM emissions from the LA92 were 3.5 time larger (on average) than the FTP
results. However, the HC emissions were only 1.2 times higher. These results seem rather
contradictory and inconclusive. The 3.5 factor also seems excessive in relation to other results,
such as the one conducted by Whitney (2000).
Finally, the California Air Resources Board conducted an extensive program over several years
comparing many different drive cycles. Unfortunately, PM was not measured in this program.
However, Figure 4-7 shows the HC emissions compared for the two cycles. The trends indicate
little difference on average between the LA92 and FTP cycles for HC.
247
-------
FTP
Figure 4-7 Hydrocarbon emissions on the LA92 versus corresponding results on the FTP cycle
Based on these studies, we conclude that there is little difference in PM emissions between the
LA92 and FTP cycles on an aggregate basis (though their bag by bag emissions may differ). We
shall demonstrate that, for the purposes of ZML analysis, the overall results will be nearly
identical even if we omit the LA92 data, thus minimizing the significance of this issue.
Figure 4-8 shows the new-vehicle emission rates from the studies listed in Table 4-1. The data
points represent each individual test, and the points with error bars represent the average for each
source. The plot presents evidence of an exponential trend (fit included) of decreasing emissions
with increasing model year. The fit is also nearly identical if we omit the two programs that
employed the LA92 cycle. We will use this exponential ZML relationship as the baseline on
which to build a deterioration model. However, the measurements from the older programs
primarily measured total particulate matter. These have been converted to PMio (for the plot),
which is nearly identical (about 97 percent of total PM is PMio). We also assume that 90 percent
of PMio is PM2.5 (EPA, 1981).67 For the older studies, we accounted for sulfur and lead directly
if they were reported in the documentation. In those cases where sulfur was not reported, the
levels were approximated using sulfur emission factors from MOBILE6 and subtracted as an
adjustment.
Unfortunately, many of the older studies used a variety of methods for measuring particulate
matter. There were many differences in filter media, sampling temperature, sample length,
dilution, dynamometer load/settings etc. It is beyond the scope of this project to normalize all of
the studies to a common PM metric. It is likely that documentation is not sufficient to even
attempt it. Therefore, no attempts at adjustment or normalization were made except for size
fraction, lead and sulfur, as described above.
248
-------
40
35
30
25
20
15
10
5
0
all
¦
Gibbs et al., 1979
o
Cadleet aL, 1979
~
Urban&Garbe, 1980
A
X
Lang et al., 1981
VW, 1991
¦
CARB, 1986
Hammerleetal., 1992
•
~
E24-1 Denver
A
E24-2UC Riverside
E24-3 San Antonio
•
O
Chrysler/Ford/GM
SwRi/NREL
X
KC-Summer LA92
+
M SAT-Tier2
• mean for each program
^—Exponential Fit
25
0 5 10 15 20
Model Year (+1975)
Figure 4-8 Particulate emission rates for new vehicles compiled from 14 independent studies
30
To estimate the ZML emission rates from these data, the next step was to separate results for cars
and trucks, and to separate cold-start from hot-running emissions. Unfortunately, the historical
data does not present PM results by cycle phase. Therefore, the 2005 hot-running ZMLs for cars
vs. trucks were calculated from the KCVES dataset, and the model-year exponential trend from
the aggregate trendline (-0.08136) is used to extend the ZMLs back to model year 1975. The
base hot running ZML emission rate for cars (LDV) (£hr,j') is:
T1 ~~0.814v
hr,v — hr,2005 Equation 4-1
where
y = model year - 1975, and
£hr,2005 = hot running ZML rate for MY 2005.
To estimate equivalent rates for trucks, we multiplied this expression by a factor of 1.43. This
value is based on an average of all the studies with new vehicles from 1992 onward (before this
model year, there were no trucks measured). It is also multiplied by 0.898 to give hot running
bag 2 rates and 1.972 to give the cold start emission rate (here defined as bag 1-bag 3 in units of
g/mi). These values were estimated by running a general linear model of bag 2 and bagl-3 with
respect to composite PM, respectively, using SPSS statistical software. The averages of these
249
-------
ratios by model year are shown in Figure 4-9, in which no clear trend is discernible. The
parameters of the model are summarized in Table 4-2.
O
o
p
~ coldstart/comp
¦ bag2/cornp
+ *
-~—
~~ *
++
+ *
1960 1970 1980 1990
model year
2000
2010
Figure 4-9 Ratios of hot-running/composite and cold-start/composite, Bag2 and Bagl-Bag3, respectively,
averaged by model year
Table 4-2 Best-fit parameters for cold-start and hot-running ZML emission rates
Parameter
Value
LDV hot-running ZML (g/mi)
0.01558
Exponential slope
0.08136
Truck/car ratio
1.42600
Bag-2 coefficient
0.89761
Cold-start coefficient
1.97218
Figure 4-10 shows the ZML emission rates. The rates are assumed to level off for model years
before 1975.
250
-------
Figure 4-10 Particulate ZML emission rates (g/mi) for cold-start and hot-running emissions, for LDV and
LDT
4.1.2.3 Aging or Deterioration in Emission Rates
In this section, a deterioration model is introduced that captures how new vehicles in all model
years deteriorate over time so that gasoline PM in any given calendar year can be modeled in
MOVES. The purpose of this model is to characterize the PM emissions from the fleet and to
backcast the past as well as forecast the future, as required in MOVES
The ZMLs determined in the previous section represent baseline emissions for new vehicles in
each model-year group. By comparing the emissions from the "aged" Kansas City vehicles in
calendar year 2005, to the new rates determined earlier, we can deduce the "age effect" for each
corresponding age. However, simple an approach as this seems, there are many ways to connect
the two points. This section describes the procedure and the assumptions made to determine the
rate at which vehicle PM emissions age.
We first break the data into ageGroups. We use the MOVES age groups which correspond to the
following age intervals: 0-3 (new), 4-5, 6-7, 8-9, 10-14, 15-19, 20+.
As a first step, the bag measurements from all of the vehicles measured in Kansas City were
adjusted for temperature using the equation derived in the analysis report.51 The equation used is:
251
-------
r r 0.03344 (12-T) „ .
pm 72 — pm T Equation 4-2
where /m>\i.72, is the adjusted rate at 72°F for cold-start or hot-running emissions, Evuj is the
corresponding measured emissions for cold-start or hot-running, respectively, at temperature T,
respectively.
The temperature-adjusted measurements are the "aged" rates, i.e., the rates in each model-year
group represent emissions for that group at the age of measurement in 2004-05, at 72°F rather
than at the actual ambient temperature.
The method adopted is to ratio the aged rates with the new rates so that the changes with
deterioration rates are all proportional. This approach will be referred to as the "multiplicative
deterioration model," and is analogous to the approach used with the gaseous emissions (Section
3.6 and 3.7).
It is likely that some of the same mechanisms that cause HC and CO to increase over time would
also result in PM increases. These factors include deterioration in the catalyst, fuel control,
air:fuel-ratio control, failed oxygen sensors, worn engine parts, oil leaks, etc. Figure 4-11 shows
trends in the natural logarithm of THC rates over approximately 10 years, based on random-
evaluation samples in the Phoenix I/M program. On a log-linear scale, the deterioration trends
appear approximately linear over this time period, suggesting that the deterioration rates are
exponential. This observation, combined with the approximate parallelism of the trends for
successive model years, implies that emissions follow a multiplicative pattern across model-year
or technology groups, calling for a multiplicative deterioration model. In such a model, the aged
rates and the new rates are converted to a logarithmic scale, after which the slopes are estimated
by fitting a general linear model. The average slope is estimated, with the ZMLs determined
earlier defining the j'-axis offsets. The result is a series of ladder-like linear trends in log scale as
shown in Figure 4-12. The lines fan out exponentially on a linear scale as shown in Figure 4-13.
The dotted lines and the points with uncertainty bars represent the Kansas City data overlaid onto
the model and indicate that the model is consistent with the data.
252
-------
LDV WEIGHTED
ln(THC) vs. Age (years), LDV
-------
Age (years)
Figure 4-13 The multiplicative deterioration model shown on a linear scale. The y-axis offsets capture the
new-vehicle ZML rates. The dotted lines and points with error bars represents the Kansas-City results (with
95 percent confidence intervals)
We applied the multiplicative deterioration factors directly to both cars and trucks, cold start,
hot-running, EC, and OC emissions, assuming that the deterioration factors are independent of
these effects. The estimation of the elemental carbon fractions, modal emission rates, and modal
start rates are discussed in the next sections.
4.1.3 Estimating Elemental Carbon Fractions
After performing the analyses described above to estimate total particulate (PM2.5), we
partitioned the total into components representing elemental carbon (EC) and non-elemental
carbon (nonECPM), respectively. Following this step, the values for EC and nonECPM were
loaded into the emissionRateByAge table, using the pollutant and process codes shown in Table
2-1 (page 17). Non-elemental carbon particulate matter (NonECPM, or pollutantID 118),
represents particulate species other than elemental carbon. For light-duty exhaust, NonECPM is
primarily composed of organic carbon (pollutantID 112), and small amounts of inorganic ions
and elements. Background and further detail on the speciation of PM2.5 is discussed in greater
detail in the MOVES TOG and PM Speciation Report.19
The initial analysis of the EC composition of the light-duty exhaust is documented in the Kansas
City analysis report.51 In the Kansas City study, EC was measured using two different methods.
The first was the technique of thermal optical reflectance (TOR). This procedure also measured
OC and total PM, but unfortunately, not all the vehicles in the study were measured using this
technique. Elemental carbon was also measured using the photoacoustic analyzer, which
254
-------
measures EC on a continuous basis. More information can be found on these techniques and their
calibration and comparison results in the contractor's report68 and Fujita et al. (2006).69 The
former reference indicates that the photoacoustic analyzer has good correlation with TOR EC
measurement especially at higher PM levels, however, at lower levels (in bag 3 for example), the
correlation is poorer. This is not surprising since all instruments have limited ability to measure
small signals. To accentuate the full range of operation, Figure 4-14 shows a plot of a
comparison of the two instruments on a natural4og scale. The plot reinforces the excellent
agreement between the two instruments in bag 1 of the test, when emissions levels are at their
highest. The correlation (and slope) is also good for the high values in Bag 2, however, as the
measurements get smaller there is relatively more variability (in log-space) between the two
measurements.
~ bag 1
¦ bag 2
a bag 3
— Linear (bag 1)
— Linear (bag 2)
— Linear (bag 3)
y = 0.982J
R2 =
ix- 0.2107
). 9417
4
~
¦
~
~
y = 1 pfififii
- 1 1R9 ft
¦
R2 = 0.
7295
8
3
4 ^ ^
~ ^
- ~ ^JSj
yw
\ {
f
= 0.6468X
R2 = 0.
- (16774
485
m
~ /
A ¦ ¦ ¦
-ft
ln(TOR EC)
Figure 4-14 Comparison of photoacoustic to TOR EC measurements on a logarithmic scale
We explored the EC/PM fraction for the four measurement techniques employed in the Kansas-
City study: photoacoustic analyzer (PM, continuous EC), Dustrak analyzer (DT, continuous
optical PM), gravimetric filter (PM), and thermal optical reflectance (TOR, which measured both
EC and total carbon, TC). Table 4-3 shows the comparison of the 3 different fractions using
results from these instruments. The values were calculated as fractions of average values in the
numerator and denominator. The TOR fractions have two major limitations: the ratios are
unexpectedly high and, after eliminating bad data points, only 75 valid measurements remain.
Due to the latter condition (primarily), the TOR fractions will not be used in subsequent analysis.
The photoacoustic to Dustrak ratios present a reasonable approach, however, since the Dustrak
and PM are not strongly correlated50, we elected to use the photo-acoustic to gravimetric filter
ratios for EC/PM fraction estimation.
255
-------
Table 4-3 Elemental to total PM ratio for 4 different measurement techniques
Instruments
All
Start
Running
PA/DT
0.128
0.188
0.105
PA/PM
0.197
0.340
0.164
EC/TC
(TOR)
0.382
0.540
0.339
In MOVES, the EC/PM fractions for light-duty gasoline vehicles are consistent with detailed
PM2.5 speciation profiles developed for all the measured PM species in the Kansas City Study.70
The EC/PM fractions are estimated using the photoacoustic analyzer to filter-based PM
emissions. The MOVES speciation analysis confirmed our previous analysis51 that the EC/PM
fraction is relatively consistent across the range temperatures measured in Kansas City study, and
across the ranges of model years in the study. For this reason, no differentiation in the EC/PM
fraction is modeled in relation to temperature or model year of vehicles in MOVES.
In developing speciation profiles for light-duty gasoline vehicles from the KCVES,70 we
discovered high concentrations of silicon in the particulate matter samples. Upon further
investigation, we determined that the silicone rubber couplers used in the sampling system
probably contributed to the filter-measured mass. The resulting contamination of filter masses
with silicon substantially impacted the Bag 2 PM2.5 emission rates, which had the highest
exhaust temperatures. No significant contribution of silicon was found in the PM2.5 start
emissions. The adjustment to the MOVES running PM2.5 emission rates based on the silicon
measurements is discussed in Appendix A. Revisions to the Pre-2004 Model Year PM2.5
Emission Rates between MOVES2010b and MOVES2014.
The silicon contamination in these measurements resulted in a positive bias in the values for OC.
In consequence, the EC and nonECPM emission rates in MOVES were revised to account for the
updated data analyses used to derive the PM2.5 profile (e.g. VMT-weighted means), and to
compensate for the silicon contamination in the PM2.5 emission rates. Upon removal of the
silicon contamination, the EC/PM fractions are not significantly different between light-duty cars
and trucks. The data from cars and trucks were pooled as documented in the speciation
analysis.70 The EC/PM2.5 fractions in MOVES are presented in Table 4-4. The EC/PM2.5 ratio is
constant across all operating modes for start and running processes.
Table 4-4 EC/PM2.5 fractions by start and running emissions processes for pre-2004 light-duty gasoline
vehicles
Emission
Process
EC/PM2.5
Running
14.0%
Start
44.4%
256
-------
4.1.4 Modal Running Emission Rates
As mentioned in section 4.1.1, the Dustrak instalments was selected as the most reliable second-
by-second PM time-series data measurement from the Kansas City Study. The Dustrak PM2.5
measurements were used to develop the PM2.5 emission rates by operating mode. The following
two figures show Dustrak PM emissions binned by VSP and classified by model year Groups.
Figure 4-15 shows this relationship on a linear scale and Figure 4-16 shows the relationship on a
logarithmic scale. It is clear from the latter plot that VSP trends for PM tend to be exponential
with VSP load, i.e. they are approximately linear on a log scale, showing similar patterns to the
gaseous emissions, particularly CO. Thus, we assume smooth log-linear relations when
calibrating our VSP based emission rates.
257
-------
VSP, kw/tonne
Figure 4-15 Particulate emissions, as measured by the Dustrak, averaged by VSP and model-year group
(LINEAR SCALE)
Cars
20 25 30
VSP, kw/tonne
Figure 4-16 Particulate emissions, as measured by the Dustrak, averaged by VSP and model-year group
(LOGARITHMIC SCALE)
In order to calculate VSP-based modal rates, we followed seven steps:
1. The LA92 equivalent hot-running emission rate (g/mi) is calculated for each age group
within each model-year group, using the deterioration model described in section 4.1.2.3.
2. Continuous emission rates (g/sec) are calculated from the Dustrak measurements for cars
and trucks. These trends are then extrapolated to higher VSP levels where data is
missing.
258
-------
3. The VSP operating-mode distribution is calculated for Bag 2 of the LA92 drive cycle for
cars and trucks separately - this step is equivalent to determining the number of seconds
in each mode.
4. The set of continuous measurements (Step 2) are then classified into the operating-mode
distribution and summed to give an aggregate emission rate representing Bag 2 of the
LA92.
5. The results from Step 4 are divided by those from Step 1 to calculate a ratio for each
combination of the model-year and age groups. The ratios are used to normalize the
modal emission rates to the aggregate filter measurements.
6. The rates from step 5 are then apportioned into EC and nonEC components to give final
rates for the hot-running process. These rates are stored in the emissionRateByAge table
under polProcessID 11201 and 11801, respectively.
The output from step 3 (operating-mode distribution) for cars and light trucks is shown in Figure
4-17. For operating-mode definitions, see Table 2-5.
160
140
120 -¦
- 100 -¦
8> 80
C 60 -¦
40 -¦
20
~ car
~ light truck
m.
III
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
VSP Bin
Figure 4-17 Operating-mode distribution for cars and light trucks representing the hot-running phase (Bag
2) of the LA92 cycle
The output of step 5 for the ZML (0-3 year age Group) in each model year is shown in Figure
4-18.
259
-------
16
14
12
10
tr
LU
— — — — —
— — —
-M-
-St
¦ i *
(j * *
¦1960-1980
1981-1982
1983-1984
s 1985
>1986-1987
-1988-1989
-1990
1991-1993
1994
1995
1996
1997
1998
1999
2000
-2001
2002
2003
2004
10
15
20 25
VSP bin
30
35
40
45
Figure 4-18 Particulate emissions for passenger cars (LDV) from Kansas City results, by model year Group,
normalized to filter mass measurements
After the rates were calculated, a quality check was performed to ensure that the aged rates in
any particular mode were not too high. A multiplicative model with exponential factors risks
excessively high emission rates under extreme conditions. For example, any rate over 100 g/sec
was considered too high, this would be an extremely high-smoking vehicle. This behavior was
corrected in only two cases for cars and trucks in the 1975 model-year group in operating mode
bin 30. In these cases, the value from operating mode 29 was replicated for operating mode 30.
4.1.5 Modal Start Emission Rates
The development of the cold start emission rates (opMode 108; soak time > 12 hours), is
discussed in Section 4.1.2.2. The cold start emission rates (g/start), as estimated using Bagl -
Bag3 of the LA92, were estimated to be a factor of 1.972 times the reported LA92 composite
g/mile emission rate from the Kansas City study. This factor was then used to estimate cold start
emissions from the zero mile level emission rates. Subsequently, the impact of deterioration on
starts was incorporated as discussed in detail in Section 4.1.2.3.
In MOVES, the start rates by operating mode account for the different soak times preceding the
start as shown in Table 2-6. Section 3.9.1.1 discusses how the start emission rates for hot starts
(opModelD 101-107; soak times < 12 hours) are estimated as a fraction of the cold start emission
rates (opModelD 108). Due to limited data on PM emissions at different soak lengths, we apply
the same ratios between start operating modes for hydrocarbon start emissions as for PM
emissions presented in Table 3-51.
260
-------
4.2 Particulate-Matter Emission Rates for Model Year 2004 and Later
Vehicles
4.2.1 Introduction
This section addresses PM running emission rates for gasoline light-duty vehicles for model
years 2004 through 2060. Previously, MOVES2014 used the same PM emission rates for model
years 2003 through 2016 and then applied phase-in assumptions to account for Tier 3 standards.
This section, therefore, represents an update to the MOVES emission rates for vehicles subject to
Tier 2 and Tier 3 standards. Since 2004, gasoline direct injection (GDI) vehicles have entered the
market. In 2016, GDI vehicles represented roughly half of new vehicles sold in the United
States.71 Additionally, several studies of vehicle emissions have been conducted since the Kansas
City study50 using vehicles newer than MY 2004 vehicles. The emission rates derived in this
section are based on the data from six such studies, including studies of GDI vehicles. The
adoption of GDI engines has been taken into account by separately calculating PM emission
rates for PFI (port fuel injection) and GDI vehicles, and then combining them to form
population-weighted average rates by model year. However, the datasets used in these analyses
do not contain enough information to derive completely new modal emission rates or
deterioration rates for these model years. Therefore, to determine the new modal rates, we
rescaled the existing modal rates used for model year 2003 in MOVES using the new data, and
retained the deterioration behavior described in Section 4.1.2.3. Finally, we applied the phase-in
of Tier 3 standards to the newly derived rates.
4.2.1.1 Dataset Description
Data from six studies was used to develop the 2004 and later PM emission rates. The dataset for
each study includes PM filter weight measurements collected on FTP or LA-92 three-phase or
"bag" test cycles. Phase 1 (bag 1) is a cold start where the vehicle has been "soaking" at a
controlled temperature for 12 or more hours with the engine turned off. Typically, vehicles are
soaked at room temperature (~72°F). Phase 2 follows Phase 1 and is used to characterize
temperature-stabilized or "hot running" conditions. At the end of Phase 2, the engine is shut off,
and the vehicle is allowed to soak for 10 minutes under the ambient test cell conditions. Finally,
the engine is restarted and Phase 3 follows the same driving cycles as Phase 1. For the LA92
cycle, Phases 1 and 3 last for 310 seconds, and Phase 2 lasts for 1,135 seconds. Phases 1 and 3 of
the FTP cycle are longer than for the LA92, taking 505 seconds. Phase 2 of the FTP cycle is
shorter at 867 seconds. PM filters were collected and weighed for each phase of the test cycles
providing a measure of the total PM mass emitted during each phase. The studies selected for
analysis are summarized in Table 4-5.
261
-------
Table 4-5 Summary of PM studies analyzed for model year 2 004 and later vehicles
Study name
Engine
Type
Number of
vehicle models
Number of unique
vehicles
EPA Tier 2 Fuel Sulfur Study72
PFI
17
72
EPAct Phase 1 FTP73
PFI
6
6
EPAct Phase 374
PFI
15
15
EPAct Phase 475
PFI
6
6
CARB LEV III PM Emissions Study76
GDI
6
6
EPA Tier 3 Certification Fuel Impacts Study77
GDI
7
8
Altogether, the dataset for PFI vehicles consists of measurements from 99 vehicles representing
19 different models. Unlike the KCVES, these studies were designed to capture properly
functioning vehicles. We assume that the vehicles in the study represent age zero emission rates
in MOVES, with no effects of emissions deterioration due to age. The dataset for GDI vehicles is
composed of measurements from 14 vehicles, and 13 models. Because of the limited number of
GDI vehicles, there was not enough data for both wall-guided and spray-guided injection
architectures to differentiate between them for this study. Only the tests conducted at room
temperature were included in this analysis in order to eliminate influences from hot or cold
temperature tests. Measurements conducted with greater than 20 percent ethanol fuels were
omitted from analysis because MOVES only handles fuel with ethanol content less than or equal
to 15 percent for gasoline vehicles.
4.2.1.2 Fuel Considerations
The four studies used to generate PM emission rates for PFI vehicles used a combined total of 27
different fuels with ethanol content less than 20 percent. In order to minimize the effects of these
fuels on the emission rate calculations, the measured rates were corrected to the equivalent rates
for Tier 2 certification fuel. The corrected rates were calculated using the EPAct fuel effects
calculator, which uses the same method used by MOVES to calculate fuel-effect adjustments.22
The EPAct calculator applies the set of statistical models developed using the EPAct Phase 3
dataset, also used for developing the particulate matter emission rates in the current analysis.
Additionally, the EPA Tier 2 sulfur study used Tier 2 based fuels and as such required negligible
correction.72 The corrections were applied to all three phases of the FTP and LA92 PM mass
measurements. The effects on the distribution of measured start and running emissions for each
test program are summarized in Figure 4-19.
262
-------
50-
-= 25-
O)
W 0-
u"
tz
E 301
(a)
•
•
•
|-
<5"
•
(—*¦
D
c
JP
M
Ej3 Fuel Corrected
Ej3 Raw Data
Figure 4-19 Boxplots of start (a) and running (b) emissions measurements with and without fuel corrections
applied
263
-------
4.2.2 Calculating FTP and LA92 Cycle Rates Using MOVES Emission Rates
The six datasets used for this analysis are not adequate to develop revised running modal
emission rates de novo for vehicles with model years 2004 and later. Therefore, the modal rates
for model year 2003 vehicles are rescaled to generate the emission rates for 2004 and later model
years. In order to develop the appropriate rescaling factors, Bag 2 emission rates are calculated
for both the FTP and LA92 drive cycles using MOVES model year 2003 emission rates.
The Bag 2 rates of both the FTP and LA92 cycles for both MOVES light-duty regulatory classes
(light-duty cars, and trucks) are calculated using the MOVES operating mode distribution
calculated for the hot running phase of each test cycle, and multiplying the time in each
operating mode with its associated emission rate. To generate an emission rate, the emission
masses calculated for each operating mode are summed, and the total is divided by the distance
driven. The MOVES operating mode distribution for Bag 2 of both the FTP and the LA92 cycles
are shown in Figure 4-20.
HI
-------
measured rates in the datasets that are analyzed in Sections 4.2.3 and 4.2.4. Additionally, these
calculated cycle rates are used in Section 4.2.5 to determine the rescale factors used to develop
the model year 2004 and later PM emission rates used in MOVES.
Table 4-6 Modeled FTP and LA92 start and bag 2 running rates for model year 2003 light-duty vehicles
Test cycle
regClassID
Cold-start mass
(mg)
Hot-running rate
(mg/mi)
FTP
LDT
8.781
1.444
FTP
LDV
6.158
2.090
LA92
LDT
8.781
2.133
LA92
LDV
6.158
1.924
4.2.3 Estimating Start Emissions for Particulate Matter
Start emissions from three-phase test cycles are calculated by comparing the measured masses of
the Phase 1 and Phase 3 PM filters. For both the LA92 and FTP drive cycles, the speed trace for
Phases 1 and 3 are identical. The difference in measured PM masses between the two phases is
attributed to the change in engine condition from cold start to hot stabilized running. Typically,
this transition results in higher Phase 1 PM mass. If the value of the Phase 1 minus the Phase 3
mass is negative, it suggests that the hot stabilized engine emitted more particulate matter than it
did when it was warming up. We observed this behavior in some of the test results. Because we
found no technical reason to exclude these points, they are included in the averaged rates. For
this analysis, we assume that cold-start PM emissions are independent of the test cycle. The
average rates from the data discussed in this section are used in Section 4.2.5 to develop the
scaling factors for constructing the PM start rates.
4.2.3.1 Start Emissions for Vehicles with Port Fuel Injection (PFI)
Figure 4-21 summarizes the cold-start results from the PFI vehicles used in this analysis, which
are drawn primarily from the EPAct Phase-3 study. The solid horizontal lines show the average
cold-start mass for light-duty cars and trucks, as averaged by vehicle model. The dashed
horizontal line shows the cold start mass for new vehicles with model year 2003 in MOVES. For
PFI light-duty cars, the average cold start mass is 2.06 mg and for PFI light-duty trucks, it is 3.75
mg. On average, the measured PM cold start emission masses for the analyzed data were
substantially lower than modeled for model year 2003 vehicles in MOVES.
265
-------
Light Duty Cars
t •
• •
//^///^/
°x#
Model
— Dataset Average
MOVES
(Model Year 2003)
Figure 4-21 Measured PFI PM start emission masses
4.2.3.2 Start Emissions for Vehicles with Gasoline Direct Injection (GDI)
Figure 4-22 summarizes the cold-start results from all of the GDI vehicles used in this analysis.
The solid horizontal lines show the cold-start mass for light-duty cars and trucks, as averaged by
each unique vehicle. The dashed horizontal line shows the cold start mass for new vehicles with
model year 2003 in MOVES. For GDI light-duty cars, the average cold start mass is 20.92 mg.
While only data from two GDI trucks is available in these studies, the average cold start mass for
these two vehicles is 38.34 mg. Generally, the measured PM start emission masses for GDI
vehicles in the analyzed dataset were significantly higher than modeled for model year 2003
vehicles in MOVES.
266
-------
Light Duty Trucks
g 40-
CL
t
(0
OT 20'
2
O
O
ffl
Light Duty Cars
~
a
11111
/$? ,3r
TAt
/
a '
&
Model (Model Year)
Dataset Average
MOVES
(Model Year 2003)
Figure 4-22 Measured GDI PM start emissions
4.2.4 Estimating Running Emissions for Particulate Matter (PM)
Running emission rates were calculated for each test in units of milligrams per mile. Because the
FTP and LA92 cycles cover different engine power ranges as shown in Figure 4-20, the average
emission rate for each vehicle model was calculated separately for each test cycle. In general, the
results for both PFI and GDI vehicles show substantially lower running PM rates than modeled
for model year 2003 in MOVES. The average rates from the data discussed in this section are
used in Section 4.2.5 to develop rescale factors for constructing the MOVES PM running rates.
4.2.4.1 Running Emissions for Vehicles with Port Fuel Injection (PFI)
For the four test programs used in the PFI analysis (Table 4-5), the running PM rates are grouped
by vehicle model. Figure 4-23 summarizes the results. The solid horizontal lines show the
average Phase 2 running mass for light-duty cars and trucks, as averaged by vehicle model. The
dashed horizontal line shows the Phase 2 running mass for new vehicles with model year 2003 in
MOVES. As Figure 4-20 demonstrates, the LA92 drive cycle has a more aggressive Phase 2 than
the FTP cycle. This difference results in a higher average emission rate for the LA92 cycle than
for the FTP cycle. This difference is reflected in both the measured datasets and the cycle
average rates calculated by combining model year 2003 emission rates and operating-mode
distributions for the two cycles.
267
-------
5-
4-
3-
2-
1
OH
5-
4-
3-
2-
1-
O-
Light Duty Trucks
•
•
—
—
ii
¦ -<
.
r -
1
I
f+WX1
r
°4r ° ° &
Light Duty Cars
•
n
T3
Dataset Average
MOVES
(Model Year 2003)
O
o ^ o
Model
Figure 4-23 Measured PFI PM running emission rates
4.2.4.2 Running Emissions for Vehicles with Gasoline Direct Injection (GDI)
The summary of running emission rate results for the GDI vehicles used in this analysis are
shown in Figure 4-24. Because the GDI vehicles were tested only using the FTP drive cycle, the
results are not split by test procedure. As with the GDI start emissions, the averages rates are
calculated weighted by test vehicle. While there is significant variation in the PM rates for the
GDI vehicles, the average running emission rates fall below the model year 2003 MOVES
average.
268
-------
5-
4-
E
O)
E_
-------
Table 4-7 Cold-start and hot-running scaling factors for PFI and GDI vehicles
Engine type
regClassID
Cold-start
scaling factor
Hot-running
scaling factor
PFI
LDT
0.427
0.382
PFI
LDV
0.335
0.260
GDI
LDT
4.367a
0.3123
GDI
LDV
3.398
0.515
Note:a See Section 4.2.5.lfor the final scaling factors for GDI LDT.
4.2.5.1 Additional Assumptions Used to Determine GDI Truck Scaling Factors
The data for the two GDI trucks included in the six datasets is not sufficient to form the basis for
revised emission rates in MOVES3. To compensate, we developed an approximation of start and
running emission rates for GDI trucks using the data analyzed for PFI vehicles, and for the GDI
light-duty cars. We assume that the apparent difference in PM emissions between GDI and PFI
vehicles are due to the change in injection technology. Additionally, we assume that the change
in injection technology will have a similar proportional emissions effect on engines in light-duty
trucks as in light-duty cars. To calculate GDI truck start emissions, we use the following
equation:
StartLDV(GDI)
StartLDT(GDI) = StartLDT(PFI) Equation 4-3
LDTy J LDTy J StartLDV(PFI)
where LDV indicates light-duty cars, and LDT indicates light-duty trucks.
For running emissions, we used a slightly different approach. Because the datasets only contain
results for GDI vehicles on the FTP cycle, it was difficult to directly compare them to the PFI
results where a significant proportion were measured on the LA92 test cycle. Therefore, we
made the assumption that the scaling of the 2003 model year MOVES rates for GDI light-duty
trucks would be the same as the scaling for light-duty cars, i.e.:
Runninq,nv(GDI)
Running ldt(GD I) = RunningLDT (MOVES)RunjUngu>vCM0VES) "quado, 4-4
Table 4-8 contains the calculated start and running rescale factors using these assumptions as
well as the average measured values from the two trucks in the studies. For start emissions, the
rates calculated from these assumptions are very similar to the measured rates from the two
trucks. The calculated running rates on the other hand show a more modest reduction relative to
the 2003 model year rate than suggested by the test results from the two trucks. The rescale
factors derived from these assumptions are the ones used to derive the final MOVES3 light-duty
truck emission rates.
270
-------
Table 4-8 Scaling factors for light-duty trucks calculated from measured data and from modeling assumptions
Cold-start
Hot-running
Unadjusted scaling factor (Table 4-7)
4.367
0.312
scaling factor calculated from Equation
4-3 and Equation 4-4
4.330
0.515
4.2.5.2 EC/NonECPM Fractions
In the MOVES EmissionRateByAge table, total PM emission rates are partitioned into elemental
carbon (EC) and non-elemental carbon (nonECPM). Section 4.1.3 describes the method for using
photo-acoustic to gravimetric filter mass ratio to determine the fraction of EC to total PM.
Because the datasets used for PFI vehicles did not have additional EC information, we retain the
EC/PM2.5 fractions calculated from the Kansas City study to represent light-duty PFI vehicles
with model years 2004 and later. The CARB LEVIII PM study used as part of the GDI rates
analysis, also included photo-acoustic PM mass measurements. As such, we used the same
method to calculate EC/PM2.5 fractions for light-duty GDI vehicles. The resulting fractions show
a significantly higher EC fraction for both start and running emissions from GDI vehicles as
compared to PFI vehicles. The start and running EC/PM2.5 fractions for both PFI and GDI
vehicles are summarized in Table 4-9.
Table 4-9 Start and running EC/PM2.5 fractions for PFI and GDI vehicles
Engine type
Start EC/PM2.5
Running EC/PM2.5
PFI
0.44
0.14
GDI
0.70
0.67
4.2.6 Calculation of Fleet-Average PM Emission Rates by Model Year, Vehicle
Age, and PM component
This section describes how the cold-start and hot-running rescale factors and the EC/PM2.5
fraction determined in Section 4.2.5 are combined to create the PM emission factors used in
MOVES for model years 2004 and later. Here, the emission rates are derived without accounting
for the implementation of new emission standards. Sections 4.2.7 and 4.2.8, describe how the
Tier 3 and LEV-III standards are applied to the PM emission rates.
Thus far, the discussion of PM rates for light-duty vehicles for model years 2004 and later has
divided these vehicles by fuel injection technology; however, MOVES does not currently
accommodate partitioning emission rates for a given regClass by engine technology. Rather,
fleet-average rates must be entered into the emissionRateByAge table. Therefore, average PM
emission rates were calculated for each model year using weights for the PFI and GDI emission
factors determined from vehicle production volumes.
271
-------
4.2.6.1 Vehicle Population Data for Model Years 2004 and Later
For model years 2004 through 2020, the annual EPA Automotive Trends Report provides data on
the relative production volumes of vehicles with different engine technologies.78 The report's
associated interactive data browser provides the proportions of the light-duty car and truck
populations that have PFI and GDI engines. For model years 2021 and later, the EPA CCEMS
Post Processing Tool was applied to data from runs of the CAFE Compliance and Effects
Modeling System (CCEMS, or CAFE model) to extract modeled future population fractions of
GDI and PFI vehicles for both light-duty cars and trucks.79 These data were used in the Revised
2023 and Later Model Year Light-Duty Vehicle Greenhouse Gas Emissions Standards
Rulemaking.80 The combined historical data, and modeled future populations are illustrated in
Figure 4-25 represented by solid and dashed lines respectively. These proportions were used
directly to weight the fleet-average PM emission rates from PFI and GDI vehicles.
1.00'
¦2 0.75 ¦
u
re
J 0.50 ¦
re
3
o" 0.251
0.
0.00'
nht niitv Trurk*;
\
\
/
Vj
n
-S
J
inht Huh/ r;ar
-------
RSFleet(MY) = SGDI * Pgdi(MY) + SPFI * Ppfi(MY)
Equation 4-5
Where S is the scaling factor for the fleet of the given engine type, and P is the population
fraction of PFI or GDI engines for each model year (MY).
Next, the EC/PM2.5 fractions for each model year were calculated as a population and emission
rate weighted sum of the EC/PM2.5 fractions for PFI and GDI vehicles using the following
equation:
EC/PM2 5 Fleet
EC/PM2.5GDI(PGDI*SGDI])
(Pgdi * $gdi) + (PpFI * SPFI) Equation 4-6
_l_ EC/PM2.S PFliPpFI * SpFl)
(PgDI * $GDI) + (PpFI * SpFI)
Where EC/PM2.5 is the EC fraction, P is the population fraction. The subscripts indicate the
values associated with the combined fleet, and for GDI and PFI vehicles. The EC/PM2.5 values
are used to estimate emission rates are portioned into two PM components (EC and nonECPM)
as discussed in Section 4.2.5.2. Finally, the scale factors and new EC/PM2.5 fractions were
applied to the start and running modal emission rates from MOVES model year 2003 light-duty
cars and light-duty trucks to generate a complete set of revised EC and nonECPM emission rates
in MOVES3 for model year 2004 through 2060. This method thus preserves the modal rate
structure as well as the deterioration effects modeled for earlier model years. Figure 4-26 through
Figure 4-28 illustrate how these emission rates change with model year. Note that these rates do
not yet account for the phase-in of the Tier 3 standards, which is handled in Section 4.2.7.
Figure 4-26 shows how the PM cold start mass for light-duty cars and trucks changes with model
year, showing increases in both EC and nonECPM as the percentage of GDI vehicles increases.
273
-------
IfflSTIiTTTTK
rnrkg
inht Huh/ fiarg
Figure
2010 2020 2030 2010
Model Year
2020
2030
4-26 Modeled cold start PM emissions by model year for age 0 vehicles- not adjusted for phase-in of
Tier 3 standards
¦ - EC
- NonECPM
— Total PM
Figure 4-27 shows calculated FTP Bag 2 running rates to illustrate how the MOVES rates for
light-duty cars and trucks change with model year. For these rates, the nonECPM portion of the
emissions decrease with GDI phase in while the EC portion increases. Together, the changes in
EC and nonECPM rates result in a net increase in Total PM with increasing model year.
1.00-
£
o>
E
-------
¦ - EC
- NonECPM
— Total PM
inht Dnh/
rnrkc
I inht Dnt\/ Ciarg
2010 2020 2030 2010 2020 2030
Model Year
Figure 4-28 Modeled FTP cycle average PM emissions by model year for age 0 vehicles - not adjusted for
phase-in of Tier 3 standards
4.2.7 Incorporating Tier 3 Emissions Standards for Particulate Emissions
Under the Tier 3 exhaust emissions standards, finalized in April, 2014, the FTP standard for
particulate emissions was reduced from its level under the Tier 2 standard (10.0 mg/mi) to a new
value of 3.0 mg/mi.81
Developing rates to represent particulate emissions from gasoline-fueled vehicles under the Tier
3 standards involved scaling down rates representing vehicles under the Tier 2 standard to a level
that assumes a reasonable compliance margin with respect to the lower standard. More
specifically, we assumed that average FTP emissions for new light-duty vehicles (age 0-3 years)
would be 1.5 mg/mi in MY 2025, corresponding to a compliance margin of 50 percent, when the
new standard was fully phased in. This assumption is independent of engine and fuel-injection
technology. The reduced rates assume that additional controls are needed to meet the new
standard for vehicles employing gasoline direct-injection technologies, but not for the declining
fraction of vehicles in the market employing port-fuel-injection. The analysis above shows that
new PFI vehicles start at about this level, and thus can virtually meet the new standard without
modification.
Additionally, as with the gaseous emissions, the regulatory useful life was increased from
120,000 to 150,000 miles. The concomitant assumption of increased durability was expressed
through a reduction in the assumed deterioration rate.
We applied these modifications to the MOVES EmissionRateByAge table in a series of three
steps.
4.2.7.1 Apply Phase-in Assumptions
The first step was to apply the phase-in assumptions applicable to PM. The phase-in begins with
model year 2017 and ends with model year 2021 for cars (LDV) and trucks (LDT). Fractions of
275
-------
new vehicles meeting the new standard during the phase-in are shown in Table 4-10. The table
also shows simulated FTP composites during the phase-in. These projections were simply
calculated as averages of the Tier 2 and Tier 3 baselines, with the phase-in fractions used as
weights. Figure 4-29 shows how the simulated Tier 3 FTP composite rates compare against the
base rates derived in Section 0, and to the rates used in MOVES2014.
Table 4-10 Phase-in Fractions and simulated FTP composites projected for the introduction of the Tier 3
Model year
Fraction
meeting Tier 3
standard
Simulated FTP
composite (mg/mi)
Cars (LDV)
Trucks (LDT)
2016
0.0
1.56
2.03
2017
0.10
1.78
2.28
2018
0.20
1.86
2.39
2019
0.40
1.84
2.30
2020
0.70
1.70
1.95
2021+
1.00
1.50
1.50
2010
2020
Model Year
2030
Trucks (LDT)
Cars (LDV)
Base Rate
MOVES2014
MOVES4
Figure 4-29 Simulated FTP composite rates for Tier 2 base line and Tier 3 phase-in. Base Rate represents age
zero emissions prior to adjustment for phase-in of Tier 3 standards.
4.2.7.2 Apply Scaling Fractions
The second step was to apply the fractions to the emission rates for running and start emissions
in the EC and nonECPM pollutant processes (11201, 11202, 11801, 11802). The fractions were
applied uniformly to rates in all operating modes, for both cars and trucks.
276
-------
Figure 4-30 shows an example of scaling, for a subset of non-elemental-carbon (nonECPM,
11801) rates for three model years, 2016, 2019 and 2021. Model year 2016 represents Tier 2
standards prior to the onset of the phase-in, 2021 shows fully phased-in Tier 3 standards, and
2019 shows an intermediate year during the phase-in period. In (a), the rates are shown on a
linear scale to show the steepness and non-linearity of the trends against power, whereas in (b),
rates are shown on a logarithmic scale to make clear that the multiplicative scaling is uniform
across the power range. Although not pictured, note that rates for elemental-carbon (ECPM,
11201) show an identical scaling pattern. Note also, that for convenience, emissions in the plot
are presented in mg/hr, whereas rates in the emissionRateByAge table are provided in g/hr.
The uniformity of the multiplicative scaling is also clear if the rates for a single model year are
viewed against age for a set of operating modes, as shown in Figure 4-31. The plot shows rates
for six modes of running operation, including idle (mode 1), with the remaining five modes
spanning a range from low to moderate power. As previously described in 4.1.2.3, the
deterioration trends are exponential (or log-linear).
277
-------
o>
£ 200'
d>
re
DC
-------
90'
U)
E
0)
| 60'
at
V)
re
CQ
§ 30'
(a) Linea
r Scale
* T* T
10 15
Vehicle Age (Years)
20
opModelD
* 1
^ 13
-±- 21
24
27
35
10 15
Vehicle Age (Years)
Figure 4-31 Non-elemental-carbon rates for trucks vs. Age for selected running operating modes in model
year 2016, presented on (a) linear and (b) logarithmic scales
4.2.7.3 Simulate the Extended Useful Life
The third and final step was to reduce deterioration for vehicles under Tier 3, relative to those for
Tier 2. The deterioration trends were scaled down such that the fleet is 1.25 times as old when a
given emissions level is reached under the extended useful life as under the original useful life.
The value of the fraction, 1.25, was calculated as 150,000 mi/120,000 mi, or 15/12.
The reduction in the deterioration trend is illustrated in Figure 4-32, which shows age trends for
cold-start non-elemental-carbon before and during the phase-in period. The upper pane (a)
shows the moderation of the exponential trend, whereas the lower pane (b) shows the reduction
in the logarithmic slope starting in model year 2017. As before, these rates are presented in
mg/start, as opposed to g/start in the database table. Note again that a similar chart for elemental
carbon would show an identical pattern.
279
-------
(a) Ltrtes
r Scale
yl
X *
* ^
W
10 15
Vehicle Age (Years)
20
Model Year
2016
-« 2017
— 2018
-• 2019
=« 2020
** 2021
(b) Log i
scale
***
>*
* *
Model Year
2016
-• 2017
=® 2018
-® 2019
-® 2020
=® 2021
5 10 15 20
Vehicle Age (Years)
Figure 4-32 Elemental-carbon rates for cars vs. Age for cold-start emissions in six model years, presented on
(a) linear, and (b) logarithmic scales
280
-------
4.2.8 Incorporating the LEV-III Standard for Particulate Matter
The Tier 3 and LEV-III standards are harmonized with respect to the light-duty standard for
particulate matter through MY 2024, at which point, a 3.0 mg/mi FTP standard will be fully
phased in. However, after MY 2025, the LEV-III program goes further, enacting a further
phased-in reduction to a 1.0 mg/mi FTP standard. This reduction is incorporated into the
emissionRateByAgeLEV table applicable to California and Section 177 states.
The assumptions used to express the transition from rates at the 3.0 mg/mi level to the 1.0 mg/mi
level are shown in Table 4-11. We assume a linear phase-in over the three years. The
calculations assume a 50 percent compliance margin with respect to the 3.0 mg/mi standard in
MY 2024, transitioning to a 25 percent compliance margin in MY 2028.
These assumptions were modeled in MOVES by applying the reduction fractions shown in the
right-most column in Table 4-11 to default MOVES rates for the LEV-III phase-in model years.
These fractions were applied uniformly to start and running emissions of EC and nonECPM, for
cars and trucks, across all operating modes.
The emissionRateByAgeLEV table including these rates is incorporated into the default MOVES
database. Instructions for use of the applicable portions of this table in a MOVES run are
available at https://www.epa.gov/moves/tools-develop-or-convert-moves-inputs. Section 3.12
details how the emission rates representing California standards were developed for criteria
pollutants.
Table 4-11 Phase-in assumptions and reduction fractions used to represent a transition to the 1.0 mg/mi PM
standard under LEV-III
Model year
Phase-in fraction
FTP composite
(mg/mi)
Reduction fraction1
At 3.0 mg/mi
At 1.0 mg/mi
2024
1.00
0.00
1.50
1.000
2025
0.75
0.25
1.31
0.873
2026
0.50
0.50
1.13
0.753
2027
0.25
0.75
0.94
0.627
2028+
0.00
1.00
0.75
0.500
1 Applied to default rates in listed model years.
4.3 Light-Duty PM Emission Rates Trends
The following graphs show trends in MOVES light-duty PM emission rates by model year.
281
-------
0 100
0.075
a>
E
O)
g
CC
a.
m
.0
0.050
0.025-
0.000-
Reg Class
10-MC
20-LDV
-*• 30-LDT
1980
2000
2020
2040
Model Year
Figure 4-33 Base PM2.5 running emission rates for age 0-3 gasoline motorcycles, light-duty vehicles, and
light-duty trucks averaged using nationally representative operating mode distributions
0.006-
0.004
a
tc
c
o
crt
E
LU
a 0.002
o.ooo
EC
NonEC
1990
1995
2000
Model Year
2005
2010
Figure 4-34 EC and NonEC PM2.5 emission rates for age 0-3 passenger cars averaged across a nationally
representative operating mode distribution
282
-------
As illustrated in Figure 4-33 and shown more clearly in Figure 4-34 the MOVES PM emission
rates for MY 2004 and later vehicles described in Section 4.2 are significantly lower than those
originally developed for MY 2003 vehicles as discussed in Section 4.1. There are several
differences in the vehicle samples, measurement methods, and data analysis methods that are
likely contributing to this difference in PM emission rates as described below:
• Vehicle samples: The most recent studies (KCVES, MSAT, and Li et al., 200666)
considered for the pre-2004 emission rates included MY vehicles between 2002-2005.
The studies used in the MY 2004 and later emission rate update included later model year
vehicles between (2007 and 2014). The decrease in PM emissions could be partially
attributed to lower PM emission rates from the newer technology vehicles.
• Measurement methods: Particulate matter emissions measurements were not conducted
with consistent methods across the studies. Uncorrected sampling artifacts could be the
cause of the large differences between the pre-2004 and the 2004+ PM emission rates. As
documented in Appendix A of this report, we corrected for a sampling issue in the
KCVES that would have caused the PM emission rates to be significantly overestimated.
Additionally, several years had passed from the last study used in to derive the pre-2004
rates (2006) and the earlier study conducted for the MY 2004+ rates (2013). In this time
there were significant improvements in particulate sampling methods, including filter
handling and filter weighing techniques. These differences in particulate matter sampling
methods could be the cause for much of the differences observed between the pre-2004
and the 2004+ model year rates.
• Data analysis methods: Different data analysis methods were used to estimate the zero-
mile emission rates for the two model year ranges. For example, we fit an exponential
curve to age 0-3 vehicles from 15 different studies (including both FTP and LA-92
cycles) by model year to estimate the pre-2004 zero-mile emission rates. For the MY
2004+ rate update, we assumed that the measured vehicles had not experienced
deterioration and simply averaged all the measured data according to sample size to
represent the zero-mile emission rates. In addition, we accounted for differences in the
MOVES operating modes between the LA-92 and FTP cycle for the recent update. These
different data analysis methods could contribute to the observed differences.
We have confidence in the more recent PM emission rates because they are based on more recent
studies and updated sampling procedures. Additionally, the data analysis methods for the most
recent rates are more straightforward than the analysis conducted for the pre-2004 MY rates.
Despite our higher confidence in the more recent PM rates, we have decided to leave the pre-
2004 MY PM rates unchanged in MOVES for these three reasons:
• Some of the differences in the pre-2004 and 2004+ emission rates may be due to the
actual differences in engine and aftertreatment differences in MY vehicles
• In a calendar year 2018 MOVES run using a draft version of MOVES3, the pre-2004
model year vehicles contribute just over 50% of PM2.5 emissionsk from all light-duty
vehicles (regulatory class LDV and LDT). In current and future years, the contribution of
k From a draft MOVES run conducted at national aggregation, using January and July to represent the entire year,
pre-2004 model years contributed 51.5% of PM2.5 exhaust emissions from regulatory class LDV and LDT vehicles.
283
-------
these older model year vehicles to the overall inventory will decrease, and no longer be
the majority of emissions from light-duty vehicles.
• Revisiting the pre-2004 model years emission rates would be a substantial effort. As
documented in this report, the pre-2004 were based on an analysis of many different
studies which measured PM emissions. The analysis of these different studies provided
data to estimate light-duty deterioration, which continues to serve the basis of the modal
VSP-trends, EC/PM ratios for PFI vehicles, and the deterioration of light-duty PM
deterioration for all model year vehicles. Additional scientific evidence is likely needed
for us to revisiting the emission rates of these older model year vehicles, which continue
to be used as a basis for the emission rates for the 2004+ model year emission rates.
5 Gaseous and Particulate Emissions from Light-Duty Diesel and
Electric Vehicles (THC, CO, NO*, PM)
This section explains the gaseous and particulate emissions from light-duty diesel vehicles and
provides some important notes on how MOVES models light-duty electric and hybrid vehicles.
Table 5-1 Fuel types and engine technologies represented for gaseous-pollutant emissions from light-duty
vehicles
Attribute
sourceBin attribute
Value
Description
Fuel type
fuelTypelD
01
Gasoline
02
Diesel
05
Ethanol (E77, E85, etc.)
Engine Technology
engTechID
01
Conventional internal combustion (CIC)
30
Electric
5.1 Light Duty Diesel
In MOVES, emission rates are calculated for each operating mode. However, for the diesel-
fueled passenger cars (LDV) and light-duty trucks (LDT), we lack the necessary continuous or
"second-by-second" measurements to directly calculate emission rates for running emissions in
relation to vehicle-specific power.
Upon additional review, we concluded that the diesel rates developed for draft MOVES and
retained in MOVES2010 were not plausible in relation to corresponding rates for gasoline
vehicles. We concluded that these rates were not adequate to retain in MOVES2014. However,
we also did not consider it a tenable option to release MOVES2014 without rates representing
diesel vehicles.
284
-------
Consequently, we decided to allow rates for light-duty gasoline vehicles to represent those for
light-duty diesel vehicles. While not an exact parallel and not desirable from a technical
standpoint, we considered it an acceptable solution, as vehicles running on both fuels would be
certified to similar standards. Also, as there are very few light-duty diesel vehicles in the U.S.
fleet, their contribution to the inventory is very small.
However, in contrast to the gasoline rates, we did not incorporate a difference in the base rates
attributable to Inspection and Maintenance. That is to say, values for meanBaseRate (non-I/M
condition) were substituted for both the meanBaseRate and meanBaseRatelM. Note, however,
that for rates representing diesel emissions, the model does not apply the fuel adjustments
applied to gasoline emissions.22
For MOVES3, we used the same approach as in MOVES2014, taking the light-duty gasoline
values for meanBaseRate and using them to populate both the meanBaseRate and
meanBaseRatelM values for light-duty diesel.
The level of detail for the rate substitution is shown in Table 5-2.
Table 5-2 Level of detail for substitution of light-duty gasoline Rates onto light-duty diesel rates
Parameter
Description
Identifier
Pollutant
THC
1
CO
2
NO,
3
EC-PM
112
NonECPM
118
Process
Running Exhaust
1
Start Exhaust
2
Regulatory Class
Passenger Car (LDV)
20
Light Truck (LDT)
30
Model-year Group
All
1960-2031
Data Source
Replicated from corresponding
4910
Rates for light-duty gasoline
5.2 Light Duty Electric Vehicles
Starting with MOVES4, electric vehicles are included in MOVES default vehicle populations.
While electric vehicles are associated with upstream and life-cycle emissions that are not
modelled by MOVES, and with energy consumption1 and brake and tire wear emissions82
described in other MOVES reports, they do not generate direct exhaust emissions. Thus,
emissions of THC, CO, NO*, NH3 and exhaust PM are modelled as "zero" in MOVES.
EPA is aware that manufacturers can include electric vehicles and hybrid electric vehicles in
their computation of average emissions for compliance with Tier 3 standards. Thus, if a
285
-------
manufacturer sells a large number of zero or low-emitting vehicles, the manufacturer would be
allowed to increase the average emissions of other vehicles. In the case of hybrid vehicles,
MOVES accounts for this by not modelling hybrids explicitly—instead, their emissions are
combined with all other vehicles in fleet averages.
MOVES takes a different approach for electric vehicles which are considered a different fuel
type. And, unlike MOVES3, MOVES4 includes electric vehicles in the default fleet.2 MOVES4
also accounts for the projected associated increases in emissions from conventional light-duty
vehicles allowed by the Tier 3 regulations. That is, THC and NO* emissions from conventional
(i.e. gasoline, diesel and E85) vehicles are adjusted to account for a less stringent "effective"
conventional vehicle standard that accounts for averaging with electric vehicles. These
adjustments are explained in the MOVES Adjustments Report.3
6 Ammonia Emissions from Light-duty Vehicles
6.1 Light Duty Gasoline
Light-duty spark-ignition vehicles are important sources of ammonia (NH3) emissions in urban
areas.83 NH3 is formed from the catalytic reduction of nitrogen oxide (NO) in the three-way
catalyst. The NO reacts with hydrogen as shown in the following reactions.
2NO + 5H2 -> 2NH3 + 2H20
2NO + 2 CO + 3 H2 -> 2NH3 + 2 C02
During slightly fuel-rich conditions, both nitrogen oxide and hydrogen are present in the exhaust
stream. Hydrogen gas is formed in the engine or in the three-way catalyst from the reaction of
carbon monoxide or hydrocarbons with water, as documented in Easter et al. (2016).84
MOVES only estimates light-duty gasoline NH3 emissions from the running emission process
when the three-way catalytic converter is active and can reduce NO to NH3. Researchers have
also measured elevated NH3 emissions from cold starts (but after catalyst light-off), however,
these data have not been incorporated into MOVES.85-86 87
The ammonia emission rates for MOVES3 and earlier versions were developed for MOVES2010
from test data from 2001 and earlier model year vehicles as documented in a MOVES2010
technical report.88 These rates continue to be used for 1960-1980 vehicles in MOVES.
Two studies suggested that the mobile ammonia emission rates developed for MOVES2010
underestimate light-duty gasoline ammonia emissions for recent calendar years.83 89 We have
updated the emission rates for model year 1981 and later vehicles in MOVES4.
6.1.1 Light-duty Model Year 1960 to 1980 Vehicles
The MOVES NH3 emission rates for model year 1960 to 1980 vehicles were developed for
MOVES2010 and documented in a MOVES2010 technical report.88 As detailed in that report,
NH3 emission rates for 1960-1974 vehicles were developed using measurements from vehicles
with no catalysts.90 Proposed NH3 1975-1980 emission rates were developed from laboratory
test data on vehicles equipped with oxidation catalysts, which oxidize hydrocarbons (HC) and
carbon monoxide (CO), but do not control NOx 90 The ammonia emission rates for these older
vehicles are significantly lower than older vehicles equipped with three-way catalytic converters,
286
-------
because three-way catalytic converters are designed to also reduce nitrogen, which enables the
formation of ammonia.90 As discussed in the next section, modern three-way-catalytic-converter-
equipped vehicles are capable of having comparably low NH3 emissions due to improved fuel
control.100
Because there are very few 1980 and earlier vehicles on the road for the calendar years of
interest for MOVES runs, for MOVES2010 and later versions we used a single set of NH3
emissions for the entire model year range between 1960-1980. The emission rates for 1960-1980
model years vehicles were calculated as a simple average of the non-catalyst (1960-1974) and
the oxidation catalyst (1975-1980) vehicles as summarized in Table 6-1.
Table 6-1. Development of the 1960-1980 NH3 emission rates used in MQVES2010 and later versions
Model Year
Range
Description
Rate for Idle (OpModelD =1)
emissions (g/lir) for all age groups
1960-1974
Emission rates scaled to non-catalyst
vehicles from 1983 EPA study.9"
Documented in a MOVES2010 report.88
Not used directly in MOVES.
0.153
1975-1980
Emission rates scaled to oxidation catalyst
vehicles from 1983 EPA study.9"
Documented in a MOVES2010 report.88
Not used directly in MOVES.
0.209
1960-1980
Average of the non-catalyst (1960-1974)
and the oxidation catalyst (1975-1980)
emission rates. Used in MOVES2010 and
later versions
0.181
6.1.2 Model Year 1981 and Later Vehicles
For MOVES4, the ammonia emissions rates for light-duty gasoline vehicles for MY 1981 and
later vehicles were updated based on remote sensing device data.
6.1.2.1 Remote Sensing Data
We analyzed ammonia emissions measurements data collected by researchers at the University
of Denver using their remote sensing device called the Fuel Efficiency Automobile Test (FEAT).
The FEAT device measures vehicle emissions across a single lane roadway—typically a freeway
on-ramp— with a light source on one side of the roadway and a detector on the other. In 2005,
University of Denver researchers added NH3 to their existing campaign measuring exhaust
carbon monoxide (CO), hydrocarbons (HC), nitrogen oxide (NO) and nitrogen dioxide (NO2).
FEAT measures emissions of individual species relative to carbon dioxide (CO2) concentrations.
Based on carbon-balance calculations, these molar ratios can be expressed as fuel-specific
emission rates, (e.g., g CO/kg fuel).91 Fuel-specific NH3 measurements from FEAT compare
well to onroad NH3 and tunnel measurements made by other researchers.83
287
-------
Several limitations of using remote sensing data to develop MOVES rates are discussed in
Section 3.2.1.1.3, including the need to combine fuel-specific emission rates with fuel rate
estimates from MOVES or other studies, and the inability to measure emissions across all
operating modes. However, NH3 emissions data are not available in the Denver IM240 dataset
used to estimate deterioration for HC and NOx discussed in Section 3.6. Also, remote sensing
NH3 data was not measured in the Colorado Department of Public Health and the Environment
campaign used to develop CO emission rates in Section 3.7.
The strength of the FEAT remote sensing device for emissions inventory development is its
ability to measure emissions from thousands of in-use vehicles, including high-emitting vehicles
that contribute disproportionately to the emissions inventory.92 The emissions data collected by
FEAT emission measurement campaigns is publicly available and contains over 335,000 light-
duty gasoline vehicle-specific N1 k observations from seven locations across the United States
over 2005 to 2020. Figure 6-1 shows the number of measurements by location and calendar year
(top panel) and by calendar year and vehicle age (bottom panel). Since this analysis was
conducted, additional data from campaigns conducted in 2020 and 2021 have been posted, which
have not been incorporated into the analysis.93
50000
40000
cn
c
0
% 30000
1
a>
CO
O 20000
10000
State, City
CA_FRES
¦ CA_LA
CA SAJO
¦ CA_VANU
CO_DENV
¦ IL_CHIC
OK TULS
2005 2008 2010 2013 2014 2015 2016 2017 2018 2019 2020
Calendar Year
2020
2019
2018
s2017
> 2016
¦g 2015
-i 2014
03
0 2013
2010
2008
2005
10
15 20
Vehicle Age
25
30
35
# Observations
1 4000
3000
2000
1000
40
Figure 6-1. Top panel: Number of Vehicle Ammonia Measurements by Location and Calendar Year. Bottom
panel: Number of Vehicle Ammonia Measurements by Vehicle Age and Calendar Year
288
-------
Table 6-2. Location and Calendar Year of University of Denver RSD sampling campaigns
Location
Year
Mean VSP
(kW/tonne)
Mean MY
Fresno, California
2008
6.4
1999.8
West Los Angeles/La
Brea Blvd. California
2008, 2013, 2015
12.2,4.6,9.8
2001.2, 2004.7, 2006.9
San Jose, California
2008
14.7
2000.6
Van Nuys, California
2010
6.2
2001.5
Denver, Colorado
2005,2013,2015,
8.1, 10.4, -1.4, 8.9,
1998.1, 2005.2, 2007.2, 2009.2,
2017, 2020
6.2
2011.6
Chicago, Illinois
2014, 2016, 2018
5.9, 6.7,4.6
2007.5,2009.6, 2011.6
Tulsa, Oklahoma
2005,2013,2015,
5.3, 7.7, 7.2, 7.8,
1999.3, 2006.3, 2008.2, 2010.1,
2017, 2019
10.3
2011.9
Using the original University of Denver remote sensing data sets listed Table 6-2 and publicly
available at the University of Denver library website,93 we developed a quality-assured dataset
with consistent data processing and naming. We combined the datasets from the different
campaigns into a single file using consistent file names for each column. From each of the files,
emissions were consistently reported in units of molar percent (%) but not all fields contained the
measurements in fuel-based units (g/kg-fuel). We re-calculated the fuel-specific rates for the
entire data set using equations provided in the University of Denver reports.94 The University of
Denver RSD data includes invalid flags by measurement and pollutant (separate for HC, NO,
NO2, NH3, and speed). Any observations that are labeled as invalid were removed from the
database. It is possible to have a valid observation for some pollutants and not others.
Each observation includes a speed and acceleration measurement. Using speed and acceleration,
we re-calculated VSP for each observation using the generic VSP equation provided in the
University of Denver reports.94 Although many observations had missing or invalid speed and
acceleration measurements, we still used these observations to develop the fuel-based emissions
inventories in the analysis below.
Each observation includes the Vehicle Identification Number (VIN). It is possible to decode the
VIN to determine make, model, and vehicle class. A subset of the original University of Denver
datasets include decoded VIN information. We generated a database with consistent decoded
VIN information, using a VIN decoder provided by Eastern Research Group (version
000.012_25octl9, data file version v25octl9, MY range 1981-2019). The ERG VIN decoder
identifies the MOBILE6 vehicle classes, which we then converted into the MOVES regulatory
classes of light-duty vehicles (LD) and light-duty trucks. All heavy-duty MOBILE6 vehicles are
removed from the dataset. Because the VIN decoder only contains vehicle models between 1981
and 2019, our analysis excluded model years before 1981.
6.1.2.2 Average Fuel-based Emission Rates by Model Year Groups
Figure 6-2 shows the average fuel-based NH3 emission rates (g/kg-fuel) by model year and
vehicle class (LDV and LDT). For context, the average rate for model years 2004-2013 for LDV
and LDT is 0.45 gNFb/kg fuel which compares well with Sun et al. 83 who reports 0.44 gNFb/kg
fuel (reported as 0.37 ppbv NFb/ppmv CO2 and converted using factors reported in that study)
for a measurement campaign performed between 2013 and 2014.
Figure 6-2 also shows a significant model year effect in ammonia emission rates.
289
-------
1.2-
1.0-
O)
0,8-
3
CO
0.6-
X
z
c
i
Xj-
O
ra
-------
Table 6-3. Model Year Averaging of NH3 rates
Model Year Range
Averaging approach
1981-1995
Across model year range
1996-2003
By model year
2004-2013
Across model year range
2014-2018
By model year
2018-2060
Same as MY 2018
6.1.2.3 Average Fuel-based Emission Rates by Model Year Group and Vehicle Age
The light-duty ammonia emission rates displayed significant aging effects for the model year
1996 and older vehicles. We believe there are aging effects for the pre-1996 vehicles, but we did
not have data from these vehicles before the age 10, and it is likely that they had already
experienced significant deterioration. Because older ages are associated with older model years,
we plotted the rates by model years groups. Figure 6-3 displays the average rates by vehicle age
plotted by different model year groups. Even for model year groups where the mean emission
rate (2004-2013) is relatively stable, there is an apparent aging effect. We plotted the points by
the remote sensing location to demonstrate that the aging effect is not due solely to an
inadvertent relationship between vehicle age and the location site.
291
-------
1.0-
0.3
a>
3
CO
X
0.1-
1981-1995
•ti
* * ; • T
_ • • ^ 0
* •
1996-2003
§1.01
0.3-
2004-2013
•. • I -1 • • 11!:'
, .«?*
0.1-
2014-2018
• 1
1 •
• f • *
1 # •
State,City
CAFRES
• CA_IA
CA_SAJO
• CA_VANU
CO_DENV
• IL_CHIC
OK TULS
10 15 20 0 5
Vehicle Age
10
15
20
Figure 6-3. Average emission rates by model year and for light-duty vehicles. Note the y-axis is plotted in a
logarithmic scale
The ammonia emission rates in MOVES are stored in the EmissionRateByAge table as described
in Section 2.2.1. The emission rates are classified according to seven age groups: 0-3, 4-5, 6-7, 8-
9, 10-14, 15-19, and 20+ years.
We first estimated average fuel-based emission rates (FERMYiaqe) for the model year ranges
presented in Table 6-3, the seven age groups, and each regulatory class (LDV and LDT). Since
we did not have data to model all combinations, we then used the following methods to estimate
fuel-based emission rates for the regulatory class, model year group and age group combinations
with missing data.
Model Years 1981-1995
For the 1981-1995 model year group, we had no remote sensing measurements for the age
groups younger than age 10. Therefore, we used scaling factors from MOVES2010 to estimate
the NH3 emissions for the missing age groups. In MOVES2010, for model years 1996-2001, we
used emissions measured in an aged catalyst study95 to calculate a ratio between emissions from
new vehicles and emissions at age 6-9 and ages 10-20; the average ammonia emission rate for
ages 0-5 was multiplied by 1.2 to estimate the emission rate for the age 6-9 groups, and by 1.5
for the age 10+ groups. In our current analysis, for model years 1981-1995, we back calculated
fuel-based emission rates for age groups 0-3 and 4-5 by scaling the measured rates by these same
factors. For age group 15-19 we used the factor of 1.5 as in Equation 6-1 for both LDV and
LDT.
292
-------
FTR _PERmy 1981—1995,age 15—19 Equation 6-1
r CtiMY 1981-1995,age 0-5, ~ ^ 4
We estimated the average emission rates for ages 6-7 and 8-9 using Equation 6-2 for both LDV
and LDT.
FERmy 1981-1995,age 6-9 — 1-2 X FERMY 1981-1995,age 0-3 Equation 6-2
Model Years 1996-2003
For the 1996-2003 model year group, we fit an ordinary least squares regression model to the
average of the light-duty ammonia emission rates by vehicle age. Using the model estimated
emission rates by model year, we calculate the mean emission rates by age group shown in Table
6-4. We then calculated aging ratios by dividing the mean emission rate for a given age group by
the mean emission rate of age group 0-3 ("Aging Ratio 1") with the intention of using this ratio
to age emissions of the youngest group. However, many of the model years in this range did not
have data for vehicles younger than 7 years old. Therefore, we calculated a second set of aging
ratios ("Aging Ratio 2") using as reference vehicles of age 15-19 because they were present for
all model years in this group. We still report the "Aging Ratio 1" in this table because it will be
used for model years 2014 and later as described below.
Table 6-4. Estimated NH3 aging effects for the 1996-2003
model year group by MOVES age groups
Age Group
Mean NH3 emissions
(g/kg) from linear
regression estimates
Aging
Ratio 1
Aging
Ratio 2
0-3
0.41
1.0
0.52
4-5
0.49
1.2
0.61
6-7
0.54
1.3
0.67
8-9
0.59
1.4
0.74
10-14
0.67
1.6
0.84
15-19
0.80
1.9
1.00
20+
1.00
2.4
1.25
For the model years, ages, and regulatory class (LDV or LDT) combinations that had missing
data between 1996-2003, we estimated values for the missing age group combinations using the
corresponding "Aging Ratio 2" for the same model year and regulatory class combination as
shown in Equation 6-3.
FERmy xage y — Aging R
-------
Model Years 2004-2013
We used a slightly different approach for model years 2004-2013. Because we combined the
2004-2013 model years into one group, we had good data coverage to generate estimates for
each age group and both regulatory classes, except for age group 20+. As shown in Figure 6-3,
the NH3 emission rates stabilize after about age 8. Based on this observation, we calculate the
aging ratio between the mean NH3 emission rates of ages 8-19 and ages 0-3 for both LDV and
LDT as 1.5 as presented in Equation 6-4.
. . _ FERmy 2004-2013,age 8-19 „ r „ „
Aging RatioMY2qq4-2013,age 20+ ~ T7T7T} — Equation 6-4
t btiMY 2004—2013,age 0-3
Table 6-5 presents the observed aging ratios for all age groups referred to age group 0-3 and the
calculated aging ratio for age group 20+ (in italic) as presented in Equation 6-4. The table also
presents the mean emission rates for each age group and the calculated mean emission rate for
age group 20+ based on Equation 6-5.
FERmy 2004-2013,age 20+
= Aging RatioMY2oo4-2oi3,age 20+ Equation 6-5
X FERmy 2004-2013,age 0-3
Table 6-5. Mean and calculated NH3 emission rates by age group and regulatory class for the MY 2004-2013
LDV
LDT
Age Group
Mean
nh3
(g/kg)
N
Aging Ratio
Mean
nh3
(g/kg)
N
Aging Ratio
0-3
0.37
27623
1.00
0.32
21591
1.00
4-5
0.49
16497
1.34
0.44
14349
1.37
6-7
0.53
15831
1.44
0.45
14460
1.42
8-9
0.55
14552
1.50
0.46
14318
1.44
10-14
0.56
14499
1.54
0.48
15987
1.48
15-19
0.51
677
1.39
0.47
977
1.47
20+
0.55
1.50
0.48
1.50
Model Years 2014-2018
For the 2014-2018 model years, there was no data for ages beyond age 6, and sparse data beyond
age group 0-3 for each model year group and regulatory class. Therefore, for the missing age
group, model year, and regulatory class combinations, we multiplied the age 0-3 emission by
"Aging Ratio 1" from the 1996-2003 group from Table 6-4 as shown in Equation 6-6.
FERmy 2014-2018,agey
= Aging RatiolMY1996_2003,age y x FERMY 2014-2018,age 0-3
294
-------
We used the 1996-2003 aging ratios because the aging estimates were based on data across a
large range of vehicle ages (ages 2 to 20+), and the multiplicative age increase is similar to the
MY 2014-2018 age group as shown in Figure 6-3.
Model Years 2019-2060
For MY 2019 and later emission rates, we had limited NFb LDV measurements, and no LDT
measurements (Figure 6-3). For these model years, we used the average MY 2018 (g/hr) rates for
all 2019 and later model year groups, as discussed in Section 6.1.2.4.
Figure 6-4 and Figure 6-5 display the fuel-based emission rates by model year group and age
group, for LDV and LDT, respectively. In general, the fuel-based ammonia emissions decrease
with model year and increase with age groups. This is not strictly the case, reflecting the mean
measured emission rates by age group and vehicle age, and our methods to estimate missing
combinations.
1.25
1.00
0.75
0.50
„0.25
|W
O)
0 1.25
To 1.00
*2 0.75
0 0.50
0.25
| 0.00
£ 0.25
jp 0.00
~a>
3 125
1.00
0.75
0.50
0.25
0.00
2003
2004-2013
2014
2015
nil hinil
2016
tad
—i 1 1 1 1 1 r—
a)
ro >0 i1^ QJ r r q
2017
¦mill
—i—i—i—i—i—i—i—
cd
co u~j r - a) +
2018
¦¦¦¦III
—i 1 1 1 1 1 1—
"sT O)
cp in hj- cp t- +
O^CDOOOIOOJ O ^ (D (3D O m (SI O 4 ® CO O If) (Nl
co m CD
-I 1—
cd
+
o
O CD 00 O ID CM
Age group
Figure 6-4. Mean and estimated fuel-based LDV NH3 emission rates by selected model year groups and
MOVES age groups.
295
-------
0.75
0.50
0.25
.*0.00
cn
ju-
ra
1981-1995
1996
nihil drill milli idiill
1997
1998
1999
2000
2001
2002
c
o
in
0.75
0.50
0.25
¦| 0.00
CO
I 0.75
-u 0.50
% 0.25
Jp 0.00
ID
"" 0.75
0.50
0.25
0.00
2003
2004-2013
2014
¦ndd
¦nidi
2016
2017
2018
¦¦¦id
¦mill
.¦¦¦III
2015
-1 1 1 1 1 1 1—
^ ai
rt W h (J) ^ ^ +
O (D CD O Ifl (M
-5J" CD ^ CF)
n m n ® Y 7 i t?u?rTc?TTo *? "? ''T T T o
o-q-coaooknoj o t ® o ifi w o 4 ffl oi o Hi (m
Age group
Figure 6-5. Mean and estimated fuel-based LDT NH3 emission rates by selected model year groups and
MOVES age groups.
6.1.2.4 Mass Rates by Operating Mode
In the EmissionRateByAge table, running emission rates are expressed as mass rates (grams per
hour) and by running operating mode. In this section, we describe how we developed MOVES
emissions rates from the fuel-based average emission rates estimated above.
In the MOVES2010 ammonia rate analysis, second-by-second ammonia emissions were
analyzed from a chassis dynamometer study conducted by CE-CERT. The study showed a strong
correlation of ammonia emissions with vehicle specific power, with higher ammonia emissions
produced at high power.88-85 95
The University of Denver remote sensing device data provide single measurements from each
vehicle. Using the vehicle speed and acceleration, we can estimate the vehicle specific power and
the MOVES operating mode for each vehicle measurement. However, the data has limitations in
estimating emissions by operating mode. The measurements at each campaign are made at a
single location - for the University of Denver the data are captured on a freeway onramp, and
thus only a limited range of vehicle speeds and accelerations are captured. Additionally, each
vehicle is only measured at one operating condition. In general, the campaigns focus on
296
-------
locations that allow for a low/medium acceleration mode, but in some years, there were
exceptional situations (e.g. traffic lights not working) that changed this pattern, as summarized in
the "Mean VSP" column of Table 6-2 and detailed in the articles and reports stored in the FEAT
archive.93
In addition, the measured vehicle emissions are likely a function of the vehicle operation before
it is measured. As such, there is an uncorrected delay between the time the measured vehicle
emissions were formed in the engine and the time at which the vehicle speed and acceleration is
measured. For these reasons, previous analysis has shown that remote-sensing data has a weaker
relationship with vehicle specific power compared to laboratory or portable emission
measurement system (PEMS) which measure tailpipe exhaust emissions and vehicle operation
simultaneously across a large range of vehicle operation for each vehicle.96
Despite the limited vehicle operating conditions sampled from single roadside location,
researchers have shown that fuel-based measurements from a single location can be
representative of area-wide emission rates for HC, NO, and CO, with bias less than 30%. The
bias can be minimized if the distribution of vehicle specific power is similar at the RSD location
and area-wide vehicle operation.97 In the NOx evaluation effort, the University of Denver RSD
locations could have less aggressive driving than is modeled for national average driving, which
can lead to significant differences in NOx measured and modeled emissions rates.98'99 Despite
these potential limitations on the representativeness of the operating conditions, we chose to use
the University of Denver remote sensing data in MOVES because it is the most robust data set
available and the fuel-based rates from multiple locations compare well with tunnel and onroad
measurements made at locations throughout the US.83
To estimate operating-mode specific ammonia emission rates using the remote-sensing data, we
multiplied the fuel-specific emission rates estimated in the Section 6.1.2.4 by model year group,
and age (FERMYage) by the MOVES4 fuel-consumption rates by model year and operating
mode (Fuel RatesMY op) as shown in Equation 6-7.
ERMY.age.op = Fuel RatesMY,op x FERMY,age Equation 6-7
Using this approach, the MOVES time-based NH3 rates, have the same relative increase in
emission rates as fuel consumption. This is a desired property of the emission rates because both
fuel consumption and NH3 have a strong positive relationship with vehicle specific power.
For 2019 and later model years, we used the emission rates that were estimated for MY 2018.
For illustration, MOVES NH3 rates for 2018 for both LDV and LDT by operating mode are
shown in Figure 6-6.
297
-------
3-
=i
o
CO
-------
0)
E
O)
£
2
c
o
(/)
E
0
cd
X
0.20
0.15
0.10
0.05
0.00
0.20
0.15
0.10
0.05
0.00
0.20
0.15
0.10
0.05
0.00
1980 1990 2000 2010
Model Year
2020
NJ
o
to
o
Regulatory Class
— LDV
— LDT
NJ
o
ro
Figure 6-7. NH3 emission rates (g/mile) by model year, regulatory class for calendar years 2010,2017 and
2024 calculated from national MOVES runs using M0VES4 default activity and the NH3 rates documented in
this report
6.1.3 Motorcycles
Motorcycle emission rates are unchanged in MOVES4 from previous versions of MOVES.88 The
motorcycle emission rates are estimated using surrogate light-duty emission rates as outlined in
Table 6-6.
Table 6-6. Motorcycle NH3 emission rates using light-duty vehicles emission rates by model year ranges
Motorcycle Model
Year Range
Surrogate light-duty emission rates
Rate for Idle (OpModelD =1)
emissions (g/lir) for all age groups
1960-1999
Non-catalyst light-duty vehicle rates
0.153
2000-2005
Oxidation catalyst light-duty emission
rates
0.209
2006-2060
MOVES2010 light-duty vehicle model
year 1981-1991 emission rates
0.516
299
-------
6.2 Light Duty E85 Vehicles
The NH3 emission rates for light-duty E85 vehicles are set equal to the emission rates from light-
duty gasoline vehicles. We believe this is a reasonable assumption because E85 vehicles also
should produce NH3 from the catalytic reduction of NO in the three-way catalytic converter. A
recent study101 suggests that there is no clear trend in NH3 emissions with different levels of
ethanol while another study observed slightly higher levels of NH3 for high ethanol blends
particularly under cold conditions102. Unfortunately, the remote sensing data analyzed from the
University of Denver cannot determine whether flex-fuel vehicles are using conventional
gasoline, E10, or E85 fuels. As such, the NH3 emissions of E85 vehicles are anticipated to be
included in the average emission rates developed for light-duty gasoline.
6.3 Light Duty Diesel Vehicles
For this version of MOVES, we updated the light-duty diesel rates to be consistent with the
newly updated heavy-duty diesel ammonia rates.39 Previous versions of MOVES (MOVES2010,
MOVES2014, and MOVES3) used light-duty diesel emission rates based on a 1983 EPA
study88 90 and did not account for emission rates from modern diesel vehicles equipped with
selective catalytic reduction (SCR) emission control systems, which actively inject urea-based
diesel exhaust fluid (DEF) into the exhaust stream and can directly release ammonia into the
atmosphere if excess urea is injected.
To develop light-duty diesel NH3 rates for MOVES, we used the same fuel-based emission rates
by the model year groups presented for heavy-duty diesel vehicles in the heavy-duty exhaust
emission report.39 We believe this is a reasonable approximation for several reasons.
First, because light-duty diesels have a very small market share of the light-duty fleet, and
because ammonia emissions from light-duty diesel are significantly lower than those from light-
duty gasoline vehicles (see Figure 6-8), we believe it is appropriate to use a simple approach
rather than a detailed analysis to estimate these rates.
Second, the adoption of selective reduction catalysts (SCR) in light-duty diesel and heavy-duty
diesel had a similar phase-in time frame. SCR was adopted in heavy-duty diesel vehicles starting
in model year 2010 and were implemented in light-duty vehicles in response to the Light-duty
Tier 2 exhaust emissions standards, which were phased in starting with model year 2004 and
full-phased in by model year 2010.
Finally, the model year 2010 and later mean heavy-duty diesel emission rate is 0.18 g/kg-fuel,
which is close to the confidence interval for all light-duty diesel values from the University of
Denver dataset, as seen in Figure 6-8. A limited number of available studies for light-duty diesel
vehicles have found values consistent with the mean heavy-duty diesel emission rate mentioned
above.103-86-87 The mean light-duty diesel rate is 0.22 g/kg-fuel with 95% confidence intervals
between -0.20 and 0.24. This is significantly lower than the rate for gasoline vehicles.
300
-------
0.5-
0.4-
-------
2.5
2.0
1.5-
1.0-
0.5-
^ 0.0H
O 2.5
2003
2017
¦¦iilLiiilllllill
F 2.0
in
w
E
aj
" 1.5
1.0-
0.5-
o.o-
.¦¦¦¦..¦¦¦iiIIhbiiiI
-i 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 r
o
1 111213141516212223242527282930333537383940 0 1 111213141516212223242527282930333537383940
Operating Mode
Figure 6-9. NHs emission rates (g/hour) by operating mode for regulatory class LDV and LDT and Model
Years 2003 and 2017 for all ages
302
-------
6.4 Summary
Reg Class
10-MC
20-LDV
30-LDT
2000 2020
Model Year
2040
1980
Figure 6-10. Base running emission rates for ammonia from age 0-3 gasoline motorcycles, light-duty vehicles,
and light-duty trucks averaged over a nationally representative operating mode distribution. The large
increase in rates in 1980 is explained by technology changes as described in Section 6.1.1
Model Year
Figure 6-11. Base running emission rates for ammonia from age 0-3 diesel light-duty vehicles and light-duly
trucks averaged over a nationally representative operating mode distribution.
0.02
re
0£
«
0.01
Reg Class
20-LDV
30-LDT
303
-------
7 Crankcase Emissions
7.1 Background
In an internal combustion engine, the crankcase is the housing for the crankshaft. The enclosure
forms the largest cavity in the engine and is located below the cylinder block. During normal
operation, a small amount of unburned fuel and exhaust gases escape around the piston rings and
enter the crankcase, and are referred to as "blow-by." These unburned gases are a potential
source of vehicle emissions.
To alleviate this source of emissions, the Positive Crankcase Ventilation (PCV) system was
designed as a calibrated air leak, whereby the engine contains its crankcase combustion gases.
Instead of the gases venting to the atmosphere, they are fed back into the intake manifold where
they reenter the combustion chamber as part of a fresh charge of air and fuel. A working PCV
valve should prevent all crankcase emissions from escaping to the atmosphere.
PCV valve systems have been mandated in all gasoline vehicles, since model year 1969.
7.2 Modeling Crankcase Emissions in MOVES
Crankcase emissions are calculated by chaining a crankcase emissions ratio to the calculators for
start, running, and extended-idle processes. Crankcase emissions are calculated as a fraction of
tailpipe exhaust emissions, which are equivalent to engine-out emissions for pre-1969 vehicles.
Crankcase emissions are calculated for selected pollutants, including THC, CO, and NOx. and the
elemental-carbon and non-elemental-carbon particulate fractions of PM2.5. For each of these
pollutants, ratios are stored in the CrankcaseEmissionRatio table.
For vehicles with working PCV valves, we assume that emissions are zero. Based on EPA
tampering surveys, MOVES assumes a failure rate of 4 percent for PCV valves.104
Consequently, for fuelType/model-year combinations equipped with PCV valves, we assume a
crankcase ratio of 0.04; i.e., emission fractions for the crankcase process are estimated as 4
percent of the emission fractions assumed for uncontrolled emissions. While this 4 percent
estimate may be pessimistic for new vehicles, and optimistic for old vehicles, available data does
not support a more detailed estimate. As older vehicles have higher overall emissions due to
deterioration effects, use of the aggregate rates may understate the impacts of crankcase
emissions.
7.3 Light-duty Gasoline and E-85 Crankcase Emissions
Very little information is available on crankcase emissions, especially those for gasoline
vehicles. A literature review was conducted to identify available data sources for emission
fractions for gasoline vehicles (Table 7-1).
Table 7-1 Selected Sources of published data on hydrocarbon crankcase emissions from gasoline vehicles
Authors
Year
Fuel
No.
Vehicles
Estimate
Units
Heinen and Bennett1"5
1960
Gasoline
5
33
% of exhaust
Bowditch1"6
1968
Gasoline
70
% of exhaust
US EPA107
1985
Gasoline
9
1.21-1.92
g/mi
304
-------
Based on these sources, we estimated emission fractions for model years without mandated PCV
valves. In absence of better information, gasoline emission fractions are a reflection of diesel
research, with the exception of the gasoline HC ratio. Given that the diesel vehicles studied are
largely heavy duty, and that most gasoline vehicles are light-duty, there is a potential mismatch
between the data sources, which is unavoidable due to the paucity of data. As noted previously,
model years with PCV valves were assigned emission fractions calculated as 4 percent of the
fractions shown in Table 7-2. The same fractions are used for E-85 vehicles.
Table 7-2 Emission fractions for vehicles without PCV systems (ratio to exhaust emissions)
Pollutant
Gasoline
(uncontrolled,
pre-1969)
Gasoline (1969 and
later)
THC
0.33
0.013
CO
0.013
0.00052
NOx
0.001
0.00004
PM (all species)
0.20
0.008
The crankcase emission fractions for THC, CO and NO.Tmay underestimate emissions. These
percentages of exhaust emissions are generally based on engine- out, uncontrolled exhaust,
which is not estimated by MOVES. MOVES produces exhaust estimates based on a number of
control technologies (such as catalytic converters). Uncontrolled exhaust in the 1970s was
considerably higher than current tailpipe exhaust.
7.4 Motorcycle Crankcase Emissions
MOVES modeling of crankcase emissions from motorcycles is detailed in a separate report.8
For motorcycles of model year 1978-and-later, MOVES models all crankcase emissions as zero.
7.5 Light-duty Diesel Crankcase Emissions
After 2001, all chassis-certified vehicles, including diesels, are required to avoid venting
crankcase emissions into the atmosphere.108 This requirement differs from turbocharged and
supercharged heavy-duty diesel engines, which are allowed to vent crankcase emissions, as long
as the crankcase emissions are included in the certification tests. As such, we modeled crankcase
emissions from light-duty diesel vehicles with two model-year groups, pre-2001, and post-2001.
The values used for the pre-2001 are the same as the LHD2b3 diesel crankcase emission ratios,
with one exception. For heavy-duty diesel vehicles, we model the same crankcase ratio for all
PM2.5 species (elemental carbon PM2.5, sulfate PM2.5, aerosol water PM2.5, and the remaining PM
(nonECnonS04PM). This is because the EC/PM fraction for light-duty diesel in MOVES is the
same as light-duty gasoline, and the PM2.5 species specific ratios are developed based on EC/PM
fractions of diesel vehicles. For 2001 and later model years, we estimate zero crankcase
emissions, consistent with how we model emissions from closed crankcase systems for heavy-
duty diesel vehicles.108 These crankcase emission ratios are located in Table 7-3.
305
-------
Table 7-3 Light-duty diesel (LDV and LPT) crankcase emission fractions (ratio to exhaust emissions)
Pollutant
Light-duty diesel
1960-2000)
Light-duty diesel
(2001-2060)
THC
0.037
0
CO
0.013
0
NO,
0.001
0
PM2.5 (all species)
0.2
0
8 Nitrogen Oxide Composition
Nitrogen oxides (NO*) are defined as NO + NO2. In MOVES, NO* includes NO, NO2, and a
small amount of nitrous acid (HONO). More information about nitrogen species and the rationale
for including HONO in NO* emissions are discussed in the heavy-duty report.39 The HONO/NO*
ratio is estimated as 0.8 percent of NOx emissions based a 2001 study that measured
concentrations of NO* and HONO from a highway tunnel in Europe.109 The HONO/NO* ratio of
0.8 percent is within the range of measurements from a gasoline vehicle by Trinh et al. (2017)110,
as well as diesel vehicles and fleet-average vehicles summarized in the heavy-duty exhaust
report.39
The NO/NO* and NO2/NO* fractions for light-duty gasoline vehicles and motorcycles were
developed from a report by Sierra Research.8 Light-duty diesel vehicles used the NO/NO* and
NO2/NOX ratios from heavy-duty diesel vehicles updated in MOVES4.100
8.1 Light-Duty Gasoline Vehicles
The NOx and HONO fractions for light-duty gasoline vehicles are presented in Table 8-1 The
HONO fraction of NO* was subtracted from the original NO2 fraction, because the HONO likely
interferes with the estimated NO2 fraction when measured with a chemiluminescent analyzer, as
discussed in the heavy-duty report.39
Table 8-1 NOx and HONO fractions for light-duty gasoline vehicles
Model Year
Running
Start
NO
NO2
HONO
NO
NO2
HONO
1960-1980
0.975
0.017
0.008
0.975
0.017
0.008
1981-1990
0.932
0.06
0.008
0.961
0.031
0.008
1991-1995
0.954
0.038
0.008
0.987
0.005
0.008
1996-2050
0.836
0.156
0.008
0.951
0.041
0.008
306
-------
8.2 Motorcycles
Motorcycle values are based on measurements on light-duty gasoline vehicles, but apply to
different model year groups, to correspond to similar exhaust emission control technologies. The
NO2 fractions reported by Sierra Research8 were adjusted to account for the HONO
measurements.
Table 8-2 NO^and HONO fractions for motorcycles
Model Year
Running
Start
NO
NO2
HONO
NO
NO2
HONO
1960-1980
0.975
0.017
0.008
0.975
0.017
0.008
1981-2000
0.932
0.06
0.008
0.961
0.031
0.008
2001-2005
0.939
0.053
0.008
0.97
0.022
0.008
2006-2009
0.947
0.045
0.008
0.978
0.014
0.008
2010-2060
0.954
0.038
0.008
0.987
0.005
0.008
8.3 Light-duty Diesel Vehicles
The NOi and HONO fractions for light-duty diesel vehicles are the same as those for heavy-duty
diesel, which were updated in MOVES4. The light-duty diesel NO* and HONO fractions apply
to start and running exhaust. Discussion of the heavy-duty diesel fractions is presented in the
corresponding report.39 These values are presented in Table 8-3 for completeness.
Table 8-3 NO* and HONO fractions for Light-duty Diesel Vehicles
Model Year
NO
NO2
HONO
1960-2003
0.9622
0.0298
0.008
2004-2006
0.9325
0.0595
0.008
2007-2009
0.7539
0.2381
0.008
2010-2060
0.8035
0.1885
0.008
307
-------
9 Appendix A. Revisions to the Pre-2004 Model Year PM2.5
Emission Rates between MOVES2010b and MOVES2014
The PM2.5 exhaust emission rates for pre-2004 model year light-duty vehicles are unchanged
between MOVES2014 and the current version, MOVES4. As noted in Section 4.1.3, we
corrected the PM2.5 light-duty gasoline emission rates between MOVES2014 and MOVES2010
to account for the silicon contamination measured in the Kansas City study, using our best
available estimates. The PM2.5 emission rates in MOVES2010 were based on a meta-analysis of
multiple studies and programs. The Kansas City study was used to estimate deterioration from
the estimated zero-mileage emission rates, to estimate the modal PM2.5 emission rates, and the
PM2.5 temperature dependency. In MOVES2014 we reduced the running PM2.5 emission rates
across all age groups and operating modes by the values shown in Table 9-1.
Table 9-1 contains the estimated contribution of silicon to the start (bag 1-bag 3) and the running
(bag 2) PM2.5 emissions measured in Kansas City. The silicone rubber contains silicon, oxygen,
carbon, and hydrogen which contribute to the measured particulate and organic carbon mass. We
estimated the contribution of the silicon to the PM2.5 emission rates by using the elemental
silicon emission rates from the set of 102 tests analyzed for elements. Additionally, we estimated
that the silicone rubber contributed particulate mass equal to 4.075 times the measured silicon
emission rates, as documented in the speciation profile analysis by Sonntag et al. (2013).70 We
applied these estimates to average silicon emission rates measured for each model year group,
and for trucks and cars. The trucks have a higher silicon contribution which is expected due to
higher exhaust temperatures and larger exhaust tailpipes which expose more silicone rubber to
the hot exhaust. The updated emission rates reflect both the reduction in total PM from the
silicon in Table 9-1 and the revised EC/PM ratios in Table 4-4.
Table 9-1 Reductions to PM2.5 in MQVES2014 compared to MOVES2Q10b due to silicon contamination
Stratum
Vehicle
type
Model group
Start
Running
1
pre-1981
0%
35.3%
2
Truck
1981-1990
0%
25.3%
3
1991-1995
0%
34.5%
4
1996-2005
0%
19.1%
5
pre-1981
0%
14.6%
6
Car
1981-1990
0%
3.5%
7
1991-1995
0%
6.1%
8
1996-2005
0%
8.5%
308
-------
10 References
1 USEPA (2023). Greenhouse Gas and Energy Consumption Rates for Onroacl Vehicles in MOVES4. EPA-420-R-
23-026. Office of Transportation and Air Quality. US Environmental Protection Agency. Ann Arbor, MI. August
2023. https://www.epa.gov/moves/moves-technical-reports.
2 USEPA (2023). Population and Activity of Onroad Vehicles in MOVES4. EPA-420-R-23-005. Office of
Transportation and Air Quality. US Environmental Protection Agency. Ann Arbor, MI. August 2023.
https://www.epa.gov/moves/moves-technical-reports.
3 USEPA (2023). Emission Adjustments for Onroad Vehicles MOVES4. EPA-420-R-23-021. Office of
Transportation and Air Quality. US Environmental Protection Agency. Ann Arbor, MI. August 2023
https://www.epa.gov/moves/moves-technical-reports.
4 USEPA Mobile Source Emissions Modeling Workgroup. EPA's New Generation Mobile Source Emissions
Model: Initial Proposal and Issues. EPA-420-R-01-007. Office of Air and Radiation, Office of Research and
Development, Region 4. April, 2001.
5 U SEPA Office of Transportation and Air Quality. Draft Design and Implementation Plan for EPA 's Multi-Scale
Motor Vehicle and Equipment Emission System (MOVES). EPA420-P-02-006. Assessment and Standards Division,
Ann Arbor, MI. October, 2002.
6 Computational Laboratory for Energy, Air, and Risk. Methodology for Developing Modal Emission Rates for
EPA's Multi-Scale Motor Vehicle and Equipment Emission System. EPA420-R-02-027. Department of Civil
Engineering, North Carolina State University, Raleigh, NC. October, 2002.
7 USEPA Office of Transportation and Air Quality. EPA's Onboard Emissions Analysis Shootout: Overview and
Results. EPA420-R-02-026. Assessment and Standards Division, Ann Arbor, MI. October, 2002.
8 USEPA (2012). Use of Data from "Development of Emission Rates for the MOVES Model, " Sierra Research,
March 3, 2010. EPA-420-R-12-022. Office of Transportation and Air Quality. US Environmental Protection
Agency. Ann Arbor, MI. August, 2012. http://www.epa.gov/otaq/models/moves/documents/420rl2022.pdf.
9 USEPA Office of Transportation and Air Quality. Cars and Light Trucks: Vehicle Standards and Regulations.
http://www.epa.gov/otaq/standards.htm.
10 Calvert, J.G.; Heywood, J.B.; Sawyer, R.F.; Seinfeld, J.H. 1993. Achieving acceptable air quality: some
reflections on controlling vehicle emissions. Science. 261:37-44.
11 Sawyer, R.F.; Harley, R.A.; Cadle, S.H.; Norbeck, J.M.; Slott, R.; Bravo, H.A. 2000. Mobile sources critical
review: 1998 NARSTO assessment. Atmospheric Environment. 34:2161-2181.
12 Harley, R.A.; Marr, L.C.; Lehner, J.K.; Giddings, S.N. 2005. Changes in motor vehicle emissions on diurnal to
decadal time scales and effects on atmospheric composition. Environ. Sci. Technol. 39:5356-5362.
13 Harley, R.A.; Hooper, D.S.; Kean, A.J.; Kirchstetter, T.W.; Hesson, J.M.; Balberan, N.T.; Stevenson, E.D.;
Kendall, G.R. 2006. Effects of Reformulated Gasoline and Motor Vehicle Fleet Turnover on Emissions and
Ambient Concentrations of Benzene. Environ. Sci. Technol. 40: 5084-5088.
14 Bishop, G.A.; Stedman, D.H. 2008. A decade of on-road emissions measurements. Environ. Sci. Technol.
42:1651-1656.
15 Ban-Weiss, G.A.; McLaughlin, J.P.; Harley, R.A.; Lunden, M.M.; Kirchstetter, T.W.; Kean, A.J.; Strawa, A.W.;
Stevenson, E.D.; Kendall, G.R. 2008. Long-Term Changes in Emissions of Nitrogen Oxides and Particulate Matter
from On-Road Gasoline and Diesel Vehicles. Atmospheric Environment. 42: 220-232.
16 McDonald, B.; Gentner, D.R. 2013. Long-term trends in motor vehicle emissions in U.S. urban areas. Environ.
Sci. Technol. 47:10022-10031.
309
-------
17 Bishop, G.A.; Stedman, D.H. 2015. Reactive nitrogen species emissions trends in three light-/medium-duty
United States fleets. Environ. Sci. Technol. 49:11234-11240.
Ix Frcy. H.C. 2018. Trends in onroad transportation energy and emissions. J Air Waste Mgmt Assoc. 68(6):514-
563.
19 USEPA (2023). Speciation of Total Organic Gas and Particulate Matter Emissions from Onroad Vehicles in
MOVES4. EPA-420-R-23-006. Office of Transportation and Air Quality. US Environmental Protection Agency.
Ann Arbor, MI. August 2023. https://www.epa.gov/moves/moves-technical-reports.
20 USEPA (2023). Evaporative Emissions from Onroad Vehicles in MOVES4. EPA-420-R-23-023. Office of
Transportation and Air Quality. US Environmental Protection Agency. Ann Arbor, MI. August 2023.
https://www.epa.gov/moves/moves-technical-reports.
21 USEPA Office of Transportation and Air Quality. Fuel Consumption Modeling of Conventional and Advanced
Technology Vehicles in the Physical Emission Rate Estimator (PERE). EPA420-P-05-001. Assessment and
Standards Division, Ann Arbor, MI. February, 2005. (http://www.epa.gov/otaq/models/ngm/420p05001.pdf).
22 U SEPA (2020), Fuel Effects on Exhaust Emissions from Onroad Vehicles in MOVES3. EPA-420-R-20-016.
Office of Transportation and Air Quality, Assessment & Standards Division, Ann Arbor, MI. November 2020.
(http://www.epa.gov/otaq/models/moves/moves-reports.htm).
23 USEPA Office of Air and Radiation, Office of Transportation and Air Quality. IM240 and Evap Technical
Guidance. EPA420-R-00-007. Transportation and Regional Programs Division, Ann Arbor, MI. April, 2000,
(page 106).
24 Heirigs, P.; Dulla, R.; Crawford, R.W. Processing ofIM240 Data for Use in MOVES. SR007-05-02. Sierra
Research, Inc., Sacramento, CA. May, 2007.
25 Singer, B.C.; Harley, R.A.; Littlejohn, D.; Ho, J.; Vo, T. 1998. Scaling of infrared remote sensor hydrocarbon
measurements for motor vehicle emission inventory calculations. Environ. Sci. Technol. 32(21) 3241-3248.
26 Bishop, G.A.; Stedman, D.H. On-road Remote Sensing of Automobile Emissions in the Denver Area: Year 6,
January 2007. Department of Chemistry and Biochemistry, University of Denver, Denver, CO. June 2007.
27
Kish, L. Survey Sampling. John Wiley & Sons, New York. 1965.
oo
Neter, J.; Kutner, M.H.; Nachtsheim, C.J.; Wasserman, W. Applied Linear Statistical Models. Irwin, Chicago.
Fourth Edition. 1996.
29 Barth, M.; An, F.; Younglove, T.; Scora, G.; Levine, C.; Ross, M.; Wenzel, T. Development of a Comprehensive
Modal Emissions Model: Final Report. National Research Council, Transportation Research Board, National
Cooperative Highway Research Program. NCHRP Project 25-11. April, 2000.
30 Annual Certification Test Results & Data: Cars and Light Trucks, https://www.epa.gov/compliance-and-fuel-
economy-data/annual-certification-data-vehicles-engines-and-equipment.
3 1
Engine and Vehicle Compliance System (VERIFY): https://www.epa.gov/ve-certification/engines-and-vehicles-
compliance-information-system-ev-cis-formerly-verify.
32 Frey, H. C.; Wei, Tongchuan. Method for Real- World Measurements of Light Duty Gasoline Vehicle Exhaust
Emissions Using Portable Emission Measurement Systems at North Carolina State University from 2008 to 2018.
Mobile Air Pollutant Emissions Laboratory, Department of Civil, Construction and Environmental Engineering,
North Carolina State University, Raleigh, NC. Unpublished Report, May, 2018.
33 Anderson D.C., et al. 2014. Measured and modeled CO and NOy in DISCOVER-AQ: An evaluation of emissions
and chemistry over the eastern US. Atmospheric Environment 96:78-87.
34 Travis K.R., et al. 2016. Why do models overestimate surface ozone in the Southeast United States? Atmos
ChemPhys 16(21): 13561-13577.
310
-------
35
McClement, D.; Dulla, R.G. Identification ofNon-I/M Vehicles in I/MProgram Vehicle Emission Datasets.
Draft Report, Sierra Research, Sacramento, CA. October, 2007.
36 Wenzel, T. 2001. Reducing emissions from in-use vehicles: an evaluation of the Phoenix inspection and
maintenance program using test results and independent emissions measurements. Environmental Science & Policy.
4:359-376.
37 Air Quality Group, Aerospace, Transportation and Advanced Systems Laboratory, Georgia Technical Research
Institute. Biennial Evaluation of the Emissions Reduction Effectiveness of the Atlanta Vehicle Inspection and
Maintenance Program for 2003-2004. Prepared for the Air Protection Branch, Environmental Protection Division,
Georgia Department of Natural Resources, Atlanta. September, 2007.
38 DeHart-Davis, L.; Corley, E.; Rodgers, M.O. 2002. Evaluating vehicle inspection/maintenance programs using
on-road emissions data. Evaluation Review. 26(2) 111-146.
39 USEPA (2023). Exhaust Emission Rates of Heavy-Duty Onroad Vehicles in MOVES4. EPA-420-R-23-027.
Office of Transportation and Air Quality. US Environmental Protection Agency. Ann Arbor, MI. August 2023.
https://www.epa.gov/moves/moves-onroad-technical-reports
40 U SEPA Office of Transportation and Air Quality. Exhaust Emission Rates for Light-Duty On-road Vehicles in
MOVES2014: Final Report. EPA-420-R-15-005. Assessment & Standards Division, Ann Arbor, MI. October,
2015.
41 Enns, P.; Brzezinski, D. Comparison of Start Emissions in the LA92 and ST01 Test Cycles. EPA420-R-01-025.
(M6.STE.001), USEPA Office of Transportation and Air Quality, Ann Arbor, MI. April, 2001.
42 Glover, E.; Carey, P. Determination of Start Emissions as a Function of Mileage and Soak Time for 1981-1993
Model-year Light-Duty Vehicles. EPA420-R-01-058 (M6.STE.003). USEPA Office of Transportation and Air
Quality, Ann Arbor, MI. November, 2001.
43 Stump, F.D.; Dropkin, D.L.; Tejada, S.B.; Loomis, C.; Pack, C. Characterization of Emissions from
Malfunctioning Vehicles Fueled with Oxygenated Gasoline-Ethanol (E-10) Fuel-Part III. EPA/600/R-01/053
(NTIS PB2004-106735). USEPA National Exposure research Laboratory (NERL). July, 2002.
44 Whitney, K.A. Collection ofIn-Use Mobile Source Emission Samples for Toxicity Testing. 08.02602. Southwest
Research Institute, San Antonio, Texas. October, 2000.
45 Sabate, S. Methodology for Calculating and Redefining Cold and Hot Start Emissions. California Air
Resources Board, El Monte, CA. March, 1996. As cited in: Glover, E.; Carey, P. Determination of Start Emissions
as a Function of Mileage and Soak Time for 1981-1993 Model-year Light-Duty Vehicles. EPA420-R-01-058
(M6.STE.003). November, 2001 (Reference 15).
46 Page 117 at: https://www.arb.ca.gov/msei/downloads/emfac2017-volume-iii-technical-documentation.pdf.
47 US FHWA, Table VM-1, "Annual Vehicle Distance Traveled in Miles and Related Data," Highway Statistics,
2018. Washington, DC: November 2019, https://www.fhwa.dot.gov/policyinformation/statistics/2018/vml.cfm.
48 U SEPA Office of Transportation and Air Quality. Sales of California Vehicles for 2011 Model Year and Beyond
(Cross-Border Sales Policy). Memorandum CISD-11-06 (LDV/LDT/HD). Compliance and Innovative Strategies
Division, Ann Arbor, MI. May 3, 2011. (http://iaspub.epa.gov/otaqpub/display_file.jsp?docid=24724&flag=l).
49 USEPA Office of Transportation and Air Quality. Control of Air Pollution from Motor Vehicles: Tier 3 Motor
Vehicle Emission and Fuel Standards; Final Rule. Federal Register. Vol. 79, No. 81. April 28, 2014. Page 23417.
50 USEPA (2008). Kansas City PM Characterization Study. Final Report. EPA Contract No. GS 10F-0036K
EPA420-R-08-009. Assessment and Standards Division Office of Transportation and Air Quality U.S. US EPA.
Environmental Protection Agency Ann Arbor,. October 27, 2006, Revised April 2008a by EPA staff.
http://www.epa.gov/oms/emission-factors-research/420r08009.pdf.
311
-------
51 NamE.; Fulper, C.; Warila, J.E.; Somers, J.; Michaels, H.; Baldauf, R.; Rykowski, R.; Scarbro, C. Analysis of
Particulate Matter Emissions from Light-Duty Vehicles in Kansas City. EPA420-R-08-010. USEPA, Office of
Transportation and Air Quality, Ann Arbor, MI; Office of Research and Development, Research Triangle Park, NC.
April, 2008.
52 Gibbs, R.E.; Wotzak, G.P.; Byer, S.M.; Kolak, N.P. Sulfates and Particulate Emissions from In-Use Catalyst
Vehicles. Regulated/Unregulated Emissions and Fuel Economy. EPA-600/9-79-047. US Environmental Protection
Agency, 1979.
53 Hammerle, R.H.; Korniski, T.J.; Weir, J.E.; Cladek, E.; Gierczak, C.A.; Chase, R.E.; Hurley, R.G. Effect of
Mileage Accumulation on Particulate Emissions from Vehicles Using Gasoline with Methylcyclopentadienyl
Manganese Tricarbonyl. SAE920731. Society of Automotive Engineers. 1992.
54 Whitney, K.A. Collection of In-Use Mobile Source Emission Samples for Toxicity Testing. SwRI 08.02602.
Southwest Research Institute, San Antonio, TX. October, 2000.
55 Cadle, S.H.; Nebel, G.J.; Williams, R.L. Measurement of Unregulated Emissions from General Motors' Light-
Duty Vehicles. SAE790694. Society of Automotive Engineers. 1979.
56 Urban, C.M.; Garbe, R.J. Regulated and Unregulated Exhaust Emissions from Malfunctioning Automobiles.
SAE790696. Society of Automotive Engineers. 1979.
57
Urban, C.M.; Garbe, R.J. Exhaust Emissions from Malfunctioning Three-Way Catalyst Equipped Automobiles.
SAE800511. Society of Automotive Engineers. 1980.
58 Lang, J.M.; Snow, L.; Carlson, R.; Black, F.; Zweidinger, R.; Tejeda, S. Characterization of Particulate
Emissions from In-Use Gasoline-Fueled Motor Vehicles. Society of Automotive Engineers. SAE 811186. 1981.
59 Volkswagen AG. Unregulated Motor Vehicle Exhaust Emission Gas Components. Volkswagen Aktien
Gesellschaft, Forschung und Entwicklung [Research and Development], Wolfsburg, Germany. 1991.
60 California Air Resources Board. Particulate Exhaust Emissions: Gasoline Powered Vehicles. Emissions Studies
Section, El Monte, CA. Project No. 2R8618. 1986.
61 Cadle, S.H.; Mulawa, P.; Hunsanger, E.C.; Nelson, K.; Ragazzi, R.A.; Barrett, R.; Gallagher, G.L.; Lawson,
D.R.; Knapp, K.T.; Snow, R. Measurement of Exhaust Particulate Matter Emissions from In-use Light-duty Motor
Vehicles in the Denver, Colorado Area. CRCE-24-1. Coordinating Research Council, Alpharetta, GA. NTIS
PB98-136401. March, 1998.
62 Norbeck, J.M.; Durbin, T.D.; Turex, T.J. Measurement of Primary Particulate Matter Emissions from Light-Duty
Motor Vehicles. 98-VE-RT2A-001-FR. University of California, College of Engineering, Center for Environmental
Research and Technology, Riverside, CA. CRC E-24-2. Coordinating Research Council, Alpharetta, GA. NTIS
PB99-151755. 1998.
63 Whitney, K. Measurement of Primary Exhaust Particulate Matter Emissions from Light-duty Motor Vehicles.
CRC E-24-3. Coordinating Research Council, Alpharetta, GA. 1998. NTIS PB99-121279. November, 1998.
64 Chase, R.E.; Duszkiewicz, G.J.; Jensen, T.E.; Lewis, D.; Schlaps, E.J.; Weibel, A.T.; Cadle, S.; Mulawa, P.
Particle mass emission rates from current-technology light duty gasoline vehicles. J Air & Waste Manage. Assoc.
2000: (50) 930-935.
65 Southwest Research Institute. VOC/PM Cold Temperature Characterization and Interior Climate Control
emissions/Fuel Economy Impact. SwRI 03.11382.04. San Antonio, TX. Prepared for USEPA, Office of
Transportation and Air Quality, Ann Arbor, MI. October, 2005.
66 Li, W.; Collins, J.F.; Norbeck, J.M.; Cocker, D.R.; Sawant, A. Assessment of Particulate Matter Emissions from
a Sample of In-Use ULEV and SULEV Vehicles. Society of Automotive Engineers. SAE 2006-01-1076. 2006.
67 U.S. Environmental Protection Agency. Study of Particulate Matter Emissions from Motor Vehicles. 1981.
312
-------
68 USEPA (2008). Kansas City PM Characterization Study. Final Report. EPA Contract No. GS 10F-0036K
EPA420-R-08-009. Assessment and Standards Division Office of Transportation and Air Quality U.S. US EPA.
Environmental Protection Agency Ann Arbor,. October 27, 2006, Revised April 2008a by EPA staff.
http://www.epa.gov/oms/emission-factors-research/420r08009.pdf.
69 Fujita, E.M.; Campbell, D.E.; Zielinska, B. Chemical Analysis of Lubrication Oil Samples from a Study to
Characterize Exhaust Emissions from Light-Duty Gasoline Vehicles in the Kansas City Metropolitan Area. Desert
Research Institute, Division of Atmospheric Sciences, Reno, NV. CRCE-69a. Coordinating Research Council,
Alpharetta, GA. December, 2006.
70 Sonntag, D. B., R. W. Baldauf, C. A. Yanca and C. R. Fulper (2013). Particulate matter speciation profiles for
light-duty gasoline vehicles in the United States. Journal of the Air & Waste Management Association, 64 (5), 529-
545. DOI: 10.1080/10962247.2013.870096.
71 USEPA (2016). Light-Duty Automotive Technology, Carbon Dioxide Emissions, and Fuel Economy Trends: 1975
Through 2016. EPA-420-R-16-00.
72 USEPA (2014). The Effects of ultra-Low Sulfur Gasoline on Emissions from Tier 2 Vehicles in the ln-Use Fleet
Final Report. EPA 420-R-14-002.
73 USEPA (2013y. EPAct Fuel Effects Study Pilot Phases 1 and 2. Memorandum to Docket EPA-HQ-OAR-2011-
0135.
74 USEPA (2013). Assessing the Effect of Five Gasoline Properties on Exhaust Emissions from Light-Duty Vehicles
Certified to Tier 2 Standards, Analysis of Data from EPAct Phase 3. EPA-420-R-13-002.
75 National Renewable Energy Laboratory (2014). Effect of Gasoline Properties on Exhaust Emissions from Tier 2
Light-Duty Vehicles - Final report: Phases 4,5, & 6, Subcontract Report. NREL/SR-5400-61099. Golden, CO.
76 California Air Resources Board (2015). Technical Support Document: An Update on the Measurement ofPM
Emissions at LEV ILL Levels.
11 USEPA (2017). Tier 3 Certification Fuel Impacts Test Program,
78 USEPA (2021). The 2020 EPA Automotive Trends Report: Greenhouse Gas Emissions, Fuel Economy, and
Technology since 1975. EPA-420-R-21-003.
79 USEPA (2021). Docket ID EPA-HQ-OAR-2021-0208; EPA CCEMS Post Processing Tool Project; Document ID
EPA-HQ-OAR-2021 -0208-013 3
80 USEPA (2021). Final Rule to Revise Existing National GHG Emissions Standards for Passenger Cars and Light
Trucks Through Model Year 2026. Final Rule. Federal Register. Vol. 86, No. 248. December 30, 2021. Page
74434.
81 Title 40: Code of Federal Regulations. Parts 79, 80, 85, etal. Control of Air Pollution From Motor Vehicle
Emission and Fuel Standards; Final Rule.
82USEPA (2020), Brake and Tire Wear Emissions from Onroad Vehicles in MOVES3. EPA-420-R-20-014. Office
of Transportation and Air Quality. Assessment and Standards Division, Ann Arbor, MI. November 2020.
83 Sun, K., L. Tao, D. J. Miller, D. Pan, L. M. Golston, M. A. Zondlo, R. J. Griffin, H. W. Wallace, Y. J. Leong, M.
M. Yang, Y. Zhang, D. L. Mauzerall and T. Zhu (2017). Vehicle Emissions as an Important Urban Ammonia Source
in the United States and China. Environ Sci Technol, 51 (4), 2472-2481. DOI: 10.1021/acs.est.6b02805.
84 Easter, J. E. and S. V. Bohac (2016). Influence ofHCCI and SACI Combustion Modes on NH3 Generation and
Subsequent Storage across a TWC-SCR System. 0148-7191. SAE Technical Paper.
85 Huai, T., T. D. Durbin, J. W. Miller, J. T. Pisano, C. G. Sauer, S. H. Rhee and J. M. Norbeck (2003). Investigation
of NH3 Emissions from New Technology Vehicles as a Function of Vehicle Operating Conditions. Environ Sci
Technol, 37 (21), 4841-4847. DOI: 10.1021/es030403+.
313
-------
86 Suarez-Bertoa, R., A. A. Zardini and C. Astorga (2014). Ammonia exhaust emissions from spark ignition vehicles
over the New European Driving Cycle. Atmospheric Environment, 97 (0), 43-53. DOI:
http://dx.doi.Org/10.1016/j.atmosenv.2014.07.050.
87 Suarez-Bertoa, R., P. Mendoza-Villafuerte, F. Riccobono, M. Vojtisek, M. Pechout, A. Perujo and C. Astorga
(2017). On-road measurement of NH3 emissions from gasoline and diesel passenger cars during real world driving
conditions. Atmospheric Environment, 166, 488-497. DOI: https://doi.Org/10.1016/j.atmosenv.2017.07.056.
88 U SEPA (2012). Use of Data from "Development of Emission Rates for the MOVES Model, " Sierra Research,
March 3, 2010. EPA-420-R-12-022. Office of Transportation and Air Quality. US Environmental Protection
Agency. Ann Arbor, MI. August, 2012. http://www.epa.gov/otaq/models/moves/documents/420rl2022.pdf.
89 Emery, C., M. Trail, C. Lindhjem, T. Shah, Y. Alvarez, S. S. Varada and G. Yarwood (2020). Investigating
Sources of Ammonia Uncertainty in Modeling the Salt Lake City PM2.5 Nonattainment Area. Prepared for Utah
Division of Air Quality. Prepared by Ramboll US Corporation. May 2020.
90 Harvey, C. A., R. J. Garbe, T. M. Baines, J. H. Somers, K. H. Hellman and P. M. Carey (1983). A study of the
potential impact of some unregulated motor vehicle emissions. SAE transactions, 280-289.
91 Bishop, G. A. and D. H. Stedman (2015). Reactive Nitrogen Species Emission Trends in Three Light-/Medium-
Duty United States Fleets. Environ Sci Technol, 49 (18), 11234-11240. DOI: 10.1021/acs.est.5b02392.
92 Ropkins, K., J. Beebe, H. Li, B. Daham, J. Tate, M. Bell and G. Andrews (2009). Real-World Vehicle Exhaust
Emissions Monitoring: Review and Critical Discussion. Critical Reviews in Environmental Science and Technology,
39 (2), 79-152. DOI: 10.1080/10643380701413377.
93 Fuel Efficiency Automobile Test Data Repository, from https://digitalcommons.du.edu/feat/.
94 Bishop, G. (2020). On-Road Remote Sensing of Automobile Emissions in the Denver Area: Winter 2020. CRC
Project No. E-123 Denver 2019. Coordinating Research Council, Inc. 5755 North Point Parkway, Suite 265,
Alpharetta, GA 30022. July 2020. https://crcao.org/published-reports-full/.
95 Durbin, T. D., J. T. Pisano, T. Younglove, C. G. Sauer, S. H. Rhee, T. Huai, J. W. Miller, G. I. MacKay, A. M.
Hochhauser, M. C. Ingham, R. A. Gorse, L. K. Beard, D. DiCicco, N. Thompson, R. J. Stradling, J. A. Rutherford
and J. P. Uihlein (2004). The effect of fuel sulfur on NH3 and other emissions from 2000-2001 model year vehicles.
Atmospheric Environment, 38 (17), 2699-2708. DOI: https://doi.Org/10.1016/j.atmosenv.2004.02.009.
96 Sonntag, D., C. Toro, D. Choi, J. Warila, E. McCurry, C. Fulper and M. Beardsley (2018). Updated Evaluation of
MOVES Light-duty Gasoline NOx Emission Rates with Real World Measurements. 28th CRC Real World Emissions
Workshop, Garden Grove, CA. March 18-21, 2018.
97 Lee, T. and H. C. Frey (2012). Evaluation of Representativeness of Site-Specific Fuel-Based Vehicle Emission
Factors for Route Average Emissions. Environ Sci Technol, 46 (12), 6867-6873. DOI: 10.1021/es204451z.
98 USEPA (2021). Overview of Progress and Findings from the Cross-EPA Coordination Effort for Understandingand
Evaluating NOx Emissions Discrepancies EPA-454-R/21-008. US Environmental Protection Agency. November
2021. https://www.epa.gov/moves/moves-technical-reports.
99 Choi, D., Sonntag, D., Warila, J., Toro, C., Beardsley, M., Brzezinski, D., Henderson, B., Timin, B., Eyth, A.,
Driver, L. (2017). Comparisons of MOVES Light-duty Gasoline NOx Emission Rates with Real-world
Measurements. American Geophysical Union Fall Conference, New Orleans, LA, December 2017.
100 Livingston, C., P. Rieger and A. Winer (2009). Ammonia emissions from a representative in-use fleet of light and
medium-duty vehicles in the California South Coast Air Basin. Atmospheric Environment, 43 (21), 3326-3333. DOI:
10.1016/j.atmosenv.2009.04.009.
101 Karavalakis, G., Durbin, T., Tang, M. (2022), Comparison of Exhaust Emissions Between E10 CaRFG and
Splash Blended El5, Final Report for the California Air Resources Board, June 2022,
https ://ww2. arb. ca.gov/resources/documents/comparison-exhaust-emissions-between-e 10-carfg-and-splash-blended-
el5
314
-------
102 Suarez-Bertoa, R., Zardini, A., Keuken, H., Astorga, C., (2015). Impact of ethanol containing gasoline blends on
emissions from a flex-fuel vehicle tested over the Worldwide Harmonized Light duty Test Cycle (WLTC), Fuel,
143,173-182, DOI: 10.1016/jiuel.2014.10.076.
103 Karavalakis, G., Durbin, T., Russell, R., (2014), Biodiesel and Renewable Diesel Characterization and Testing in
Modern LD Diesel Passenger Cars and Trucks, CRC Project No. AVFL-17b, Coordinating Research Council, Inc.
5755 North Point Parkway, Suite 265, Alpharetta, GA 30022, November 2014, https://crcao.org/wp-
content/uploads/2019/05/CRC-Final-Report_A VFL-17b.pdf
104 US Environmental Protection Agency. EPA Motor Vehicle Tampering Survey. EPA 420-A-90-001. 1990.
105 Bennet, P. A. et al. Reduction of Air Pollution by Control of Emissions from Automotive Crankcase. SAE Paper
142A, January 1960.
106 Bowditch, F.W. The Automobile and Air Pollution. SAE Paper No. 680242. Presented at SAE Mid-Year
Meeting, Milwaukee, WI. May, 1968.
107USEPA (1985). Crankcase Emissions with disabled PCVsystems. EPA 460/3-84-011. Office of Mobile Source
Air aPollution Control. US Environmental Protection Agency. Ann Arbor, MI. March 1985.
https://nepis.epa.gOv/Exe/ZyPDF.cgi/9100Y26S.PDF?Dockey=9100Y26S.PDF.
108 Title 40: Code of Federal Regulations. Part 86- Protection of Environment. Control of Emissions from New and
In-Use Highway Vehicles and Engines. 86.1810-01 Subpart S—General Compliance Provisions for Control of Air
Pollution From New and In-Use Light-Duty Vehicles, Light-Duty Trucks, and Complete Otto-Cycle Heavy-Duty
Vehicles. General standards; increase in emissions; unsafe conditions; waivers.
109 Kurtenbach, R., K. H. Becker, J. A. G. Gomes, J. Kleffmann, J. C. Lorzer, M. Spittler, P. Wiesen, R. Ackermann,
A. Geyer and U. Piatt (2001). Investigations of emissions and heterogeneous formation of HONO in a road traffic
tunnel. Atmospheric Environment, 35 (20), 3385-3394. DOI: http://dx.doi.org/10.1016/S1352-2310(01)00138-8.
110 Trinh, H. T., K. Imanishi, T. Morikawa, H. Hagino and N. Takenaka (2017). Gaseous nitrous acid (HONO) and
nitrogen oxides (NOx) emission from gasoline and diesel vehicles under real-world driving test cycles. Journal of
the Air & Waste Management Association, 67 (4), 412-420. DOI: 10.1080/10962247.2016.1240726.
315
------- |