Development of Emission Rates for

            Light-Duty Vehicles in the Motor Vehicle

            Emissions Simulator (MOVES2010)


            Final Report
&EPA
United States
Environmental Protection
Agency

-------
       Development of Emission Rates for
   Light-Duty Vehicles in the Motor Vehicle
      Emissions Simulator (MOVES2010)

                        Final Report
                    Assessment and Standards Division
                   Office of Transportation and Air Quality
                   U.S. Environmental Protection Agency
      NOTICE

      This technical report does not necessarily represent final EPA decisions or
      positions. It is intended to present technical analysis of issues using data
      that are currently available. The purpose in the release of such reports is to
      facilitate the exchange of technical information and to inform the public of
      technical developments.
United States
Environmental Protection
Agency
EPA-420-R-11-011
August 2011

-------
                                    Table of Contents

1.    Gaseous Exhaust Emissions from Light-Duty Gasoline Vehicles (THC, CO, NOx)	1
  1.1    Introduction	1
     1.1.1   MOVES Background	1
     1.1.2   Light-Duty Vehicles	1
     1.1.3   Differences between MOVES and MOBILE	2
     1.1.4   Overview	7
  1.2    Emissions Sources (sourceBinID) and Processes (polProcessID)	9
     1.2.1   The emissionRateByAge Table	11
       1.2.1.1    Age Groups (ageGroupID)	11
  1.3    Exhaust Emissions for Running Operation	13
     1.3.1    Operating Modes (opModelD)	13
     1.3.2   Scope	15
     1.3.3   Emission-Rate development: Subgroup 1 (Model years through 2000)	15
       1.3.3.1    Data Sources	15
         1.3.3.1.1    Vehicle Descriptors	15
            1.3.3.1.1.1    Track Road-Load Coefficients: Light-Duty Vehicles	16
         1.3.3.1.2    Test Descriptors	17
         1.3.3.1.3    Candidate Data Sources	17
       1.3.3.2    Data Processing and Quality-assurance	19
         1.3.3.2.1    Sample-design reconstruction (Phoenix only)	20
       1.3.3.3    Source selection	21
       1.3.3.4    Methods	22
         1.3.3.4.1    Data-Driven Rates	22
            1.3.3.4.1.1    Rates: Calculation of weighted means	23
            1.3.3.4.1.2    Estimation of Uncertainties for Cell Means:	23
         1.3.3.4.2    Model-generated Rates (hole-filling)	25
            1.3.3.4.2.2    Rates	25
               1.3.3.4.2.2.1    Coast/Cruise/Acceleration	27
               1.3.3.4.2.2.2    Braking/Deceleration	30
            1.3.3.4.2.3    Estimation of Model Uncertainties	32
            1.3.3.4.2.4    Reverse transformation	32
         1.3.3.4.3    Table Construction	33
       1.3.3.5    Verification and Adjustment for High-Power Operating modes	34
       1.3.3.6    Estimating Rates for non-I/M Areas	43
       1.3.3.7    Stabilization of Emissions with Age	53
         1.3.3.7.2    non-I/M Reference Rates	58
     1.3.4   Emission-Rate Development: Subgroup 2 (MY 2001 and later)	60
       1.3.4.1    Data Sources	60
         1.3.4.1.1    Vehicle Descriptors	60
       1.3.4.2    Estimating I/M Reference Rates	60
         1.3.4.2.1    Averaging IUVP Results	61
         1.3.4.2.2    Develop Phase-In Assumptions	65
         1.3.4.2.3    Merge FTP results and phase-in Assumptions	69
         1.3.3.2.4    Estimating Emissions by Operating Mode	75
            1.3.4.2.4.1    Running Emissions	75
         1.3.4.2.5    Apply Deterioration	81
            1.3.4.2.5.1    Recalculate the logarithmic mean	81
            1.3.4.2.5.2    Apply a logarithmic Age slope	81

-------
            1.3.4.2.5.3    Apply the reverse transformation	84
          13.4.2.6    Estimate non-I/M References	85
   1.4     Exhaust Emissions for Start Operation	86
     1.4.1   Subgroup 1: Vehicles manufactured in model year 1995 and earlier	86
        1.4.1.1    Methods	86
          1.4.1.1.1    Data Sources	86
          1.4.1.1.2    Defining Start Emissions	87
          1.4.1.1.3    Relationship between Soak Time and Start Emissions	87
     1.4.2   Subgroup 2: Vehicles manufactured in MY1996 and later	89
     1.4.3   Applying Deterioration to Starts	90
        1.4.3.1    Assessing Start Deterioration in relation to Running Deterioration	90
        1.4.3.2    Translation from Mileage to Age Basis	99
        1.4.3.3    Application of Relative Multiplicative Deterioration	101
   1.7     Replication and Data-Source Identification	108
2.    Particulate-Matter Emissions from Light-Duty Vehicles	112
   2.1     Introduction and Background	112
     2.1.1   Particulate Measurement in the Kansas City Study	113
     2.1.2   Causes of Gasoline PM Emissions	116
   2.2     New Vehicle or Zero Mile Level (ZML) Emission Rates	118
     2.2.1   Longitudinal Studies	120
     2.2.2   New Vehicle, or ZML Emission Rates and Cycle Effects	121
     2.2.3   Aging or Deterioration in Emission Rates	126
        2.2.2.4    Age Effects or Deterioration Rates	126
   2.3     Estimating Elemental and Organic Carbon Fractions	129
   2.4     Modal PM Emission Rates	135
     2.4.1   Typical behavior in particulate emissions as measured by the Dustrak and Photoacoustic
     Analyzer	136
   2.5     Conclusions	144
3.    Gaseous and Particulate Emissions from Light-Duty Diesel Vehicles (THC, CO, NOx, PM)	146
   3.1     Gaseous Emissions: MY2009 and earlier,  Particulate  Emissions: MY2003 and earlier	146
        3.1.1.2    Estimating Bag Emissions:	147
        3.1.1.3    Assigning Operating Modes for Starts (Adjustment for Soak Time)	150
     3.1.2   Running  Emissions by Operating Mode	152
   3.2     Gaseous Emissions: MY2010 and Later, Particulate Emissions: MY2004 and Later	156
     3.2.1   Gaseous  Emissions	156
     3.2.2   Particulate Emissions	156
   3.3     Particulate Emissions: Estimating Elemental and Organic Carbon Components (EC, OC)... 156
     3.3.1   Group  1: MY 2003 and earlier	156
     3.3.2   Group 2: MY 2004 and later	157
4.0     Crankcase Emissions	158
Appendix A: Peer-Review Comments and Response: Reviewer 1	161
Appendix B: Peer-Review Comments and Response: Reviewer 2	182
References	190
                                               11

-------
1.     Gaseous Exhaust Emissions from Light-Duty Gasoline
Vehicles (THC, CO, NOx)


1.1    Introduction

1.1.1   MOVES Background

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 evaporation emissions,
estimation of usage and activity patterns for vehicles, 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 Proposal 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 predecessor1.
   •   A subsequent "Draft Design and Implementation Plan" describes the MOVES design and
       introduces the reader to concepts and terminology developed for the new model2.
   •   Readers wishing to  further understand the development of the modal design for running
       emissions can consult the "Methodology for Developing Modal Emission Rates, "3 as well
       as the "Shoot Out"4 conducted among several candidate approaches.
   •   This document focuses on development of inputs to the MOVES Database. Readers
       interested in further understanding the processes used by the model to process inputs into
       inventory estimates can consult the MOVES  Software Design Reference Manual
       (SDRM)5.

A large volume of additional documentary and supporting materials can be obtained at
http://www.epa.gov/otaq/models/moves/movesback.htm. In general, the most recent and
relevant materials are at the top of the page, with older material located 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.
1.1.2 Light-Duty Vehicles

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 (NOx). The resulting model inputs are included in the
MySQL database supporting the MOVES2010 model.

Light-duty vehicles are defined as cars and trucks with gross vehicle weight ratings (GVWR) of
less than 8,500 Ibs. For purposes of emissions standards "cars" are designated as "LDV" or
"passenger cars" (PC), and are distinguished from "trucks" which are further subclassified as
"light light-duty trucks" (LLDT) and "heavy light-duty trucks" (HLDT), on the basis of GVWR
< 6000 Ibs and GVWR > 6000 Ibs, 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 truck classes for
purposes of inventory estimation, we will refer to "cars" and "trucks" throughout.

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 forty
years. The Clean Air Act (CAA), passed in 1970 (and amended in 1977 and 1990), set "National
Ambient Air-Quality Standards" (NAAQS) for HC, CO and NOx.  Carbon monoxide is targeted
for its respiratory toxicity, and HC and NOx largely for their roles in production of ground-level
ozone, another pollutant targeted under the CAA. 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, and Tier-2 standards starting in 2004.  Concurrently, the state of California and
additional  states electing to adopt "California" in lieu of "Federal" standards have implemented
the "LEV-I" and "LEV-II" 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 Inspect!on-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 a primary 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, and on-board diagnostic systems (OBD).
Development of emission rates thus largely involves constructing a "numerical" account of this
history. However, 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
database. However, this history has been well described elsewhere, and we refer interested
readers to  the EPA website6'7, as well as to the peer-reviewed literature8'9'10'11'12'13.
1.1.3  Differences between MOVES and MOBILE

At the outset, it is useful to highlight four important different between MOVES and MOBILE.
(1) While intending to estimate average emissions across the entire vehicle fleet, MOVES does

-------
not distinguish between "normal" and "high emitters," (2) MOVES inverts MOBILE'S approach
to inspection and maintenance, (3) MOVES is a "modal" model, whereas MOBILE is "non-
modal," and (4) emission rates developed for MOVES are expressed in time-specific, rather than
distance specific terms, i.e., mass/time (g/hr), rather than mass/distance (g/mi, g/km).

 1.  A fundamental difference between MOVES and MOBILE is that MOVES does not classify
vehicles into "emitter classes." The MOBILE model(s) provided different sets of emission rates
for "normal" and "high"  emitters. While arbitrary, this distinction made qualitative and practical
sense because the emission rates were themselves averages of FTP test results.

We didn't attempt a similar approach in MOVES for several reasons, some conceptual, some
practical. The main conceptual reason is that in review of data, we did not see clear evidence of
distinct "high emitter" subpopulations. Rather, review of emissions data seems to show highly
skewed but continuous distributions with long tails, which we treat as log-normal for modeling
purposes.  Clearly, the vehicles in the upper percentiles of the distributions make
disproportionate contributions to the inventory, assuming similar driving patterns to cleaner
vehicles in the lower percentiles.  Based on these observations, our approach has been to capture
the mean of the entire distribution, including the upper tail. We illustrate these concepts using
two examples, based on aggregate cycle means from the Phoenix I/M program, measured on the
EVI147 cycle.

Figure 1 shows  cumulative distributions of NOx emissions for "young" cars, aged 0-3 years,
representing two sets of emissions standards. The blue distribution represents "Tier 0" vehicles,
manufactured prior to  1994; the green represents "Tier 1" vehicles, manufactured in 1996-97,
and the red represents a mix of the two, during the Tier-1 phase-in period (1994-95). Note that
the combination of reduced standards and improved technology pushes the entire distribution
"leftward" or towards lower emission  levels.

-------
Figure 1. Cumulative distributions of running NOx for cars, Age 0-3, measured on the IM147 cycle (Source:
Phoenix I/M program).
  a-
  §
  I
100-

so

80-

70-

60

50-

40-

30-

20-

10;

 0-
        bengrojp
                              345
                              NOx Mass Rate,  (g/mi)

                          Tier-0   	 Tier-0/Tsr—1
                                                      Tier-1
A similar example, Figure 2, shows NOx distributions for Tier-1 vehicles (MY1996-97) at two
different age levels, 0-3 and 8-9 years old, shown in blue and red, respectively. Qualitatively, the
picture looks very similar to Figure 1, except that in this case we can see the effect of age in
pushing the entire distribution "rightwards," towards higher emission levels. Note that the entire
distribution shifts, including the lower percentiles, not only the "high emitters" in the upper
percentiles.

-------
Figure 2. Cumulative distributions of running NOx for Tier-1 cars, at two age levels, measured on the IM147
cycle (Source: Phoenix I/M program).
  fr

  I
  It
  I
 o
     ttC-J
      ic-
Group
 45

Mots Rata,

 0-3 yr
                                               9—9 yr
A pattern not necessarily apparent in Figure 1 emerges if we view the same distributions on a
logarithmic scale, as shown in Figure 3. In the logarithmic view, we can see that the distribution
at 8-9 years is the same is that at 0-3 years, but shifted to the right; that is, the shapes (variances)
of the two distributions are very similar, but the means are shifted. These figures illustrate the
"logarithmic" or "multiplicative"  scaling typical in emissions data. The utility of logarithms in
modeling follows from the fact that multiplicative patterns representing actual changes can be
represented and projected very conveniently as additive changes in logarithmic space.  These
patterns obtain whether the data are analyzed with respect to technology, age or power.  The
development of emission rates, as described in this chapter (and for PM in Chapter 2), relies
heavily on these concepts.  Figure 4 shows a similar picture to Figure 2, except for THC; what is
notable is that the THC distributions are even more skewed than the NOx distributions.

-------
Figure 3. Cumulative distributions of running NOx for Tier-1 cars, at two age levels, measured on the IM147
cycle (LOGARITHMIC SCALE) (Source: Phoenix I/M program).
      0.0001         Q.0010         0.0100         0.1000
                                 NOx Mass Rate,  (g/mi)
                     Age Group   	 0—3 yr  	  8—9 yr
                                                            1.0000       10,0000
Figure 4.  Cumulative distributions of running THC for Tier-1 cars, at two age levels, measured on the
IM147 cycle (Source: Phoenix I/M program).
  Q)
  ••f
  o
mo-






vSJ-

0:
/•"
/
7







x- — '









_ . ^ —















































































        0       1
                                  THC  Mass Rate,  (g/mi)
                      Age Group   	  0-3 yr  	 8-9 yr

-------
In addition to the conceptual reason just illustrated, there were practical reasons for not creating
one or more "high-emitter" classes: (1) a vehicle or test showing high emissions for one
pollutant need not show high emissions for other pollutants, (2) high emissions may be a
transitory phenomenon in many cases, i.e., vehicles with high results for one set of
measurements may not show similar results if re-measured; in such cases it is very difficult to
determine whether the apparent change is due to an actual change in the vehicle or the
notoriously high variability of emissions measurements, (3) given that rate development for
MOVES operating modes is not coupled to the FTP (or any particular cycle), convenient and
non-arbitrary definitions of "high emitter" are not readily available, and (4) distinction of emitter
classes would require that the intensive process of rate development be repeated for each class,
including the projection of emissions by age and power, and development of distinct adjustments
for temperature and fuel (performed separately). The detailed data required for these analyses
and their projection into the future is not available.
2. A second important difference between MOVES and MOBILE is that MOVES inverts
MOBILE'S approach to inspection and maintenance. That is, the emission rates provided with
MOBILE represented "non-I/M" conditions, and MOBILE represented I/M conditions by
making adjustments during model runs. By contrast, the MOVES input table contains two sets
of rates, representing "I/M reference" and "non-I/M" conditions, respectively. In development
of these rates, "I/M conditions" were assigned as the default case, and rates representing "non-
I/M" conditions were developed in relation to rates representing "I/M conditions." During model
runs, MOVES represents particular I/M programs as a function of both sets of rates, modified by
adjustments calculated to represent the parameters of specific programs.  These topics are
discussed in greater detail in 1.3.3.6.

3. A third major difference between MOVES and MOBILE is that MOVES is modal, whereas
MOBILE was not.  This feature gives MOVES tremendous flexibility, allowing users to
represent any driving pattern, across a range of temporal  and spatial scales. The modal emission
rates are applied consistently at the different analysis scales under which MOVES operates  -
national, county and project.

4. Finally, emission rates in MOVES are expressed as "time-specific" rates (mass/time, g/hr), as
opposed to "distance-specific" rates (g/mi), as were rates in MOBILE. With respect to model
design, the purpose for this change was to introduce a measurement basis that would be
applicable to all emissions sources, processes, and operating modes, including those for which a
distance-specific basis is not applicable. Examples include all emissions for nonroad equipment,
which are expressed on a mass/work basis in the NONROAD model (g/hp-hr, g/kW-hr), and idle
or "hotelling" emissions for all sources, which occur while the source is stationary.

1.1.4  Overview

Section 1 describes the structure of the MOVES emissionRateByAge table, as it applies to
gaseous-pollutant emissions from gasoline-fueled light-duty vehicles.  The values in this table
describe the "base rates" (meanBaseRate). These values represent mean emissions on the
MOVES reference fuel on a temperature range of 68-86 °F, and unadjusted for the effects of
temperature, humidity, air-conditioning and inspection-and-maintenance programs (I/M). The
adjustments for these factors, applied during MOVES runs, are described in a separate report:

-------
"MOVES2010 Highway Vehicle Temperature, Humidity, Air Conditioning, and Inspection and
Maintenance Adjustments."34

The emissionRateByAge table includes rates representing start and running operation, defined as
distinct "processes" in MOVES.  Rates representing "running operation" are described in Section
1.3, and those for "start operation" are described in section 1.4.

For running emissions, section 1.3.3 describes the development of emission rates for vehicles
manufactured prior to model year 2000. Sub-sections 1.5.1 and 1.5.2 describe the process of data
selection and quality assurance. Rates were generated either directly from available data (sub-
section 1.3.3) or by development and application of statistical "hole-filling" models (sub-section
1.3.4).  These rates were derived using data from the Phoenix I/M program and represent rates
characteristic of a program with features similar to those in the Phoenix program14.

Because the analyses described in sub-sections 1.3.3 and 1.3.4 relied on data collected on EVI240
and EVI147 cycles, we thought it appropriate to evaluate the extrapolation with power to high
levels beyond those covered by the I/M cycles. The development and application of adjustments
to rates in operating modes at high power is discussed in sub-section 1.3.3.5.

As mentioned, the rates described in 1.3.3 and 1.3.4 represent emission rates for vehicles under
the requirements of an inspect!on-and-maintenance program, specifically the program in
Phoenix, AZ, during calendar years 1995-2005. For this reason, we refer to these rates as "I/M
reference rates." With respect to the I/M reference rates, we describe the approach taken to
estimating rates in non-I/M areas, designated as the "non-I/M reference rates", in 1.3.3.6. For
runs representing areas without an I/M program, MOVES uses the non-I/M reference rates.  For
runs representing areas with I/M programs, MOVES adjusts the I/M reference rates to account
for the particular aspects  of the program(s) represented. It is important to note the I/M reference
rates assume full compliance with program requirements within the area. MOVES discounts
estimated emissions for non-compliance during a model run, which is then represented in the
results34.

We have observed, as have other researchers, that emissions deterioration tends to follow
exponential, or log-linear trends over the first 8-9 years.  However, after this point, the trends
enter a declining phase, during which increases in mean emissions continue at a reduced rate. For
the I/M reference rates, we assume that rates  stabilize between 12 and 15 years of age. For the
non-I/M reference rates, we assume that they continue to increase at reduced rates through 20+
years of age. The analyses guiding these assumptions are described in 1.3.3.7.

For start emissions, we also applied different methods to different datasets to derive two sets of
rates. For vehicles manufactured in 1995 and earlier, the process of rate development is
described in 1.4.1. For vehicles manufactured in 1996 and later, the process of rate development
is  described in 1.4.2. We assume that emissions deterioration affects start as well as running
emissions. Sub-section 1.4.3 describes how we estimate deterioration in start emissions in
relation to deterioration in running emissions.

-------
1.2    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
pollutants are the gaseous criteria pollutants: total hydrocarbons (THC), carbon monoxide (CO)
and oxides of nitrogen (NOx). 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 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."

Note that this document describes only emission rates for exhaust hydrocarbons. Modeling of
emission rates for evaporative hydrocarbons is described in a separate report: Development of
Evaporative Emissions Calculations for the Motor Vehicle Emissions Simulator15.

For these pollutants and processes emissions sources include light-duty vehicles (cars and
trucks). Note that the engine-size and weight-class attributes are not used to classify vehicles.
For light-duty vehicles, these parameters are assumed not to influence emissions, as these
vehicles are required to meet applicable standards irrespective of size and weight.

In the emissionRateByAge table, the emissions source is described by a label known as the
"sourceBinID".  This identifier is constructed as a pattern variable incorporating the attributes
shown in Table 2. Assignment of the attributes just described allows assignment of the source-
bin identifier. The identifier is a 19-digit numeric label, of the form "\fftteeyysssswwwwQQ"
where each component is defined as follows:

    1 is the literal value "1," which serves as a leading value to set the magnitude of the entire
        label,
    ^represents the fueltypelD,
    tt represents the engTechID,
    ee represents the regClassID,
    yy represents the shortModYrGrpID,
    ssss represents the  engSizelD,
    wwww represents the weightClassID, and
 00 is the literal value "00," which serves to provide two trailing zeroes at the end of the label.

The individual attributes are assembled in the proper sequence by constructing the sourceBinID
as a pattern variable, where

-------
                         sourcebinID=lxl018
                                     + fuelTypeIDxl016
                                     + engTechID xlO14
                                     + regClassID xlO12
                                     + shortModYrGroupID x 1010
                                     + engSizeID  xlO6
                                     + weightClassIDxl02
Equation 1
As an example, Table 3 shows the construction of sourceBin labels for light-duty gasoline
vehicles, manufactured in model years 1998 and 2010.
Table 1.  Combinations of pollants and processes for gaseous pollutant emissions.
pollutantName1
HC
CO
NOX
pollutantID1
1
2
o
J
processName2
Running exhaust
Start exhaust
Running exhaust
Start exhaust
Running exhaust
Start exhaust
processID2
1
2
1
2
1
2
polProcessID3
101
102
201
202
301
302
Section






1 as shown in the database table "pollutant."
2 as shown in the database table "emissionProcess."
3 as shown in the database table "emissionRateByAge."
Table 2.  Construction of sourceBins for Exhaust Emissions for light-duty vehicles.
Parameter
Fuel type
Engine Technology
Regulatory Class
Model-Year group
Engine Size Class
Vehicle Test Weight
MOVES Database Attribute1
fuelTypelD
engtechid
regClassID
shortModYrGroupID
engSizelD
weightClassID
Values
Gasoline = 01
Diesel = 02
Ethanol =05
01= "Conventional internal
Combustion"
20 = "Car" (LDV)
30 = "Truck" (LOT)
Varies2


1 as used in the database table "emissionRateByAge."
2 as defined in the database table "modelYearGroup."
                                            10

-------
Table 3 Examples of sourceBinID construction for cars and trucks in model years 1998 and 2010.
fuelTypelD
1 (Gasoline)
1
1
1
engTechID
1 (conventional)
1
1
1
regClassID
20 (Car)
30 (Truck)
20 (Car)
30 (Truck)
shortModYrGroupID
98 (MY 1998)
30 (MY 20 10)
98 (MY 1998)
30 (MY 20 10)
sourceBinID
1010120980000000000
1010130980000000000
1010120300000000000
1010130300000000000
1.2.1  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 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 are expressed as coefficients of
variation for the mean (meanBaseRateCV); this term is synonymous with the "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.
1.2.1.1
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. These groups assume that the
most rapid change in emissions as vehicles age occurs between 4 and 9 years.
                                          11

-------
Table 4. Description of the EmissionRateByAge Table.
Field
SourceBinID
PolProcessID
opModelD
ageGroupID
meanBaseRate
meanBaseRateCV
meanBaseRatelM
meanBaseRatelM
CV
dataSourcelD
Symbol




p
•^cell
cv-E



Description
Source Bin identifier. See Table 2
and Table 3 and Equation 1.
Combines pollutant and process. See
Table 1.
Operating mode: defined separately
for running and start emissions. See
Table 5.
Indicates age range for specific
emission rates. See 1.2.1.1.
Mean emission rates in areas not
influenced by inspection and
maintenance programs.
Coefficient of variation of the cell
mean (relative standard error, RSE),
for the meanBaseRate.
Mean emission rate in areas subject
to an I/M program with features
similar to the Phoenix program . See
1.3.7.
Coefficient of variation of the cell
mean (relative standard error, RSE),
for the meanBaseRatelM.
Numeric label indicating the data
source(s) and method(s) used to
develop specific rates.
                                              12

-------
1.3    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.

1.3.1        Operating Modes (opModelD)

For running emissions, the key concept underlying the definition of operating modes is "vehicle-
specific power" (VSP, /V )•  This parameter represents the tractive power exerted by a vehicle to
move itself and  its cargo or passengers16.  It is estimated in terms of a vehicle's speed and mass
(commonly referred to as weight), as shown in Equation 2
                                    Avt + Bvf + Cvf + mvtat
                              Pv t = - i - i - '- - t—t-                     Equation 2
                                             m

In this form, VSP (Pv,t, kW/metric ton) is estimated in terms of vehicles' :
       •  speed at time t (vt, m/sec),
       •  acceleration at (m/sec2) ,
       •  - mass m (metric ton) (usually referred to as "weight, "),
       •  - track-road load coefficients A, B and C3, representing rolling resistance, rotational
          resistance and aerodynamic drag, in units of kW-sec/m, kW-sec2/m2 and kW-sec3/m3,
          respectively.

This version of the equation does not include the term accounting for effects of road grade,
because the data used in this analysis was measured on chassis dynamometers. Note that during
model operation, MOVES does account for grade when characterizing vehicle activity. For a
description of this process, see the "Vehicle Population and Activity" report17.

On the basis of VSP, speed and acceleration,  a total of 23 operating modes are defined for the
running-exhaust process (Table 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/metric ton. For reference, each mode is identified by a numeric label, the
"opModelD."
                                           13

-------
Table 5. Definition of MOVES Operating Modes for Running-Exhaust operation.
Operating
Mode
0
1
11
12
13
14
15
16
21
22
23
24
25
27
28
29
30
33
35
37
38
39
40
Operating Mode
Description
Deceleration/Braking
Idle
Coast
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Coast
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Cruise/ Acceleration
Vehicle-Specific
Power
(VSPf, kW/metric
ton)


VSP,< 0
0 
-------
1.3.2  Scope

In estimation of energy consumption for MOVES2004, it was possible to combine data from
various sources without regard for the residence locations for vehicles measured. In contrast,
when turning attention to the regulated gaseous pollutants, it is essential to know with some
degree of confidence whether vehicles had been subject to inspect!on-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. This
approach is an inversion of that used in MOBILE6, in which "non-I/M" is the "default
condition" relative to which "I/M" emissions are calculated during a model run.

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.
1.3.3  Emission-Rate development:  Subgroup 1 (Model years through 2000)

1.3.3.1       Data Sources

For emissions data to be eligible for use in MOVES development, 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., e.g., continuous or
          "second-by-second" measurements.
       •  To make allowance for application of temperature adjustments (developed
          separately), it was necessary to know the temperature at the time of test.
       •  Vehicles were subject to I/M program requirements at the time of measurement.


1.3.3.1.1            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 6.
                                           15

-------
Table 6. Required Vehicle Parameters.
Parameter
VIN
Fuel type
Make
Model
Model year
Vehicle class
GVWR
Track road-load power
Units






Ib
hp
Purpose
Verify MY or other parameters



Assign sourceBinlD, calculate age-at-test
Assign sourceBinlD
Distinguish trucks from cars (LDV)
Calculate track road-load coefficients^, B and C
1.3.3.1.1.1
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 318.
                                           TRLHP.c,

                                          ^   V50'C2
                                          TTRLHP-C,
                                                                               Equation 3
                                  --PIT    TRLHP-c.
                                  ^  -L -L n \   /      \->
where:
    PF^ =  default power fraction for coefficient^ at 50 mi/hr (0.35),
    PF# =  default power fraction for coefficient B at 50 mi/hr (0.10),
    PFC =  default power fraction for coefficient C at 50 mi/hr (0.55),
    ci = a constant, converting TRLP from hp to kW (0.74570 kW/hp),
    v50 = a constant vehicle velocity (50 mi/hr),

    c2 = a constant, converting mi/hr to m/sec (0.447 m-hr/mi-sec)).

In the process of performing these calculations, we converted from English to metric units, in
order to obtain values of the track road-load coefficients in SI units, as listed above. Values of
TRLP were obtained  from the Sierra I/M Look-up Table.19
                                           16

-------
1.3.3.1.2
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 7.
Table 7. Required Test Parameters.
Parameter
Date
Time of day
Ambient temperature
Test Number
Test duration
Test result
Test weight
Units


°F

sec
pass/fail
Ib
Purpose
Determine vehicle age at test
Establish sequence of replicates
Identify tests on target temperature range
Identify 1st and subsequent replicates
Verify full-duration of tests
Assign tests correctly to pass or fail categories
Calculate vehicle-specific power
1.3.3.1.3
Candidate Data Sources
In addition to the parameters listed in Table 6 and Table 7, 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.

At the outset, a large volume of emissions data was available, representing over 500,000 vehicles
when taken together (Table 8).  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 8. Datasets available for use in estimating Running emissions from cars and trucks.
Dynamometer
I/M
AZ (Phoenix)
IL (Chicago)
MO (St. Louis)
NY (New York)



non-I/M







Remote-Sensing (RSD)
I/M
AZ Phoenix
IL (Chicago)
MO St. Louis
Maryland/N Virginia
GA (Atlanta)


non-I/M



VA (Richmond)
GA (Augusta/Macon)
NE (Omaha)
OK (Tulsa)
                                            17

-------
Several remote-sensing datasets received consideration.  However, we elected not to use remote-
sensing data directly to estimate rates, for several reasons: (1) For the most part, the RSD
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
RSD 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 HC measurements by other methods.20 (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 data21, mass rates cannot be calculated without an independently
estimated CC>2 mass rate. It followed that RSD would not provide rates for any MY* Age
combinations where dynamometer data were not available. In these cases, RSD 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 describe emissions only selected cruise/acceleration operating modes. However, RSD
cannot provide measurements for coasting, deceleration/braking or idle modes. For these
reasons we reserved the RSD for additional roles, such as verification of results obtained from
dynamometer data.

Table 9. Characteristics of candidate Datasets

Type
Network
Exempt MY
Collects random
sample?
Program Tests
Fast-pass/Fast-fail?
Test type (for
random sample)
Available CY
Size (no. tests)
Chicago
Enhanced
Centralized
4 most recent
YES
Idle, IM240, OBD-II
YES
IM240
2000-2004
8,900
Phoenix
Enhanced
Centralized
4 most recent
YES
Idle/SS, IM240,
IM147, OBD-II
YES
IM240, IM147
1995-1999
2002-2005
62,500
NYIPA
Basic/Enhanced
De-centralized
2 most recent
n/a
IM240
n/a
IM240
1999-2002
8,100
St. Louis
Enhanced
Centralized
2 most recent
NO
IM240
YES
n/a
2002-2005
2,200,000
Dynamometer datasets that received serious consideration are described below and summarized
in Table 9.

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 EVI240 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 EVI240 tests in rapid succession, obviating concerns about
conditioning prior to conduction of EVI240 tests. A second is the "full-duration" random sample,
in which selected vehicles received a single full-duration EVI240. 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
                                           18

-------
light on age trends in emissions. Both samples were 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,
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%
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% 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
EVI240 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.
1.3.3.2        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

                                            19

-------
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
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% 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.
1.3.3.2.1            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 5.  Within both strata, sample vehicles
then received three replicate EVI147 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 4.
                        f     _  f,MY,CY        f      _   p,MY,CY
                       AMY,CY - ~ -       /P,MY,CY - ~ -                Equation 4
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 5.
                                           20

-------
                        w
                                 A,MY,CY
                                               W
                                                p.MY.CY
                                          /p.MY.CY
                                                                                Equation 5
Figure 5. Stratified sampling as applied in selection of the random evaluation sample in the Phoenix I/M
Program (CY 2002-05).
               Official Test
               MY 1980 and prevails. Loaded-mode * Idle
               MY 1981 -1995: IMU7
               MY 1996 and later 080 I!
   Failing Stratum
   Oversarnpled
   "higher" sampling rate
                                 Dassing Stratum
                                 "lower" sampling rate
                          Tripncace IM147
1.3.3.3
Source selection
After excluding the St. Louis dataset, and comparing the Phoenix, Chicago and NY datasets,
analysis, 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 9 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
                                           21

-------
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.

Finally, the selection of the Phoenix data provided a relatively consistent basis for specification
of a "reference fuel," and development of associated fuel adjustments34.
1.3.3.4
Methods
1.3.3.4.1
       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 5).  The model-year groups used are shown in Table 7, along
with corresponding samples of passing and failing tests.
Table 10. Test sample sizes for the Phoenix random evaluation sample (n = no. tests)
Model-year
group1

1981-82
1983-85
1980-89
1990-93
1994-95
1996-98
1996
1997-98
1999-2000
Total
Cars
fail2
562
1,776
3,542
2,897
997
1,330


176
11,285
pass
539
2,078
6,420
8,457
4,422
3,773


753
26,478
Trucks
fail
340
1,124
1,745
1,152
703

526
858
136
6,589
pass
495
1,606
3,698
4,629
3,668

1,196
2,320
624
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, HC or NOx, as applicable.
                                           22

-------
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 '/z'.
1.3.3.4.1.1          Rates: Calculation of weighted means

The emission rate (meanBaseRate) in each cell is a (Eh) simple weighted mean
                                      Eh = — -                             Equation 6
where wt is a sampling weight for each vehicle in the cell, as described above, and Rt:t is the
"second-by-second" emission rate in the cell for a given vehicle at a given second t.
1.3.3.4.1.2          Estimation of Uncertainties for Cell Means:

A new feature of MOVES is its ability to estimate uncertainty in emissions projections. In the
emissionRateByAge table, uncertainties for individual rates are stored in the
"meanBaseRateCV" fields (Table 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 variability22.  To enable estimation of variances under this approach, we
calculated a set of summary statistics, as listed below:

Test mean (Ei): 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 (n,): the number of measurements in a cell representing an individual
test on an individual vehicle.
                                           23

-------
Cell sample size («&,/•): 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").

Test variance (sf ): 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.

Weighted Between-Test variance component (si)', 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
                                                                                Equation 7
Weighted Within-Test Variance Component (s^): the variance component due to variability
within tests, or the variance of measurements within individual tests (Rj,t) about their respective
test means,  calculated in terms of the test variances, weighted and summed over all tests in the
cell:
                                                                                Equation 8
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.

                                           s2   s2
                                      SE  = -- \ — —                             Equation 9
                                           »*   »*,,•

Coefficient-of-Variation of the Mean (CV^/,): 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
                                                                               Equation 10
                                            24

-------
Note that the term CV^/, is synonymous with the term "relative standard error" (RSE).
1.3.3.4.2
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."  Empty cells occur for age Groups not covered by available data (Figure
6). In the figure, "age holes" are represented by un-shaded areas. Filling in these un-shaded areas
required "hind-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 EVI147 or EVI240, meaning operating
modes with power greater than about 24 kW/metric ton.
Figure 6. Model-year by Age Structure of the Phoenix I/M Random Evaluation Sample.
  MY                          Vehicle Age at Test (years)
                               8  9  10  11 12 13 14 15 16 17 18  19  20  21  22  23 24
1.3.3.4.2.2
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 described in section 1.5.5.

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/metric ton ( 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
                                           25

-------
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 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 5).  Overall, we fit
three models for each combination of cars and trucks, for the model-year groups shown in Table
10, 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 11). This sampling was performed to fit models on smaller volumes of data that
a standard desktop computer could handle. 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
vc 8^, calculated as

                           /   _   strat      ,..    _  -*-   _  'strat                  _    .
                           /strat - ~	 >     Wstrat  ~ ~7	 ~	                  Equation 11
                                 •^ * strat             /strat   "st
                                                         'strat
where «strat and Nstrat 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
5.

                                                                              Equation 12
                                           26

-------
Table 11. Sample sizes for statistical modeling, by regulatory class and test result.
Model-year group

1981-82
1983-85
1980-89
1990-93
1994-95
1996-98
1996
1997-8
LDV
fail
645
569
375
260
406
663


pass
554
631
828
944
1,995
1,738


LDT
fail
476
508
343
209
378

346
671
pass
723
691
856
991
2,021

854
1,730
Each model included two sub-models, one to estimate means and one to estimate variances, as
described below.
1.3.3.4.2.2.1   Coast/Cruise/Acceleration
Means model

For the means sub-model, the dependent variable was the natural logarithm of emissions
where :
In E  =
V + p2Pj + j93Pv3
p6Pvs + J1tl + £
                                                                              Equation 13
               = natural -logarithm transform of emissions (in cell /z),
       •  Py, Pv2, Py3   = first-, second- and third-order terms for vehicle-specific power
          (VSP, kW/metric ton),
       •  a  = vehicle age at time of test (years),
       •  s  = speed class (1 -25 mph, 25-50 mph and 50+ mph),
       •  t  = test identifier (random factor)
       •  e= random or residual error
       •  ft = regression coefficients for the intercept and fixed factors /V, « and s.
       •  y  = regression coefficients for the random factor test.

The model includes first-, second- and third-order terms in /Vto 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 In-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
                                           27

-------
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 InE1/,, i.e., the logarithmic variance
s/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 11). To get replicate variances, we calculated In-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
                              sf = a0+ a^a + a2Pv + a3Pva + e
Equation 14
where /V and a are VSP and age, as above, and a are regression coefficients. After fitting we
examined similar diagnostics as for the means model.
Model application

Application of the model was simple.  The first step was to construct a cell matrix including all
emission rates to be calculated, as shown in Table 12.
Table 12. Construction of emission -rate matrix for light-duty gasoline vehicles.


X
X
X
X
X
=
Count
1
2
10
21
7
3
9,660
Category
Fuel (gasoline)
Regulatory Classes (LDV, LOT)
Model-year groups
Operating modes
Age Groups
Pollutant processes (running HC,
CO, NOx)
TOTAL cells
MOVES Database attribute
fuelTypeID = 01
regClassID = 20, 30
As in Table 1 1
opModelD = 11-16, 21-30, 33-40
ageGroupID = 3, 405, 607, 809,
1014, 1519,2099
polProcessID=101,201, 301

                                           28

-------
Next, we constructed a vector of coefficients for the means sub-model (P) and merged it into the
cell matrix.
                       = (A, A A A A4 A(0-25) A(25-50) A(50+) A]
                                                                         Equation 15
Then, for each table cell, we constructed a vector of predictors (X/,).  Equation 16 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 13.
                                      v  Pv2Pv3a 1  00 Pv\
Equation 16
Table 13.  Values of VSP used to apply statistical models.
opModelD
11,21
12,22
13,23
14,24
15,25
16
27,37
28,38
29,39
30
40
33
35
Range
<0
0-3
3-6
6-9
9-12
12 +
12-18
18-24
24-30
30 +
30 +
<6
6-12
Midpoint
-2.0
-2.5
4.5
7.5
10.5
14.5
15.0
21.0
27.0
34.0
34.0
0.5
9.0
The final step was to multiply coefficient and predictor vectors, which gives an estimated
logarithmic mean (InE1/,) for each cell h.
                                                                                Equation 17
                                            29

-------
The application of the variances model is similar, except that the vectors have four rather than
nine terms

                                   a = [ao aia2(X3  ]                          Equation 18

                                   Xh = [ 1  Pv a Pva ]                         Equation 19

Thus, the modeled logarithmic variance in each cell is given by

                                        slh = XA$                               Equation 20

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 99-00 model-year group, the available data represented young vehicles without sufficient
coverage of older vehicles.  We considered it reasonable to adapt the age slope for the 96-98
model-year group for cars, and the 1997-98 model-year group for trucks.

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
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 13 to evaluate the model in
operating mode 24, using the age slope from the previous model-year group (/?4*) and an
estimate of In-emissions from the available dataset at the earliest available age (\nEa*) at age a*.
In operating mode 24, the midpoint of the VSP range (6-9) is 7.5 kW/metric ton and the speed
class is 25-50 mph.

               y8; = ln^-7.5A-7.52A-7.53y83-y8>*-A(25-5o)-V.5y86          Equation21

On a case by  case basis, age slopes were adopted from earlier or later model-year groups.  In a
similar way, In-variance models or estimates could be adopted from earlier or later model years.


1.3.3.4.2.2.2   Braking/Deceleration


Means model

We derived models similar to those used for coast/cruise/acceleration. 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:

                                           30

-------
                                                 y7tt + £                        Equation 22
Variances model
In addition, we fit variances models for these operating modes, which were also simple functions
of age.
                                     sf =(X0+ U^a + £                           Equation 23
Model Application
In these operating modes, rates were to be modeled for a total of 840 cells. This total is
calculated as in Table 12, 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 = [ A> A ]                             Equation 24

and

                                       XA = [ 1 a ]                             Equation 25
 respectively.

For the variances model the coefficients vector is

                                       | = [«„ «!  ]                             Equation 26

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 In-variance from a previous or later model-year group.  In model-year groups
where the purpose was to hindcast 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 \nEh calculated from the sample data for the youngest available age class. In this
case, Equation 27 is a rearrangement of Equation 22.

                                    /J0* = In Ea« - j34a *                           Equation 27

After these steps, the imputed values of InE1/, were calculated, as in Equation 19.
                                           31

-------
1.3.3.4.2.3
                     Estimation of Model Uncertainties
We estimated the uncertainty for each estimated \nEh in each cell. During each model run, we
saved the covariance matrix of the model coefficients (s/). This matrix contains covariances of
each of the nine coefficients in relation to the others, with the diagonal containing variances for
each coefficient.
                    CT
                   ^6.0
                                      ov,
                                       6,4
                                                                   °"0,4-
                                             2
                                           "5(0-251
                                                   ^5(25-
                                                      5-50)
                                                             2
                                                           "5(50+)
                                                                                Equation 28
Using the parameter vectors X/, and the covariance matrix
for each cell was calculated as
                                                         , the standard of error of estimation
                                      sinEh = XASpXA                            Equation 29

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 cell23. 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.
1.3.3.4.2.4
                     Reverse transformation
To obtain an estimated emission rate Eh in each cell, the modeled means and variances are
exponentiated as follows
                                         _    h
                                         — e   e
                                                                                Equation 30
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.
                                            32

-------
The estimate of In-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 In-transformed data.  A third option,
also based on available data, was an estimate calculated from averaged emissions data and the
mean and variance of In-transformed emissions data. This process involves reversing Equation
30 to solve for ss2. If the mean of emissions data is xa and mean of In-transformed data is xh then
the logarithmic variance can be estimated as
                                               -^                              Equation 31
                                             v.e  )

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 InE/,. 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 InE1/,, based on
a normal distribution with a mean of 0.0 and variance equal to s^. We applied Equation 30 to
reverse-transform each variate, and then calculated the variance of the reverse-transformed
variates.  This result represented the variance-of-the-mean for Eh(s2Eh), as in Equation 9. Finally,
we calculated the CV-of-the-mean (CV^) for each modeled emission rate, as in Equation 10.


1.3.3.4.3             Table Constru ction

After compilation of the modeling results, the subset  of results obtained directly from the data
(Equation 6 to Equation 10), shaded area in Figure  6) and the complete set generated through
modeling (Equation 13 to Equation 31) 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 (nh  > 3), and
(2) the C\Eh (relative standard error, RSE) of the data-driven Eh must be less than 50% (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
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 14.
                                           33

-------
Table 14. Mapping "Analytic" Model-year Groups onto MOVES database Model-year groups
"Analytic"
Cars
1981-82
1981-82
1983-85
1983-85
1986-89
1986-89
1990-93
1990-93
1994-95
1994-95
1996-98
1996-98
1996-98
1996-98
1996-98
Trucks
1981-82
1981-82
1983-85
1983-85
1986-89
1986-89
1990-93
1990-93
1994-95
1994-95
1996
1997-98
1997-98
1997-98
1997-98
"MOVES database"

1 980 and previous
1981-82
1983-84
1985
1986-87
1988-89
1990
1991-1993
1994
1995
1996
1997
1998
1999
2000
modelYearGroupID

19601980
19811982
19831984
1985
19861987
19881989
1990
19911993
1994
1995
1996
1997
1998
1999
2000
shortModYrGroupID

1
61
62
85
63
64
90
65
94
95
96
97
98
99
20
1.3.3.5
Verification and 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/metric ton for speed and VSP, respectively.
Some coverage 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
above were used to extrapolate up to about 34 kW/metric ton.

Based on initial  review and comment on this aspect of the analysis, we thought it advisable to
give 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 EVI cycles, namely, the US06 and the "Modal Emissions Cycle" or "MEC." Much of the data
was collected in the course of the National Cooperative Highway Research Program (NCHRP)24
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 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
                                          34

-------
                                                 \24
Comprehensive Modal Emissions Model (CMEM)  . Driving traces for the US06 and MEC
cycles are shown in Figure 7 and Figure 8.  Both cycles range in speed up to over 70 mph and in
VSP up to and exceeding 30 kW/metric ton.
Figure 7. Example Speed Traces for the US06 and MEC Drive Cycles.
 •g.
  £
  8.
70-


60;


50;


40-


30-


20-


10


 O-l
       0     200    400    600   800   1000   1200   1400   1600   1800  2000

                                 Time (sec)

                     Dms Cycle   ^"^ rose   — us06


Figure 8.  Example vehicle-specific power (VSP) traces for the US06 and MEC cycles.
      30:
      20:
      10
 0_    0:
     -10
     —20:
     —30-
             200    400    600   800   1000   1200   1400   1600   1800  2000
                                 Time (sec)

                     Drive Cycle   ^^~ rrec   —— usoe
                                               35

-------
Table 15 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 15. Sample sizes for US06 and MEC Samples (No. tests)
Model-year group

1 980 & earlier
1981-85
1986-89
1990-93
1994-95
1996-99
Total
Car
US06
4
15
21
54
49
58
201
MEC
14
23
24
57
45
28
191
Truck
US06

8
13
22
22
56
121
MEC
6
19
31
36
30
17
139
Total

24
65
89
169
146
159
652
Figure 9, Figure 10 and Figure 11 show trends in emissions vs. VSP for CO, HC and NOx for
LDV and LDT by model year group. Both cycles were averaged and plotted as aggregates.
Figure 9.  CO emissions (g/sec) on aggressive cycles , vs. VSP, by regulatory class and model-year group.
 ? 5
 •a
       Reg/MYG
                 1 LDT—0080
                 ' LDT—9435
                  LEV— 86*3
     20

 VSP (kW/tonne)

 LDT-6185  ——
' LOT-9639  — —
 LEW— SOSO  — —
' LDT—
 LDv-ooeo
 LDV—9496
LDT-3393
LCV-8186
LCV-9S93
                                             36

-------
Figure 10. THC emissions (g/sec) on aggressive cycles , vs. VSP, by regulatory class and model-year group.
    0.13
    0.12
    0.11
    0.10
    0.09
    0.08
    0.07
    0,06
    0.05
    o,04
    O.C0
    0.02
    0.01
    0.00
                           10        20        30
                                VSP (kW/tonne)
       Reg/MYG
                  ' LDT-OC8D
                  ' LDT-9495
                   LEV-9689
 ' LOT-8185
 1 LDT-9S9S
  LCW-93S3
 ' LDT-8683
  LEW—0060
  LD₯-94S6
  LDT-9033
  LDV-8185
  LDV-96S9
Figure 11. NOx emissions (g/sec) on aggressive cycles, vs. VSP, by regulatory class and model-year group.
    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
    0.09
    0.07
    0.03
    0.05
    0.04
    O.OS
    0.02
    0.01
    0.00
                                   20
                               VSP (kW/tonne)
       Reg/MYG
                 • LOT—ooao
                 ' LDT-S4S5
                  LQ/— 8689
 LOT—8186
• LOT-mm
 LC₯— 9393
' LOT—8689
 LEW—0080
 LDV— 9-456
" LOT—9093
 LCy-8185
 LC₯—9699
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 15.  After averaging, we
calculated ratios from high-power operating modes to a selected reference mode. Specifically,
we selected two modes covered by the EVI cycles (27 and 37) to serve as reference points. The
midpoint VSP for each is -15 kW/metric ton. With mode 27 as a reference, we calculated ratios
to modes 28, 29 and 30.
                                                 37

-------
                                      77
                                     -T- , f°r / = 28, 29, 30                      Equation 32
and with mode 37 as a reference, we calculated ratios to modes 38, 39 and 40.

                                      77
                               R,37 = -T- , for / = 3 8, 3 9, 40                     Equation 33
After calculating the ratios, we calculated ratio-based emissions estimates (ER) as the products of
their respective ratios and the initial rate for modes 27 or 37

                            J7R  _ P   Z7 initial     77-R _  r>   T? initial                         .
                            hh,, ~ R,-.2ihh,2i •> or  ^v - R,-3ihh,n                     Equation 34

respectively, where Eh'mtial 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 has changed substantially for the final rates. In the draft,
we calculated upper and lower confidence limits for ER and replaced the initial rate withER 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 EVI147. 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 on
page 33) 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 VSP 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.
                                            38

-------
In the THC example (Figure 12), 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 13). 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 HC. 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 14), 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,
whereas that for 30 falls above the upper confidence limit. In the final, the ratio is used more
sparingly, as in the HC 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 EVI147. 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 EVI147 overlaps somewhat the range of the aggressive cycles. For this reason, the degree
of extrapolation is lower and the power trends are similar.
                                           39

-------
Figure 12.  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.







n
la\ Draft





V
	 »f


c
/
/
/J
4



1


T
•/*
...**
        0  1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
      10-
           (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
   18
   17-
   16
   15:
_  14
€  13
21  12-
2  11
£  10
3.  9
tU  Q
vi  8
&  7-j
S  6

-  :
   3
   2-
   1
   0
           (c) Final: options selected
        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
                                                  40

-------
Figure 13.  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.

4,500 -











a) Draft






r
/


~J
^/ i
1
/-








]

I
L /
I *
&

0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33
5000
1
^ 3000
o>
1

-------
Figure 14. NOx 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.
    500
    450
    400
    350
    300
    250
    200
    150
    100
     50
      0
(a) Draft
                                  t
                             7/r
         0  1  11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
    _ 300
    £
    m
    |
(b)












Fi






lal






Of






)tic





i
i
ns



t

*
avail.












ible





i




i
i
I
• i •
, . » * *
4
k
1



1


4

1


»




•
,


















•
A
A , t




A B
•
t





         0  1  11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
                                 opModelD

   I
   01

             (c) Final: options selected
         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    ^ * * Model      : Ratio
                                                     42

-------
1.3.3.6        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 inspect!on-and-maintenance (I/M) programs.  In this regard a
fundamental difference between MOVES and MOBILE is that MOVES inverts MOBILE'S
approach to representing I/M.  In MOBILE, the emission rates stored in the input data tables
represent non-I/M conditions.  During a model run, as required, emissions for I/M conditions are
modeled relative to the original non-I/M rates.

In MOVES, however, two sets of rates are stored in the input table (emissionRateByAge). One
set represents emissions under "I/M conditions" (meanBaseRateEVI) 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. Development of the I/M reference rates
is discussed above in sections  1.3.3.2 and 1.3.3.3.  As the I/M references represent Phoenix, the
program characteristics implicitly reflected in them include:

   •   A four-year exemption period,
   •   transient tailpipe tests for  MY 81-95,
   •   OBD-II for MY 96+,
   •   Biennial test frequency.

In addition, the Phoenix program provides a relatively stable basis against which to represent
other program designs and for application of fuel adjustments.

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 Figure 15(a).  A second and
less common approach is to compare emissions between two groups of vehicles 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 Figure 15(b).

                                           43

-------
Figure 15. General approaches to estimating differences attributable to I/M programs: (a) comparison of
subsets of vehicles between two geographic areas, with and without I/M, and (b) comparison within a
program area.
  (a)  Comparison between a program .Area
       and a non-program area
(b) Comparison within a program
    A~ea
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. 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 16.

Table 16.  Criteria used to identify vehicles migrating into the Phoenix Program.
logic

OR
AND NOT
AND
AND
Criterion
The vehicle comes from
From a non-I/M county
from out-of-state
in AZ
From other I/M areas
Receiving very first test
in Phoenix program
Selected for 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
                                            44

-------
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 program25.

Figure 14 shows the distribution of incoming vehicles, by Census Region. Most vehicles
migrating to Phoenix came from the Midwest (47%), followed by the South (32%), the West
(20%) and the Northeast (1%). The low incidence from the NE may be attributable to the large
number of I/M programs in that region.

Figure 16. Geographic Distribution of Vehicles migrating into the Phoenix I/M area, 1995-2005.
To assess the 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 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.
       E
Ratio = —^
                                                                              Equation 35
For purposes of verification, we compared our results to previous work. An initial and obvious
comparison was to previous work based on an out-of-state fleet migrating into Phoenix that
provided a model for our own analysis7. 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 35.
                                           45

-------
Another valuable source for comparison was remote-sensing data collected in the course of the
                                                  0^ T7
Continuous Atlanta Fleet Evaluation (CAFE) Program ' .  Unlike our own analysis, this
program involves a comparison between two geographic areas. The "I/M area" is the thirteen-
county Atlanta area, represented by measurements for approximately 129,000 vehicles. The other
(the non-I/M area)  is the twelve-county non-I/M area, surrounding Atlanta, represented by
measurements for approximately 28,000 vehicles. Both areas have 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 RSD 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 17. 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 HC 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 HC, 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-DVI  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 on top of this, the second reason is that the ratios themselves increase with age
(Figure 15).  A third implication is the absolute differences would be smaller for successive
model-year groups as tailpipe emissions decline with more stringent standards.
                                           46

-------
Figure 17.  Non-I/M : I/M ratios for CO, HC and NOx for the Phoenix Area (this analysis) compared to
remote-sensing results for Atlanta and N. Virginia, and previous work in Phoenix (diamonds).
                       5-9
                    Age Class
10+
         i AZ I/M •'//, GA RSD (CY04) = VA RSD (CY04)
    1.80
                                  10+
         lAZI/M VGARSD(CY04) Z VA RSD (CY04)
    1.60
            0-4         5-9        10+
                    Age Class

      • AZI/M  -.GARSD(CY04)  - VA RSD (CY04)
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.
                                             47

-------
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
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 18. 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.

Figure 18. 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.
                                                          : 6-9,,
            1.0.
                      1.5
5,0
7.5
    MOVES AgeGroup
  Analysis AgeGroi p
o n
2 |3
IP
P 0-4 years
4
5
4-5 years
8 |7
8 |9
6-7 years 8-9 years
. ' -5 • 9 years ' ' • . '
Figure 19 shows final values of the non-I/M ratios for CO, THC and NOx, with error-bars
representing 95% confidence intervals.  The values for each pollutant start at 5.0% and increase
with age, stabilizing at maximum values at 6 years (for NOx) and 10 years (for HC and CO).
                                           48

-------
Figure 19.  Final non-I/M ratios for CO, HC and NOx, by MOVES AgeGroups, with 95%
confidence intervals.
    1.60
    1.40
    1.20
    1.00
    0.80
    0.60
    0.40
    0.20
    0.00
-(a) CO
           0-3
                  4-5
                          6-7
                                 8-9     10-14
                              Age Class
                                               15-19
                                                       20+
    0.00
           0-3
                  4-5
                          6-7     8-9    10-14
                              Age Class
                                               15-19
    1.60
    1.40
    1.20
    1.00
    0.80
    0.60
    0.40
    0.20
    0.00
-(c)NOx
           0-3     4-5     6-7     8-9    10-14
                              Age Class
                                               15-19
                                                       20+
                                                49

-------
The ratios shown in Figure 17 are applied to the I/M reference rates to derive non-I/M reference
rates.

                               Eh,non-VM = Ratio * Eh,uu                           Equation 36

The uncertainty in E/^non-i/M was calculated by propagating the uncertainty in the Ratio with that
of the corresponding I/M rate
                                                              'I/                Equations?
                       v2
                       "

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.

Figure 20 shows an example of the reference rates vs. operating mode, for all three pollutants.
Note that not all the modes are shown, to allow examination of differences between non-I/M and
I/M rates at lower VSP.  Figure 21 shows  corresponding trends by age for two operating modes.
The first is opmode 11, (speed = 1-25 mph, VSP <0 kW/metric ton) and 27 (speed = 25-50 mph,
VSP = 12-18 kW/metric ton). An clear observation from both plots is that the I/M difference is
much larger in the more aggressive mode  (27) than in the less aggressive one (11), with the
inference that I/M differences will be more strongly expressed for more aggressive than less
aggressive driving, in absolute (but not relative), terms.
                                           50

-------
Figure 20. Non-I/M and I/M Reference Rates by Operating Mode (Example: Cars, MY 1994, at 8-9 years of
age)
T Rn
€
5 50 -
V
s. m ~
§ 30 '
i5 70
m 10 -
n -
« I/M Reference •
• non-l/M Reference
(,_.\ Til/"*
a) I Hu i
m
: "


         0  1  11  12 13 14 15 16 21  22 23 24 25 27 28 29  30 33 35 37 38 39 40
                                     opModelD
:
2500 :
onnn
i^nn -
mnn
500 ;
o i
• I/M Reference
• non-l/M Reference *

(D) UU ^
5'
• • » '

           0  1  11 12 13 14  15 16 21  22 23 24 25 27 28  29 30 33 35 37 38 39 40
                                       opModelD
       300
    8.
    £  50




n
n
n I
* I/M Reference •
• non-l/M Reference «

• * * <
(c) NOx _ " f
• * •
* * . * • *
I • * ~ * T - T 	
          0  1  11 12 13 14 15  16 21 22 23 24 25  27 28 29 30 33 35  37 38 39 40
                                      opModelD
                                                 51

-------
Figure 21. Non-I/M and I/M Reference Rates vs. Age for Two Operating Modes (Example: Cars,
MYG 1994).
     40
     35
     30
     25
     20
 -•O---I/M Ref:opMode 11
 - D- - • non-l/M Ref: Mode 11
  *	I/M Ref: Mode 27
  •—non-l/M Ref: Mode 27
                             =H?
                                     ...D	Q	0-
                                .-••:.- -Q	- - - g-	o
             i	=0= = = =g= = ::
                                 10           15
                                    Age (years)
                                                          20
                                                                       25
     800
     700
     600
     500
     400
     300
     200
     100
       0
--«•-• I/M Ref: opMode11
- - D - • non-l/M Ref: Mode 11
—•	I/M Ref: Mode 27
—•—non-l/M Ref: Mode 27
                                            (b)CO
                                 10           15
                                    Age (years)
                                                          20
                                                                       25
   iu


120 =




20 :
n :

(c) NOx
<___— — — '
• — • — — _— — —
/S 	 » 	 *"

^^^"^
*
--D. -



n 	 -p 	 n 	 p -r 	 D 	 | 	 O-




I/M Ref: opMode 11
non-l/M Ref: Mode 1 1







	 n
                                 10
                                              15
                                                          20
                                                                       25
                                    Age (years)
                                                      52

-------
1.3.3.7        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 studies12. Figure 22 and
Figure 23 show age trends by model year for cars and trucks, respectively. The values shown are
aggregate mass rates over the EVI147 expressed as g/sec for CO, THC and NOx.  Incorporating
stabilization of emissions with age is another departure with the approach used on MOBILE,
which allowed emissions to increase indefinitely.

From  these figures, as well as Figure 1, it is clear that no data were available at ages older than
10 years of age for model years later than 1995, and that no data was available at ages older than
15 years for model years older than 1990.  Thus for model years more recent than about 1995 it
was necessary to project emissions for ages greater than 8-10 years.

However, it is not appropriate to simply extrapolate the statistical models past about 8-10 years.
As described above, emissions were modeled as In-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 15-19 year ageGroups, 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 24. 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.
                                                                              „    ,.  ~0
                                                                              Equation 38
                                                      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 (/?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 37, but did not propagate
the uncertainty through to the final result.
                                           53

-------
Table 17.  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

Cars
Cars
Trucks
Trucks
ageGroup

10-14
15-19
10-14
15-19
Ratios CRage)
THC
1.338
1.571
1.301
1.572
CO
1.226
1.403
1.220
1.479
NOx
1.156
1.312
1.156
1.312
Variances (VR)
THC
0.000000032
0.00000411
0.00000173
0.0000518
CO
0.000160
0.00268
0.000758
0.0666
NOx
0.00000009
0.00000261
0.00000138
0.0000499

                                                 54

-------
Figure 22. Aggregate IM147 Emissions (g/sec) for Cars, by Model year and Age, for the
Phoenix Random Sample.
                         LDV; WEIGHTED
                        CO vs. Age (years)
   0.20-
   013-
        (a) CO
                i-Mc-  O--B--B
                         LDV; WEIGHTED
                       ThC vs. Age (years), LDV
 O 000?
 I
        (b) THC
                         LDM WSGHTED
                       NOx \s. Age {years), LDV
        (c) NOx
                           Vehicle age (years)
                                               55

-------
Figure 23. Aggregate IM147 Emissions (g/sec) for trucks, by Model year and Age, for the Phoenix
Random Sample.
                         IDT WEIGHTED
                        CO vs. Age (years)
        (a) GO
                         LDT WEIGHTED
                      THC vs. Age (years), LDT
         (b) THC
                          \fehicte ags (years)
                         LDT WEIGHTED
                      KDx vs. Age (years), LDT
         (c)  NOx
                        » -K  , -

                    ,-   -'A'
                          \fehicle ag© (
                                               56

-------
Figure 24. Aggregate IM147 Emissions (g/mi) by Model-Year Group and Age Group.


      30.0
                         10      15

                       Age (years)
                                            20
25
                       10       15

                      Age (years)
                                           20
25
V
  co
  '«
  .2
    0.0
                                                        -•-81-82


                                                        -*- 83-85


                                                        ^-86-89


                                                        -W- 90-93


                                                        -•-94-95


                                                        -•-99-00


                                                        -»- 96-98
                                                        -•-81-82


                                                        -*-83-85


                                                        -K- 86-89


                                                        ^-90-93


                                                        -•-94-95


                                                        -•-99-00


                                                        -•-96-98
                                                            81-82


                                                            83-85
                                                              -93


                                                          •-94-95


                                                          •-99-00


                                                          ^96-98
0
5
10 15
Age (years)
20
25
                                        57

-------
1.3.3.7.2             non-I/M Reference Rates

The ratios developed in 1.3.3.7.1 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. The program places some pressure on high-emitting vehicles to improve
their emissions, leave the fleet, leave the area, or, it could be added, evade requirements in some
way. 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 18 shows the deterioration stabilization ratios for both the I/M and non-I/M references
rates. As mentioned above, all ratios are  applied by multiplication by values for the 8-9 year
ageGroup in all operating modes. The ratios for I/M areas (Rage,i/M) are identical to those in Table
17. The center column shows the ratio of values ofR^^yu for the 15-19 to the 10-14 year
ageGroups. Ratios for the non-I/M references (7?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.
                                           58

-------
Table 18. Deterioration-stabilization ratios as applied to I/M and non-I/M reference rates.
Pollutant
THC
CO
NOx
Regulatory
Class
Cars
Trucks
Cars
Trucks
Cars
Trucks
ageGroup
10-14
15-19
20+
10-14
15-19
20+
10-14
15-19
20+
10-14
15-19
20+
10-14
15-19
20+
10-14
15-19
20+
RageJM
1.338
1.571
1.571
1.301
1.572
1.572
1.226
1.403
1.403
1.220
1.479
1.479
1.159
1.312
1.312
1.159
1.312
1.312
Ratio (15-19: 10-14)

1.174


1.206


1.144


1.213


1.132


1.132

D
-**-age,non-I/M
1.338
1.571
1.845
1.301
1.572
1.898
1.226
1.403
1.606
1.220
1.479
1.795
1.159
1.132
1.486
1.159
1.132
1.486
Values in this column are identical to those in Table 17.
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.
                                                   59

-------
1.3.4  Emission-Rate Development: Subgroup 2 (MY 2001 and later)

1.3.4.1       Data Sources

Data for vehicles in model years 2001 and later was acquired from results of tests conducted
under the In-Use Verification Program. This program, initiated in 2003, is run by manufacturers
and administered by EPA/OTAQ through the Compliance and Innovative Strategies Division
(CISD).

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.
1.3.4.1.1
       Vehicle Descriptors
In addition to the parameters listed above in Table 7, the IUVP data provides engine-family
information. Using engine family, 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 HC, CO or NOx, where HC 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 1<
). Vehicle Descriptors available in IUVP files and certification test records.
Parameter

VIN
Fuel type
Make
Model
Model year
Engine Family
Tier
Emissions Standard
Units









Source
IUVP
Y
Y
Y
Y
Y
Y


Cert. Records


Y
Y
Y
Y
Y
Y
Purpose

Verify MY or other parameters



Assign sourceBinID, calculate age-at-test


Assign Vehicle Class
Combining data from both sources allows individual test results to be associated with the correct
standard level and emissions standard, which allows inference of the correct vehicle class.
1.3.4.2
Estimating I/M Reference Rates
The goal of this process is to represent 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 Section 1.3.4.2.1, below.
                                          60

-------
1.  Average IUVP results by standard level and vehicle class.

2.  Develop phase-in assumptions for MY 2001 - 2021, 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). Then calculate emissions
by operating mode in each model year by multiplying the MY2000 emission rates by the
weighted ratio for each model year.

4.  Estimate Emissions by Operating Mode. 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).

5.  Apply Deterioration to estimate emissions for three additional age Groups (4-5, 6-7 and 8-9).
We assume thatNLEV and Tier-2 vehicles will deteriorate similarly to Tier-1 vehicles, when
viewed in logarithmic terms. We therefore apply In-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 53.

6.  Estimate non-I/Mreference 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 (Figure 19).

Each of these steps is described in greater detail in the sub-sections below.
1.3.4.2.1            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 25
shows FTP composite results in relation to applicable certification and useful-life standards. For
THC and NOx, the data show expected compliance margins in the range of 40-60% in most
cases. For CO, compliance margins  are even larger, ostensibly reflecting the concomitant effects
of HC or NOx control on CO emissions.

Figure 25. Composite FTP Results for Tierl, NLEV and Tier 2 Vehicles, as measured by IUVP, in
relation to corresponding certification and useful-life standards.
                                           61

-------
Oonn
1 °-250 -
2 n onn
.w n 1 ^n
UJ
Q- n 1 nn
u_
n nnn
10.000
.~ 9.000
£ 8.000
•2 7.000
c 6.000
•5; 5.000
| 4.000
UJ 3.000
£ 2.000
"- 1.000
0.000
0.700 -
~ 0.600 -
S 0.500 -
(A
§ 0.400 -
'in
w 0.300 -
E



/a\ Lip
> \"/ riO - - Q - -Certification Standai


ds (g/mi)
\ > Useful Life Standards (g/mi)
\ '. 	 A 	 FTP Composite esti


AV b;:..
\ * . * f - .
nates (g/mi)









^v "'H'- •' 'a" "l-1" "n'--...

T1 JLEV LEV DLEV bin8 bin7 bin6 bin5 bin4
T1 NLEV NLEV NLEV T2 T2 T2 T2 T2











— A-~I_I_i
bin3 bi
T2 T


£
n2
2

(D) UCJ - - D - -Certification Standards (g/mi)
- - D - • Useful Life Standards (g/mi)
	 A 	 FTP Composite estimates (g/mi

nnn n n n n
n 	 n 	 D**. .-'.n 	 n- — n — •n/«.
* * ~"n'' ' " **-r\ -
± 	 tr U-
~~~— A . x



--B - --
*• 	 A 	 * 	 A 	 A 	 A 	 A 	 A-^
T1 JLEV LEV DLEV bin8 bin7 bin6 bin5 bin4
T1 NLEV NLEV NLEV T2 T2 T2 T2 T2
n fr^ NOv
- - Q - • Certification Sts

Useful Life Star
1-1 1-1 	 A 	 FTP Composite
\

bin3
T2

ndards (g/mi)
dards (g/mi)
estimates (g/




-A
bin2
T2


mi)

"•%
62

-------
Figure 26. Cold-start (Bag 1 - Bag 3) and Hot-running (Bag 2) FTP emissions for Tier 1, NLEV and
Tier 2 vehicles, as measured by IUVP  (g/mi).

^ 0.400 -
c Q '^'sn
"5)
w
w 0 POO

uj U.loU -
t n 1 nn


19 nnn

— 1 o 000
s a ooo
M o.uuu
c
.2 R nnn
_(/)
IM 4 nnn
t 9 nnn
0.000
1 nnn -,

~" n pnn
~5) n vnn
52 n finn -
.2 r\ cnn


H n 9nn
01 nn
n nnn -


* cold start (g/mi) — • — Hot Kunnmg (g/mi)


»-~~"'~~^\

/ \ lip \ Jt 	 ^ 	 ^ 	 ^^
v a/ ^ \^^^ ^^^^*-^-^-^
•^ *^^

T1 TLEV LEV ULEV binS bin? bin6 binS bin4 binS bin2
T1 NLEV NLEV NLEV T2 T2 T2 T2 T2 T2 T2
^.^^^
^v > Cold start (g/mi) — • — Hot Running (g/mi)
\.
(bO CO ^v ^^*^-
\yj 1 ^**S \ 	 ^~-~+ » » 
-------
Figure 27 Composite, Cold-start (Bag 1 - Bag 3) and Hot-running (Bag 2) FTP emissions for Tier 1,
NLEV and Tier 2 vehicles, as measured by IUVP, normalized to respective Tier-1 levels.
      1.000
      0.900
      0.800
      0.700
      0.600
      0.500
      0.400
      0.300
      0.200
      0.100
      0.000
    0)
   H
    O
   +*
    O
   '•»
   a:
1.000
0.900
0.800
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.000
                                         Cold Start—*—Hot Running     Composite
                                               64

-------
Figure 26 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
rate, in g/mi. Additionally, Figure 23 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.
1.3.4.2.2            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
     ^.f\
years  .  These records contain information on certified vehicles, including model year, 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 20.

After compling lists of engine families by standard, model year and vehicle class, we obtained
estimates of final sales from the EPA VERIFY database for MY 2001-200729. 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 21. Sales-weighted
phase-in scenarios for each vehicle class are shown in Figure 28 through Figure 31.  As noted, the
results in the Figures reflect the certifications in the "Fed"  or "Both" groups shown in Table 21.

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.
                                           65

-------
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.

Table 20. Examples of Information obtained from Certification Test Records, with Vehicle Class inferred
from combinations of Standard, and FTP Certification values.
Standard

LEV
LEV
Tierl
Tierl
Engine Family

2HNXV02.0VBP
2MTXT02.4GPG
2CRXT05.95B2
2CRXT05.96BO
Sales Area

NLEV all states
NLEV all-states
Federal all-altitude
Federal all-altitude
FTP Standard
50,000-mi
0.075
0.100
0.32
0.39
100,000-mi
0.09
0.13


120,000-mi


0.46
0.56
Vehicle-Class

LDV, LDT1
LDT2
LDT3
LDT4
Table 21. Approximate Numbers of Engine Families Certified, by Model Year and Age Group, Model Years
2001-2007.
Sales Area

California
Clean Fuel Vehicle
California + NLEV (all
states)
Federal All Altitude
Federal + CA Tier 2
Clean Fuel Veh +
NLEV(ASTR)2
+ CA
NLEV (All States)
TOTAL
Code

CA
CF
CL
FA
FC
NF
NL

Group1

CA
Fed
Both
Fed
Both
Both
Fed

Model Year
2001
114
38
149
79

57
31
468
2002
116
46
140
75

56
47
480
2003
118
81
129
86
16
45
74
549
2004
240
76

209
81


606
2005
251
69

219
41


580
2006
275
61

271
33


640
2007
255
55

274
16


600
Total

1,369
426
418
1,213
187
158
152
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 orLEV-II standards, including the "section 177" states, "Both" denotes vehicles certified for
Federal or California Sales Areas.
2. "ASTR" = "All-state trading Region."
                                           66

-------
Figure 28. Phase-in Assumptions for Tier 1, NLEV, and Tier 2 standards, for LDVand  LDT1




"c
0)
0)
0_
30% -

n°/n -


























•

1






• _








•
•


















































































• Tier-2(Bin2)
• Tier-2(Bin 3)
Tier-2(Bin 4)
• Tier-2(Bin5)
• Tier-2(Bin 6)
• Tier-2(Bin7)
LEV-II (ULEV)
• LEV-II (LEV)
NLEV(ULEV)
NLEV(LEV)
• NLEV(TLEV)
Tier!


                                                       t-t-t-
           CMCMCMCMCMCMCMCMCMCMCMCMCM
                                 Model Year
Figure 29.  Phase-in Assumptions for Tier 1, NLEV and Tier 2 standards, for LDT2.
     100%

      90%

      80%

      70%

      60%
  •s
  a>
  o   50%
  a>
  a.
      40%

      30%

      20%

      10%

      0%
ITier-2(Bin3)
 Tier-2(Bin 4)
ITier-2(Bin 5)
ITier-2(Bin7)
ITier-2(Bin8)
 NLEV(ULEV)
 NLEV(LEV)
I NLEV(TLEV)
 TieM
           CM    CM    CM    CM
                                   CM    CM    CM
                                    Model Year
                                                  CM   CM   CM    CM
                                                  67

-------
Figure 30.  Phase-in Assumptions for Tier 1 and Tier 2 standards, for LDT3
90% -
80% -
70% :
60% :
'c
a)
o 50% -

-------
1.3.4.2.3            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 32.

Figure 33 shows an example of the Phase-in calculation for NOx 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 22 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 (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 34 to Figure 36 below.
                                           69

-------
Figure 32.  Relative Fractions of Truck Classes, by Model Year.



60% -
1 '
0)
0_






























	












































































































ST-CslCO^LOCDr^CQO)Ot-
-------
Table 22. Weighted Average FTP Values Projected for Trucks and Cars for MY 2001-2010.
regClass

Trucks










Cars










MY

2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
CO
Composite
(g/mi)
2.28
1.43
1.41
1.47
0.92
0.78
0.70
0.66
0.65
0.63
0.62
1.62
0.856
0.821
0.808
0.714
0.672
0.657
0.621
0.611
0.601
0.591
Start
(g)
17.90
12.56
12.40
12.73
7.92
7.05
6.12
5.85
5.75
5.67
5.58
11.40
7.68
7.27
7.05
6.16
5.91
5.85
5.63
5.55
5.47
5.38
Running
(g/mi)
1.01
0.566
0.552
0.586
0.393
0.315
0.296
0.281
0.270
0.260
0.251
0.805
0.287
0.284
0.299
0.298
0.274
0.257
0.234
0.232
0.231
0.229
HC
Composite
(g/mi)
0.175
0.0965
0.0941
0.100
0.0535
0.0440
0.0378
0.0361
0.0356
0.0350
0.0345
0.126
0.0361
0.0333
0.0340
0.0356
0.0358
0.0350
0.0341
0.0341
0.0339
0.0339
Start
(g)
1.87
1.23
1.21
1.25
0.786
0.703
0.612
0.587
0.580
0.571
0.564
1.53
0.954
0.893
0.839
0.664
0.634
0.633
0.608
0.592
0.574
0.557
Running
(g/mi)
0.104
0.0400
0.376
0.0424
0.0123
0.00574
0.00511
0.00490
0.00479
0.00470
0.00462
0.0571
0.00509
0.00451
0.00462
0.00488
0.00477
0.00462
0.00443
0.00443
0.00442
0.00442
NOx
Composite
(g/mi)
0.304
0.171
0.169
0.181
0.0849
0.0596
0.0381
0.0315
0.0285
0.0258
0.0233
0.209
0.0948
0.0898
0.0824
0.0461
0.0351
0.0335
0.0271
0.0248
0.0224
0.0201
Start
(g)
1.12
0.843
0.836
0.863
0.473
0.367
0.264
0.226
0.208
0.192
0.177
0.888
0.586
0.573
0.530
0.315
0.248
0.239
0.201
0.187
0.172
0.158
Running
(g/mi)
0.174
0.0876
0.0865
0.0934
0.0434
0.0291
0.0183
0.0148
0.0130
0.0115
0.0101
0.127
0.0457
0.0421
0.0394
0.0220
0.0161
0.0150
0.0112
0.0101
0.00896
0.00784
                                               71

-------
Figure 34. Weighted Ratios for Composite, Start and Running CO Emissions, for (a) Trucks and (b) Cars.
            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
                                                  72

-------
Figure 35.  Weighted Ratios for Composite, Start and Running THC Emissions, for (a) Trucks and (b) Cars.
   1.0
   0.9 -
   0.8 -
°  „ -,
*=  0.7
ro
&  0.6 H
1  0.£
a 0.4 H
'5  n'
Sa2H
   0.1
   0.0

                  vs.
                                  	
                                        0.118
                  n       '       '      '       '  0.055  ' 0.049 '  0.047  '  0046  '  0045  ' 0044
             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
                                                  73

-------
Figure 36. Weighted Ratios for Composite, Start and Running NOx Emissions, for (a) Trucks and (b) Cars.
                                                                       U.UDD  i  U.Ubb
            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
                                                74

-------
1.3.3.2.4     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). To project emissions for NLEV and Tier-2
vehicles, we divided the operating modes for running exhaust into two groups. These groups
represent the ranges of speed and power covered by the FTP standards (< -18 kW/metric ton),
and the ranges covered by the US06 cycle. For convenience, we refer to these two regions as
"the hot-running FTP region" and "US06 region," respectively (See Figure 37). Data measured
on the SC03 cycle did not play a role in emission rate development.

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 the FTP operating modes, we applied the ratios shown in
Figure 34 to Figure 36 above.

1.3.4.2.4.1           Running Emissions

For the "US06" operating modes, we followed a different approach from that described above in
1.3.4.2.3. At the outset, we noted that the degree of control in the FTP standards increases
dramatically between MY 2000 through MY 2010, following phase-in of the Tier-2 standards,
giving pronounced declines in emissions on the FTP. For our purposes, we are referring
specifically to declines in running emissions, as shown by changes in Bag-2 emissions. However,
it was not obvious that the degree of control would increase as dramatically for the  SFTP
standards, as shown by the US06. Thus, in preparation of the draft rates, we adopted a
conservative assumption that emissions in the "US06" region would not drop as sharply as those
in the "hot-running FTP" region.

It was therefore necessary to estimate different sets of ratios. Two alternative approaches were
developed.

The first involved returning to the Phoenix I/M data.  To create pre- and post-SFTP estimates, we
pooled tests for two model-year groups, 1998-2000, representing Tier 1 vehicles not subject to
SFTP requirements, and 2001-2003, representing NLEV vehicles subject to the SFTP.  For each
group, we calculated means for each pollutant for the US06 operating modes (as a group), and
calculated ratios between the two groups.

                                         77
                                   n      -c/poll,SFTP,01-03
                                  /VgFTp = -=	                         Equation 39
                                         -^poll.SFTP, 98-00

The resulting ratios were used for CO and HC, as shown in Figure 38.

The second approach  involved compilation of results on the US06 cycle and calculation of ratios
in a manner similar to that used for FTP data as described in 1.3.4.2.3 above.  It was possible to

                                          75

-------
obtain data representing US06 tests representing vehicles in MY 1996-97 from the Mobile-
Source Observation Database (MSOD), developed and maintained by EPA/OTAQ.30  For NLEV
and Tier-2 vehicles manufactured after MY2000, US06 results were available from the IUVP
program. Emissions results on the US06 by standard were weighted by the phase-in assumptions
for MY 2001-2007 as with the FTP results. Resulting ratios were used for NOx, as shown in
Figure 38.

Figure 39 and Figure 40 show application of the ratios to the hot-running FTP and US06
operating modes in model years 2000 (the reference year), 2005, and 2010, both calculated with
respect to 2000. The sets of ratios shown in Figure 38 are used for both sets of modes. Note that
the values for the SFTP modes are equal in 2005 and 2010 for HC and CO, because the SFTP
ratios are constant by model year. 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 between MY 2005 and 2010 in the lower power modes.
                                          76

-------
Figure 37. Operating modes for running Exhaust Emissions, divided broadly into "hot-running FTP" and
"US06" regions.


30 +
| 27-30
2 24-27
1 21-24
fc 18-21
£ 15-18
Jg 12-15
£9-12
i«
•S 3-6
0^3
<0
Speed Class
7-25 ,^25-50
16



30
29
28

27
^H
15 25
14 24
13 23
12 22
11 21
(mph)
50 +
40
jg^x
38

37

35
« 	

33
. 	 •*


The "US06 Region"
' High Power





«


The "hot-running FTP Region"
Low to Moderate
Power
(also includes braking (0)
and idle (1)


                                               77

-------
Figure 38.  Weighted Ratios for Cars, for hot-running emissions, representing the "hot-running FTP Region''
(FTP) and the "US06 Region" (US06), for (a) CO, (b) THC and (c) NOx.
                     (a) CO
                            0.551   0.550   0.550   0.550   0.550   0.550   0.550   0.550
                                        0.340   0_320
                                                    0.290   0.288   0.286  Q.284
          2000  2001  2002  2003  2004  2005 2006  2007  2008  2009  2010
                                     Model Year
                     (b) THC
                                         0.347  0.347   0.345   0.345   0.345   0.345
      0.2
      0.1
      0.0
0.089   Q.Q79   O.Q81   0.085  0.084  0.081   0.078   0.078   0 n?7   n (177
           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
                                                       78

-------
Figure 39. Projected Emission Rates for Cars, by Operating mode for ageGroup 0-3 years, for three model
years (LINEAR SCALE).
1 ftnn

D) -| 4QQ

ik -i nnn ^
c -
15 fioo ^

onn


zuuu
• 2005
D2010

/0\ ^>o
(a) uu
•




4
•

k C
• •••••^•nfiftfti*i"
±±
n n n fi 5
           0  1  11  12 13 14 15 16 21  22 23 24 25 27 28 29 30 33 35 37 38 39 40
                                 Operating Mode
     18
     16 ^
     14
     10
  'in   6
  in
  E   4
      2 -.
  LLI
»2000
• 2005
D2010
(b)THC
                                      *  *
           *AJh
                                             ^ »
         0  1  11  12 13 14  15 16 21 22 23 24 25 27  28 29 30  33 35 37 38 39 40
                                Operating Mode
i^u :
-r^ -i nn
^)
— on :

c :
o
"in 40 :
in :
m 20 :
n :
*2000 «
• 2005 (C) NUX
D2010


*«*. ..•**.•••**.••
n n n rYn n n n ,rYri ,n n n D n r
          0  1  11  12 13 14 15 16  21 22 23 24 25  27 28 29 30 33  35 37 38 39 40
                                Operating Mode
                                              79

-------
Figure 40.  Projected Emission Rates for Cars, by Operating mode for ageGroup 0-3 years, for three model

years (LOGARITHMIC SCALE).
"C~ 1 nnn
aj inn
13 IUU ;
o: :
o m
u> :
UJ 1 ;
n
42000 (a) CO
• 2005 2
02010 «. n ,. •
* k * • n n
* * * * n * * * B * *
• a a a a ! a *
o a 11 12 13 14 15 16 21 22 23 24 25 27 28 29 3° 33 35 37 38 39 4°
                                   Operating Mode
•f
2 :
J3
(0
£ 1
0 :
'in
tn
E'~ 0 -
u !
UJ
n
:r ™c
D2010 * • • *
.»•*.. ***"•" .«•!•'
i i ^ i i i i i i i i i i i i i i i i i i i i
0 1 11 ft 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
p»H ••' ."
._
! H
H
                                  Operating Mode
  I
   tn
   tr>


  ill
1 nn
-i n
•i
n
n
*2000
2005
D2010

* * *
(c) NOx
4

• n n n L
w ' ' '•'r-il~l ' ' • ' ~ ' U 	
0 t 11 12 Q T4 15 16 21 £| S 24 25 27 28 29 30 33 35 37 38 39 40
• n n D
n •
n
                                   Operating Mode
                                             80

-------
1.3.4.2.5            Apply Deterioration

Based on review and analysis of the Phoenix I/M data, 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 natural logarithms 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 NLEV and 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.
1.3.4.2.5.1           Recalculate the logarithmic mean

Starting with the values of the arithmetic mean (xa) calculated above, we calculate a logarithmic
mean (x/), as shown in Equation 40. Note that this equation is simply a rearrangement of
Equation 30.

                                                 2
                                     Xj = In xa	                            Equation 40


The values of the logarithmic variance are intended to represent values for young vehicles, as the
estimates for xa represent the 0-3 year age Group.  The values of o/2used for this step were 1.30,
0.95 and 1.60 for CO, THC andNOx, respectively.
1.3.4.2.5.2           Apply a logarithmic Age slope

After estimating logarithmic means for the 0-3 age class (x/,o-3), we estimate additional
logarithmic means for successive age classes (x/,age), by applying a linear slope in In-space (mi).

                                *z,age = */,o-3 + mi (a§e -* •5)                       Equation 41

The values of the logarithmic slope are adapted from values developed for the 1996-98 model -
year group. The values applied are shown in Table 23 . 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,
as shown in Equation 41.

Figure 41 shows an example of the approach, as applied to THC from LDV in the 1996-98
model-year group.  The upper plot (a) shows InTHC vs Age, by VSP, where the VSP acts as a
surrogate for operating mode. The defining characteristics of the plot are a series of parallel
lines, with the gaps between the lines reflecting the magnitude of the VSP differences between
them. Similarly, the lower plot shows InTHC vs. VSP, by Age, where age acts as a surrogate for
                                           81

-------
deterioration. In this view, deterioration appears as the magnitude of the gaps between a family
of similar trends against power.

Table 23. Values of the logarithmic deterioration slope applied to running-exhaust emission rates for MY
following 2000.
pollutant
CO
THC
NOx
opMode Group
"hot-running FTP"1
"US06"2
"hot-running FTP"
"US06"
"hot-running FTP"
"US06"
Logarithmic slope (mj)
0.13
0.06
0.09
0.09
0.15
0.15
1 Includes opModelD = 0,1, 11-16, 21-25, 27, 33,35,37.
2 Includes opModelD = 28,29,30, 38,39,40.
                                             82

-------
Figure 41.  Example of Logarithmic Deterioration Model for THC (Cars, MYG 96-98): (a)
InTHC vs age, by VSP level (kW/metric ton), and (b) InTHC vs. VSP, by Age (yr).
                                                               -12
                                                               -21
                                                               •30
          QO     2.0    4.0    6.0     8fl
                             Age (Years)
100   12.0    14.0
       -9.0
      -10.0
          0.0     5.0    10.0   15.0    20.0    25.0    30 D    35.0    40.0
                               VSP (kW/tonne)
                                            83

-------
1.3.4.2.5.3
Apply the reverse transformation
After the previous step, the values of x/;age were reverse-transformed, as in Equation 30. The
values of the logarithmic variance used for this step were adapted from the Phoenix I/M results
and are intended to represent emissions distributions for "real-world" vehicle populations,
meaning that the values are higher than the value used in step 1.3.4.2.5.1 and may vary with age.
Values of logarithmic variances for all three pollutants are shown in Table 24.

Table 24.  Values of Logarithmic Variance Used to Calculate Emissions  Deterioration by Reverse
Transformation of Logarithmic Means.
Age Group

0-3 years
4-5
6-7
8-9
Pollutant
CO
1.30
2.05
2.00
1.80
THC
0.95
1.50
1.70
1.90
NOx
1.60
1.60
1.40
1.40
No values are presented in the table for the 10-14, 15-19 and 20+ year age Groups.  This
omission is intentional, in that we did not want to extrapolate the deterioration trend beyond the
8-9 year age Group. Extrapolation beyond this point is incorrect, as we know that emissions tend
to stabilize beyond this age, while the In-linear emissions model would project an increasingly
steep and unrealistic exponential emissions trend. For the 10-14, 15-19 and 20+ age Groups, the
"stabilization of emissions with age" was estimated as described in section 1.3.3.7.

Figure 42 shows the same results as Figure 41, following reverse transformation. The families of
parallel logarithmic trends are replaced by corresponding "fans" of diverging exponential trends.
An implication of this model is that as deterioration occurs, it is expressed more (in absolute
terms) at high power.  Similarly, the relationship between emissions and VSP becomes more
pronounced, in absolute terms, with increasing age.
                                           84

-------
Figure 42. Example of Reverse Transformation for THC (LDV, MYG 96-98): (a) THC vs. Age, by
VSP level (kW/metric ton), (b) THC vs. VSP, by Age (yr).
      110.0
                 3 kW/ton ne
                 6 kW/ton ne
                 9 kW/ton ne
                 12 kW/to nne
                 21 kW/to nne
                 30 kW/to nne
                20
                       4.0
                              6.0     8.0
                              Age (Years)
                                           10.0
                                                  12.0
                                                         14.0
      200.0 -f
      180
                50
                     10.0
                           15.0   20.0   25.0
                            VSP (kW/tonne)
                                             30.0
                                                   35.0   40.0
1.3.4.2.6
Estimate non-I/M References
Completion of steps 1.3.4.2.1 - 1.3.4.2.6 provided a set of rates representing I/M reference rates
for MY 2001-2021.  As a final step, we estimated non-I/M reference rates by applying the same
ratios applied to the I/M references for MY 2000 and previous, as described above.
                                              85

-------
1.4    Exhaust Emissions for Start Operation

1.4.1  Subgroup 1: Vehicles manufactured in model year 1995 and earlier

In EPA's previous emissions model (MOBILE6), start emissions for passenger cars and light-
duty trucks, were dependent upon three factors:

        1.  the (base) emissions of that vehicle at 75 degrees Fahrenheit31,
                                                           10
        2.  an adjustment factor based on the length of soak time , and
        3.  an adjustment factor based on the ambient temperature33

Within the MOVES modal structure, operating modes for start emissions are defined in terms of
soak time (preceding an engine start). The following sections will discuss the development of
base rates for "cold starts" (operating mode 108), as well as those for "warm" or "hot" starts
following seven soak times of varying length (operating modes 101-107).

Note that the development and application of temperature adjustments is discussed in a separate
report. 34

1.4.1.1       Methods


1.4.1.1.1     Data Sources
Data used in these analyses come 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 tens of thousands of vehicles under various conditions. EPA has
            stored those test results in its Mobile Source Observation Database (MSOD).

            We identified (in the MSOD) 549 gasoline-fueled vehicles (494 cars and 55
            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 EVI240
            cycle at temperatures of: 75, 40, 20, 0 and -20 °F35.

        3.  Under a contract with EPA, 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 [citation?]
                                          86

-------
        4.  During 2004-05, USEPA OTAQ and 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 cycle36.
1.4.1.1.2
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
Board37 Based on these data, we derived a correction factor "A" as shown in Equation 42  and
Table 25.
                         Cold - start Emissions =
                                               (Bag 1-Bag 3)
                                                   \-A
Table 25. Correction Factor A for application in Equation 39.
                                                               Equation 42
Vehicle Type
No Catalyst
Catalyst Equipped
Heated Catalyst
HC
0.37101
0.12090
0.05559
CO
0.34524
0.11474
0.06937
NOx
1.57562
0.39366
1.05017
1.4.1.1.3     Relationship between Soak Time and Start Emissions

In the MOVES input database, "operating modes" for start emissions are defined in terms of
soak time preceding an engine start.  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 the soak-time/start relationships mentioned above. The
specific values used are adapted from the MOBILE6 soak-effect curves for catalyst and non-
catalyst equipped vehicles15. 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, as
shown in Table 26.
                                          87

-------
Table 26. Operating-Mode Definitions for Start Emissions, Defined in terms of Soak Time.
Nominal Soak Period
(min)
3
18
45
75
105
240
540
720
OpModelD
101
102
103
104
105
106
107
108
OpModeName
Soak Time < 6 minutes
6 minutes <= Soak Time < 30 minutes
30 minutes <= Soak Time < 60 minutes
60 minutes <= Soak Time < 90 minutes
90 minutes <= Soak Time < 120 minutes
120 minutes <= Soak Time < 360 minutes
360 minutes <= Soak Time < 720 minutes
720 minutes <= Soak Time
       We have adapted and applied soak time adjustments used in MOBILE6.2 for gasoline-
fueled vehicles, as shown in Table 27.  Additionally, all pre-1981 model year passenger cars and
trucks use the catalyst equipped soak curve adjustments, although some of these vehicles are not
catalyst equipped.

Table 27. Calculated soak-time adjustments, derived from MOBILE6 soak-time coefficients for catalyst-
equipped vehicles.
opModelD
101
102
103
104
105
106
107
108
Soak
period
(min)
3
18
45
75
105
240
540
720
HC
0.051
0.269
0.525
0.634
0.645
0.734
0.909
1.000
Adjustment
CO
0.034
0.194
0.433
0.622
0.728
0.791
0.914
1.000
NOx
0.093
0.347
0.872
1.130
1.129
1.118
1.053
1.000
       Model-year groups used to calculate start rates for vehicles in model year 1995 and
earlier are shown in Table 28. In some cases, model-year groups were adjusted to compensate for
sparsity of data in narrower groups.  For example, the average NOx emissions for MY 1983-
1985 trucks are slightly negative. This result is possible, 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 MY1994-1999 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.
                                           88

-------
Table 28. Cold-start emissi
Model-year
Group «
Years
ms (Bag 1 - Bag 3,) for gasol
Mean (g)
THC CO NOx
ine-powered cars and trucks
Standard deviation (g)
THC CO NOx
CV-of-the-Mean (RSE)
THC CO NOx
Cars
1960-1980 1,488
1981-1982 2,735
1983-1985 2,958
1986-1989 6,837
1990-1993 3,778
1994-1995 333
5.172 75.832 0.608
3.584 52.217 1.118
2.912 34.286 0.922
2.306 21.451 1.082
1.910 17.550 1.149
1.788 16.233 1.027
6.948 83.812 2.088
7.830 60.707 1.682
5.216 44.785 1.321
2.740 32.382 1.034
1.728 13.953 1.034
1.203 31.648 0.742
0.035 0.029 0.089
0.042 0.022 0.029
0.033 0.024 0.026
0.014 0.018 0.012
0.015 0.013 0.015
0.037 0.107 0.040
Trucks
1960-1980 111
1981-1985 910
1986-1989 1192
1990-1995 1755
9.008 115.849 0.155
4.864 94.608 0.0412
3.804 45.918 2.107
3.288 40.927 2.192
9.179 113.269 2.682
4.992 67.871 1.797
2.298 36.356 2.152
4.211 42.478 2.158
0.097 0.093 1.641
0.034 0.024 1.445
0.017 0.023 0.030
0.031 0.025 0.024
1.4.2  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 EPA In-use Verification Program (IUVP), as with running rates for MY 2001 and later
(see 1.3.4 above).

For model years 1996-2000, rates for vehicles at 0-3 years of age (ageGroup=0003), are shown
above in Table 22, in the row for MY2000.

For MY 2001 and later, cold-start rates (opModeID=108) were estimated as described in 1.3.4
above, using the data and approaches described in steps 1-4 and step 6 (as described on page 60).
We applied the FTP averages as shown in Figure 26 and Figure 27,  and the phase-in assumptions
shown in Figure 28, Figure 29, Figure  30, Figure 31 and Figure 32.  As with running emissions,
Figure 33 illustrates the calculation of weighted average FTP results by model year.

To estimate start emissions for the remaining seven operating modes, we applied "soak
fractions" to the "cold-start" emissions, as described above. The soak fractions were adapted
                                            on
from the approach applied in the MOBILE model . Specifically, the part-wise regression
equations used in MOBILE6 for "conventional catalyst" engines were evaluated at the midpoint
of the soak period for each operating mode. For each mode, the start rate is the product of the
cold-start rate and the corresponding soak fraction. Figure 43 shows the soak fractions for HC,
CO and NOx, with each value plotted  at the midpoint of the respective soak period.
                                          89

-------
Figure 43. Soak Fractions Applied to Cold-Start Emissions (opModelD = 108) to Estimate
Emissions for shorter Soak Periods (operating modes 101-107).
       1.40
       0.00
                  120      240      360      480

                            Soak Time (minutes)
600
720
1.4.3  Applying Deterioration to Starts

1.4.3.1        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 information on deterioration for start emissions.
After some consideration, it occurred to us that the data from the IUVP program, used to develop
running rates for NLEV and Tier-2 vehicles, could also be useful to evaluate the relationship
between deterioration trends for start and running emissions. A valuable aspect of these 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 evaluate trends in emissions vs.
mileage.

At the outset, we plotted the data for NMOG and NOx vs. odometer reading, on linear and
natural log scales. Scatterplots of start and running NMOG emissions are shown in Figure 44 and
Figure 45; corresponding plots for InNMOG are shown in Figure 46 and Figure 47. Similarly,
scatterplots of start and running NOx emissions are shown in Figure 48 and Figure 49;
corresponding plots for InNOx are shown in Figure 49 and Figure 50.

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 the

                                           90

-------
purpose of the IUVP program, 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.  In the log plots, the degree of skew is shown by the variability of the transformed
data, with the In(start) spanning 3-3.5 factors of e, and the In(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 In-transformed data. To illustrate, we will focus on models run on
vehicles certified to LEV standards, as shown in Table 31 and Table 32. The model structure
includes a uniform intercept for all vehicle classes (LDV, LDT1-4), with 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 Bin5 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. For InNOx, the running slope is 1.65
times that for starts, and for InNMOG, the running slope is 1.25 times that for starts.
                                           91

-------
Figure 44. Cold-start FTP emissions for NMOG (g) vs. Odometer (mi), for LEV vehicles, from the IUVP
program
         •'
                                                       laaws
Figure 45. Hot-running (Bag 2) FTP emissions for NMOG (g/mi) vs. Odometer (mi), for LEV vehicles, from
the IUVP program
                                              92

-------
Figure 46. Cold-start FTP emissions for In(NMOG) vs. Odometer (mi), for LEV vehicles, from the IUVP
program
                                   a  -1
     -=5


Figure 47. Hot-running (Bag 2) FTP emissions for In(NMOG) vs. Odometer (mi), for LEV vehicles, from the
IUVP program
          it
           §
                   rtslas °D n 1»T2  o Q o LOVT1 -•:• -'* -     '" * * MDV1 *-*•+•
                                               93

-------
Figure 48.  Cold-start FTP emissions for NOx (g) vs. Odometer (mi), for LEV and ULEV vehicles, from the
IUVP program
                              1QQQ0               UW5
                                                       UOOffi      UOOOQ
Figure 49.  Hot-running (Bag 2) FTP emissions for NOx (g/mi) vs. Odometer (mi), for LEV and ULEV
vehicles, from the IUVP program
                                             94

-------
Figure 50. Cold-start FTP emissions for In(NOx) vs. Odometer (mi), for LEV vehicles (Source: IUVP
program).


                                                    i     * •* "-
Figure 51. Hot-running (Bag 2) FTP emissions for In(NOx) vs. Odometer (mi), for LEV vehicles (source:
IUVP program).
                        scc-x   ^>»M:=   ?]&:'}


                wisdas  n D D IST2  ° '" Q LUVTI  - '* & MK^  v s" 1i" M3₯3  * •*• H- J«iJ₯4  •'• * f ^SW
                                                95

-------
Table 29.  Model fit parameters for InNMOG, for LEV vehicles
Parameter
Predictor
Estimate
Standard error
Denom. D.F.
/-value
Pr>/
Cold-Start (Bag 1 - Bag 3) (residual error = 0. 1942)
Slope
intercept
intercept
intercept
intercept
Odometer (mi)
LDV-T1
LDT2
LDT3 (MDV2)
LDT4 (MDV3)
0.000004982
-1.9603
-1.7353
-1.5735
-1.2937
0.0
0.02224
0.02429
0.03520
0.03233
2,404
2,404
2,404
2,404
2,404
co
-88.14
-71.43
-44.70
-40.01
O.0001
O.OOOl
O.0001
O.OOOl
O.OOOl
Hot-Running (Bag 2) (residual error = 1.3018)
Slope
intercept
intercept
intercept
intercept
Odometer (mi)
LDV-T1
LDT2
LDT3 (MDV2)
LDT4 (MDV3)
0.000008237
-6.1604
-6.2554
-5.9018
-5.5949
0.0
0.05961
0.06577
0.09239
0.08766
2,225
2,225
2,225
2,225
2,225
co
-103.34
-95.11
-63.88
-63.83
O.OOOl
O.OOOl
O.OOOl
O.OOOl
O.OOOl
Table 30.  Model fit parameters for InNOx, LEV+ULEV vehicles ,
Parameter
Predictor
Estimate
Standard error
Denom. D.F.
/-value
Pr>/
Cold-Start (Bag 1 - Bag 3) (residual error = 0.68)
Slope
intercept
intercept
intercept
intercept
Odometer (mi)
LDV-T1
LDT2
LDT3 (MDV2)
LDT4 (MDV3)
0.000009541
-2.6039
-2.4538
-2.0769
-1.645
0.0
0.05231
0.06056
0.08173
0.08882
1,657
1,657
1,657
1,657
1,657
co
-50.74
-40.52
-25.41
-18.52
O.OOOl
O.OOOl
O.OOOl
O.OOOl
O.OOOl
Hot-Running (Bag 2) (residual error = 2.9643)
Slope
intercept
intercept
intercept
intercept
Odometer (mi)
LDV-T1
LDT2
LDT3 (MDV2)
LDT4 (MDV3)
0.000012
-4.7396
-4.9527
-4.3144
-4.1214
0.00000165
0.1092
0.1304
0.1740
0.1835
1,622
1,622
1,622
1,622
1,622
7.13
-43.40
-37.98
-24.80
-22.47
0.0001
0.0001
0.0001
0.0001
0.0001
                                            96

-------
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 30 (page 32) to obtain predicted geometric and
arithmetic means with increasing mileage, as shown in Table 3 Ifor NMOG and Table 32 for
NOx.

We normalized the predicted means at each mileage to the value at 0 miles to obtain a
"deterioration ratio" Rdet, by dividing each predicted value at a given mileage by the predicted
value at 0 miles (i.e., the intercept); Rdet for the intercept =1.0 (Equation 43 ).

                                             V
                                        T-)   _ a, miles
                                       -'Met ~ ~^. -                             Equation 43
                                              *a,o

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 R^i for start to that for
running, designated as Rre\
                                             n
                                             "'Met, start
                                             Met,
                                                                                Equation 44
                                              t, running
Values or R^i and ^i for NMOG and NOx are shown in Table 3 1 and Table 32, respectively,
with corresponding results shown graphically in Figure 52 and Figure 53, respectively.
                                            97

-------
Table 31. Application of Models for NMOG, representing emissions trends for LDV-T1 vehicles certified to
LEV standards.
Parameter

Odometer (mi, x 10,000)
0
1.5
2.5
3.5
4.5
5.5
6.5
7.5
8.5
Cold Start
InNMOG
Geometric mean
Arithmetic mean
Deterioration
ratio (Rdei)
-1.960
0.141
0.156
1.000
-1.886
0.152
0.168
1.078
-1.836
0.159
0.176
1.133
-1.786
0.168
0.185
1.190
-1.736
0.176
0.195
1.251
-1.686
0.185
0.205
1.315
-1.636
0.195
0.215
1.382
-1.587
0.205
0.226
1.453
-1.537
0.215
0.238
1.527
Hot Running
InNMOG
Geometric mean
Arithmetic mean
Deterioration
ratio (Rdet)
Relative Ratio (Rie{)
-6.160
0.00211
0.00404
1.000
1.000
-6.037
0.00239
0.00458
1.132
0.9952
-5.954
0.00259
0.00497
1.229
0.922
-5.872
0.00282
0.00540
1.334
0.892
-5.790
0.00306
0.00586
1.449
0.864
-5.707
0.00332
0.00636
1.573
0.836
-5.625
0.00361
0.00691
1.708
0.809
-5.543
0.00392
0.00750
1.855
0.783
-5.460
0.00425
0.00815
2.014
0.758
Table 32. Application of Models for NOx, representing emissions trends for LDV-T1 vehicles certified to
LEV standards
Parameter

Odometer (mi, x 10,000)
0
1.5
2.5
3.5
4.5
5.5
6.5
7.5
8.5
Cold Start
InNOx
Geometric mean
Arithmetic mean
Deterioration
ratio (Rdet)
-2.604
0.0740
0.1039
1.000
-2.461
0.0854
0.1199
1.154
-2.365
0.0939
0.1319
1.269
-2.270
0.1033
0.1452
1.396
-2.175
0.1137
0.1597
1.536
-2.079
0.1250
0.1757
1.690
-1.984
0.1376
0.1933
1.859
-1.888
0.1513
0.2126
2.045
-1.793
0.1665
0.2339
2.250
Hot Running
InNOx
Geometric mean
Arithmetic mean
Deterioration
ratio C^det)
Relative Ratio (RKi)
-4.740
0.0087
0.0385
1.000
1.000
-4.560
0.0105
0.0461
1.097
0.964
-4.440
0.0118
0.0520
1.350
0.940
-4.320
0.0133
0.0586
1.522
0.918
-4.200
0.0150
0.0660
1.716
0.895
-4.080
0.0169
0.0745
1.935
0.874
-3.960
0.0191
0.0840
2.181
0.852
-3.840
0.0215
0.0947
2.460
0.832
-3.720
0.0242
0.1067
2.773
0.811
                                                98

-------
Figure 52.  Deterioration Ratios for Cold-Start and Hot-Running NMOG Emissions.

       2.5
       2.0 --

       1.5 --
  £    1.0
       0.5
       0.0
         •Cold-start

         •Hot-running

         •StartRunning
           0.0      1.0       2.0      3.0      4.0      5.0
                                      Mileage (mi, x 10,000)
                                                     6.0
7.0
3.0
9.0
Figure 53.  Deterioration Ratios for Cold-Start and Hot-Running NOx Emissions.

      3.0  -,

      2.5

      2.0
    -  1
    -.p  I.
    (0
      1.0
      0.5
      0.0
        •Cold-start

        -Hot-running

        •StartRunning
0.0      1.0      2.0      3.0     4,0     5.0      6.0
                          Mileage (miles, x 10,000)
                                                                7.0
 8.0
9.0
For NOx (Figure 53) we decided to assign start NOx the same multiplicative relative
deterioration as running NOx.  However, for HC, the difference between running and start
deterioration was greater, large enough that we reduce to reduce start deterioration relative to
running deterioration.
1.4.3.2
     Translation from Mileage to Age Basis
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
                                             99

-------
assigns deterioration on the basis of age. It was therefore necessary to translate the Rre\ 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 year, from which it
follows that the RTe\ 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 ^1 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 (ARrei) over the corresponding change in time
(Atime).
                                   0.675-1.0   -0.325
                                                       = -0.30952
                           Atime    12.5-2     10.5

The calculation of the slope lets us estimate a value of ^i for each ageGroup.
Equation 45
                                                                               Equation 46
The results, as applied for hydrocarbons and CO, are shown in Table 33 and Figure 54. The net
result is a 15-40% reduction in multiplicative start deterioration, relative to running deterioration.

Table 33.  Relative deterioration ratios (Rrel), for NMOG (and CO), assigned to each ageGroup.
AgeGroup
0-3
4-5
6-7
8-9
10-14
15-19
20 +
Age (years)
2
5
7
9
12.5
17.5
23
Relative Ratio (Riel)
1.000
0.845
0.783
0.721
0.613
0.613
0.613
                                           100

-------
Figure 54. Relative Deterioration Ratios (Rrel), for NMOG (and CO), assigned to each ageGroup.


S3
4
.2
Of
>
"jB
15
CD
ft"

f-UUU
fj CffiQ -



U.600 -

U.4UU
U.JUU
0.2uO
U. 100 "
-\
"Nii^^^
^"^^-^^






l . . . . l




















                                         10             15
                                                        20
25
1.4.3.3
Application of Relative Multiplicative Deterioration
An advantage of the modal approach is that any driving cycle can be represented as a weighted
average of the MOVES emission rates and the "operating-mode distribution" for the cycle. In
this case, we applied an operating-mode distribution for the "hot-running" phase of the FTP. This
phase is 860 seconds long that represents urban driving over a 3.86 mile route after the engine
has stabilized at its normal operating temperature.  We estimated an operating-mode distribution
using the "PhysicalEmission Rate Simulator" (PERE)16. This distribution, shown in Table 34,
represents a "typical" car, with engine displacement and test weight of 2.73 L and 3,350 Ib. A
corresponding "typical" truck was represented with displacement and test weight of 4.14 L and
4,364 Ib, respectively.

Combining emission rates for hot-running emissions with the operating-mode distributions, we
calculated aggregate  cycle emissions for the second hot-running phase of the FTP (g/mi), for all
model-year and age groups.  Figure 55 and Figure  56 show resulting cycle aggregates for THC
and NOx. Note that the underlying rates for model years 1995 (representing Tier 0) and 2000
(representing Tier 1) were derived using data and methods described above in Section 1.3.3
(starting on page 15), and those for model years 2005 and 2010 were derived using data and
methods described above in Section 1.3.4  (starting on page 60).

It is important to note that this step is performed both for vehicles in inspection-and-maintenance
areas (I/M, using the meanBaseRateEVI) and for vehicles outside I/M areas (using
meanBaseRate). Because deterioration is  represented differently for the non-I/M and I/M
reference rates (see 1.3.3.7.2, page 58), and this difference is carried into deterioration for the
start rates, the result is that the MOVES rates represent that I/M programs have effects on start as
well as running emissions.
                                           101

-------
                                                   77
                                                _ -C/FTP2,MYG,Age
                                     •*Met,MYG,Age "~ ~^
Equation 47
Table 34.  Operating-mode distributions for running emissions, representing a "typical" car and truck on the
hot-stabilized phase of the FTP (Bag 2).
opModelD
0
1
11
12
13
14
15
16
21
22
23
24
25
27
28
29
30
o o
JJ
35
37
38
39
40
Cars (LDV)
7//we /'« mode
(sec)
97
155
77
121
83
59
22
4
42
111
62
18
7
2
0
0
1
0
0
0
0
0
0
Time in mode
(%)
11.27
18.00
8.94
14.05
9.64
6.85
2.56
0.46
4.88
12.89
7.20
2.09
0.81
0.23
0.00
0.00
0.12
0.00
0.00
0.00
0.00
0.00
0.00
Trucks (LOT)
Time in mode
(sec)
97
155
74
112
88
66
19
7
41
102
69
21
7
2
0
0
1
0
0
0
0
0
0
Time in mode
(%)
11.27
18.00
8.59
13.01
10.22
7.67
2.21
0.81
4.76
11.85
8.01
2.44
0.81
0.23
0.00
0.00
0.12
0.00
0.00
0.00
0.00
0.00
0.00
                                                 102

-------
Figure 55.  Cycle-aggregate THC emission rates by age, projected from MOVES running-exhaust emission
rates, for the hot-stabilized phase of the FTP, representing vehicles in I/M areas.
                        I , t ,  , ,>.  ,  I .  ft  ,  I .  , t
                                                             —4—1995

                                                             -•—200O

                                                             -•—2005

                                                             —e—2010
                             10       15

                             Age (years)
20
25
Figure 56.  Cycle-aggregate NOx emission rates by age, projected from MOVES running-exhaust emission
rates, for the hot-stabilized phase of the FTP, representing vehicles in I/M areas.
                                                                •1995

                                                                 2000

                                                                 2005

                                                                •2010
                             10       15

                             Age (years)
20
25
For each model-year group, we divided cycle aggregate for each ageGroup (£1FTP2,MYG,Age)  by the
estimate for the 0-3 year ageGroup (EWp2,MYG,o-3), to obtain a deterioration ratio (^det,MYG,Age) as
shown in Equation 47. As examples, ratios for cars are shown for THC in Figure 57(I/M) and
Figure 59 (non-I/M). Corresponding ratios for NOx are shown in Figure 59 (I/M) and Figure 60
(non-I/M). The ratios show that, in relative multiplicative terms,  the MOVES rates represent
greater deterioration for running exhaust THC than for NOx.
                                            103

-------
Figure 57. Deterioration Ratios for THC, representing the hot-stabilized phase of the FTP (Bag 2),
Representing vehicles in I/M areas.
                                                                —4—1995

                                                                -•—200O

                                                                -•—2005

                                                                —T—2010
                              10
15
20
                             Age (years)
Figure 58. Deterioration Ratios for THC, representing the hot-stabilized phase of the FTP (Bag 2),
Representing vehicles in non-I/M areas.
                                                                    -1995

                                                                    •2000

                                                                    •2005

                                                                    -2010
                              10
15
20
25
                             Age (years)
                                               104

-------
Figure 59. Deterioration Ratios for NOx, representing the hot-stabilized phase of the FTP (Bag 2),
representing vehicles in I/M areas.
tg
'£
ID

c

K
0


I
                                                                    •1995

                                                                     2000

                                                                     2005

                                                                    •2010
                               10       15

                               Age (years)
                                              20
Figure 60. Deterioration Ratios for NOx, representing the hot-stabilized phase of the FTP (Bag 2),
representing vehicles in non-I/M areas.
                                                                    -1995

                                                                     2000

                                                                     2005

                                                                    • 2010
                               10       15

                               Age (years)
                                              20
25
                                               105

-------
At this point, projecting deterioration for start emissions is a simple matter of multiplying the
start rate for the 0-3 year ageGroup in each relevant operating mode (opModelD =101-108) by
the deterioration ratio (Rdet) and the relative deterioration ratio (7?rei) for each ageGroup. The
projected start rate in each agegroup (Estaitage) is
    -'start, age
                                          'start,0-3-n-det,age-n-rel,age
                                                     Equation 48
Note that for NOx the values of ^i are 1.0 for all agegroups, i.e., relative multiplicative
deterioration for start emissions is the same as for running emissions.  For THC and CO,
however, Rre\ takes the values shown in Table 33, which reduces reduced relative start emissions
in comparison to relative running emissions. To illustrate the results, Figure 61 and Figure 62
show deterioration for cold-start emissions (opModeID=108) for THC and NOx, respectively.

Figure 61. Projected Deterioration for Cold-start THC Emissions (opModeID=108), in four Model years,
representing vehicles in I/M areas.
                                                                 •1995

                                                                 •2000

                                                                 •2005

                                                                 •2010
           o
10       15

Age (years)
20
25
                                             106

-------
Figure 62. Figure 63. Projected Deterioration for Cold-start NOx Emissions (opModeID=108), in four Model

years, representing vehicles in I/M areas.
^E

_j>g

 u

 ra
a:
a.

t


£
 E
 c
            0
                                                                     •1995



                                                                      2000



                                                                      2005



                                                                     •2010
                           10        15



                           Age (years)
20
25
                                                107

-------
1.7    Replication and Data-Source Identification

The rates developed as described in Section 1 represent gasoline-fueled conventional-technology
engines. For purposes of the draft version of the emissionRateByAge table, we replicated these
rates to represent other fuels and technologies.

At the outset, we replicated the entire set of gasoline rates for ethanol blends. However, for
ethanol blends, 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 document34.
Table 35. Fuel types and Engine technologies represented for gaseous-pollutant emissions from light-duty
vehicles.
Attribute
Fuel type


Engine Technology

sourceBin attribute
fuelTypelD


engTechID

Value
01
02
05
01
30
Description
Gasoline
Diesel
Ethanol
Conventional internal combustion (CIC)
Electric
Throughout the process, we assigned dataSourcelDs to subgroups of rates, which identify the
data and methods used to develop particular subsets of rates. The dataSourcelDs developed for
these analyses are listed and described in Table 36.  Note that the table also lists the numbers of
records in each dataSourcelD and relevant report section describing rate development for each
dataSourcelD.

Finally, Table 37 shows the accounting for all rates developed for light-duty gaseous-pollutant
emissions and included in the emissionRateByAge table. The leftmost four columns delineate
subsets of rates by the pollutant processes included (Running, Start), and the respective fuel
types (fuelTypelD), engine technologies (engTechID) and dataSourcelDs. The next seven
"accounting" columns show the construction of subtotals corresponding to combinations of
fueltype,  engtech, and dataSource. The values in these columns represent numbers of groups or
categories covered, as follows. Fueltype and engtech always represent single categories, as only
one of each is represented in a single row.  Two regClasses in each row always refer to two
categories, cars (LDV) and trucks (LDT).  The numbers of model-year groups (MYG),  age
groups and operating modes (opModes) covered varies with combinations of process and
dataSource.   Each row always represents the three gasous pollutants (CO, THC and NOx).
The rates for datasourcelD = 4400 - 4601 were summed as a single category, as these groups
represent the outcome of a set of interrelated processes,  as described in "Section 1.3.3
       Emission-Rate development:  Subgroup 1 (Model years through 2000)." The count of 15
modelyeargroups includes MY 2000 and earlier.  The dataSourcelDs 4800 and 4801 represent
running emissions for MY 2001+, as described in "Section 1.3.4  Emission-Rate Development:
                                          108

-------
Subgroup 2 (MY 2001 and later)."  For these rows, a count of one agegroup refers to the 0-3 year
ageGroup, whereas a count of three refers to the 4-5, 6-7, and 8-9 year ageGroups.
For these rates, the total of 21modelyeargroups represent groups 2001 - 2021-2050.

DataSourcelDs 4805 - 4807 represent start emissions for MY2001 - 2031-2050. For these
groups, counts of 26 or 36 modelyeargroups denote MY 1996-2031 and 1980 and earlier -
20212050, respectively. Counts of one or six ageGroups refer to the 0-3 ageGroup and the
remaining six ageGroups, respectively. Counts of one or seven opModes refer to the cold-start
emissions (opmode 108) or the remaining seven start modes, respectively.

The dataSourcelD 4900 refer to the replication of the gasoline/conventional rates to provide
base rates for ethanol blends. The Count of 36 modelyeargroups includes all groups from 1980
& earlier through 2021-2050.  The count of 31 opModes includes all modes for both the start
and running processes; 8 modes for start emissions and 23 modes for running emissions.

The count for dataSourcelD  4910 is similar, except that the 12 modelyeargroups include only
2010 - 2021-2050, as mentioned previously.
                                          109

-------
Table 36. Description of data sources and methods used in development of gaseous-pollutant emission rates
for light-duty vehicles.
DataSourcelD
4400
4027
4037
4427
4500
4527
4601
SUBTOTAL
4601
4800
4801
4805
4806
4807
4900
4910
Description
Data driven rates: averaged from continuous (second-by-second) IM240/IM147
data from Phoenix random evaluation sample, CY1 995-99 and CY2002-05, on
temperature range of 68-86 °F.
For opModelD = 29,39 only; use meanBaseRatelM calculated by ratio relative
to value in opMode 27; neither data mean or model prediction eligible
For opModelD = 30,40 only; use meanBaseRatelM calculated by ratio relative
to value in opMode 27; neither data mean or model prediction eligible.
For opModelD = 29,30,39,40 only; use meanbaseRatIM calculated by ratio
realative to value in opMode 27; either data mean or model prediction, or both
is eligible
imputed using statistical hole-filling models.
For opModelD = 29,30,39,40 only; use meanbaseRatelM calculated by ratio
relative to value in opMode 27; model prediction eligible, data mean not
eligible.
calculated by ratio relative to ageGroup 8-9, modefyeargroups 2000 and
earlier, ageGroupID 10-14, 15-19 and 20+ only)
calculated by ratio relative to ageGroup 8-9, modefyeargroups 2001 and later,
ageGroupID 10-14, 15-19 and 20+ only)
calculated by ratio from MY2000 rates, with ratios calculated from IUVP FTP
Bag-2 data, (modelyeargroups 2001 and later only, ageGroup 0-3 only).
calculated by applying deterioration to 4800 values, (modelyeargroups 2001 and
later only, ageGroups 4-5 through 8-9)
calculated from IUVP FTP results, as Bag 1 - Bag 3 mass (cold start, opMode
108 only, ageGroup 0-3 only).
calculated by applying deterioration ratios to 4805 values (cold start, opMode
108 only, ageGroup 4-5 and older).
calculated by applying soak fractions and deterioration ratios to 4805 values
(opModes 101-107 only, all ageGroups).
replicated from gasoline rates (fueltypeid = 1 ) to represent ethanol blends
(fueltypeid= 5).
replicated from gasoline rates for all engine technologies to represent rates for
tier-2 light-duty diesel engines (MY 2010 and later only).
No.
Records
6,968
118
252
138
3,572
268
3,174
14,490
8,694
2,898
8,694
156
1,296
10,584
46,872
15,624
Report
Section















                                                110

-------
Table 37.
Accounting for the segment of the emissionRateByAge table contributed by rates for gaseous-pollutant emissions for light-duty vehicles.
Process(es)

Start
Running
Running
Running
Running
Running
Running

Running
Running
Start
Start
Start
fuelTypelD

01
01
01
01
01
01
01

01
01
01
01
01
engTechID

01
01
01
01
01
01
01

01
01
01
01
01
dataSourcelD

101
4400
4027
4037
4500
4527
4601
4601
4800
4801
4805
4806
4807
Accounting (No. classes or groups)
fueltypes
1
engTechs
1
regClasses
2
MYG
10
ageGroups
1
opModes
1
pollutants
3
No. records

60
11 2 15 7 23 3 14,490
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
21
21
21
26
36
36
3
1
3
1
6
7
23
23
23
1
1
7
3
3
3
3
3
3
8,694
2,898
8,694
156
1,296
10,584
SUBTOTAL 46,872
Running & start
Running & start
05
02
01
01
4900
4910
1
1
1
1
2
2
36
12
7
7
31
31
3
3
46,872
15,624
TOTAL 109,368

111

-------
2.     Particulate-Matter Emissions from Light-Duty Vehicles
2.1    Introduction and Background

A large body of research is available on the formation and measurement of paniculate matter
(PM) emissions from combustion engines.  This chapter describes the process by which
emissions measured in a subset of the PM research programs on light-duty gasoline vehicles was
employed to generate emission rates for MOVES.  The emission rates developed by this
approach embody a strictly "bottom-up" method whereby emission rates are developed from
actual vehicle measurements following intensive data analysis, and are then structured as inputs
to the emissions inventory model. This is in contrast to a "top-down" approach which uses
measurements of ambient PM  concentrations from local regions and may apportion these
emissions to vehicles (and other sources), which are then input into inventory models.

The primary study that this chapter relies on is the "Kansas City Characterization Study"
conducted in 2004-200538.  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. This study is the most comprehensive and
representative study of its kind. In the context of a rigorous recruitment plan, strenuous efforts
were made to procure a representative sampling of the fleet. 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"39  This
"analysis report" (which is the partner to this chapter) presented preliminary emission rates for
PM,  elemental carbon fraction (EC), 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 hindcast 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 2005 Kansas City study.  The rates from this deterioration model
allow both forecasting and hindcasting as required by MOVES.

In addition, the previous analyses did not attempt to translate results measured on the LA92 cycle
(used in Kansas City) 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

                                          112

-------
operating modes defined for PM are the same as for the gaseous emissions (see Table 5, page
10).  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 Kansas City.

2.1.1  Particulate Measurement in the Kansas City Study

For measurements conducted on the dynamometer, vehicles were operated over the LA92
Unified Driving Cycle (see Figure 64). 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
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.
Figure 64.  Phases 1 and 2 of the LA92 Cycle, representing "cold-start" and "hot-running" operation,
respectively.
   70.0
   60.0
               200
400
600
800
1000
1200
1400
                                      time
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 an
Booker Systems Model RPM-101 Quartz-crystal microbalance (QCM) manufactured by Sensors,
                                           113

-------
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 |im
cutpoint pre-classifier. Further details and a schematic of the sampling instrumentation are
shown in Figure 65 and Figure 66.
Figure 65. Schematic of the constant-volume sampling system used in the Kansas-City Study.
       Dessicator

     J
   Air Conditioner
                                    Diluted exhaust    to aldehyde sample   aldehyde
                                    ~t 46 C          flow comtroller  A    cartridge
    from vehicle
    tailpipe

       T
                             4U
  to particle sample
  flow controller A

particle filter—t
                                                                          vent
                                background
                                                   T
\s. '^ sample line
air \F~~
46 C) ^ 	 Jj
i^
Backgrd HC analyzer



LI tf|J '-! | 	
pump - P I

filter — ^ |
•
flow _^-"J^*"
measurement
and control
                                flow
                                measurement
                                and control
                                                                         Positive
                                                                         Displacement
                                                                         Pump (POP)
                                                  HC analyzer
                                                                       Dilute exhaust
                                                                       collection bags
low CO
analyzer
        *       A       *
       igh CO     |    NOx analyzer
       nalyzer
                             high
                             analyze
                                    CO2 analyzer
                                              114

-------
Figure 66. Continuous PM analyzers and their locations in the sample line.
 <-Dyno
  cvs
                            PM2.EIMPACTORS
1.51pm.
ATER
TROL
i
i
i
i
i 	
QCM

HE ATE
CONTRC


1 m, d ~) Ipn
:ART SYSTEM
(massj
                                                 PHOTO ACOUSTIC
                                                   (black cation)
                             Data RAM
                               (light
                           scatteringi
  DusTrak
   (light
s cafe ring'
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 report40. As of the date of this program,
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 MOVES at this time, 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 more reliable of the 3 instruments,
and mass correction at low loads was not judged to be worth the effort given the uncertainties
                                          115

-------
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 report40.  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 with one another. 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 real-time PM at the time of
the Kansas City study, these instruments are 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.
2.1.2  Causes of Gasoline PM Emissions

In gasoline-powered spark-ignition engines, particulate matters is formed during incomplete fuel
and oil combustion (although the amount of oil consumed in combustion and its contribution to
PM varies greatly from vehicle to vehicle). During operation, numerous distinct technologies
used in vehicles are in various states of repair or disrepair which also affect PM emissions.  Even
brand new vehicles emit PM from combustion but at very low levels. A complete description of
the causes of PM emissions and associated mechanisms is beyond the scope of this report, as
many aspects of the science that are still not well understood. We will briefly summarize factors
that contribute to gasoline PM in the vehicle fleet in this section. Where appropriate, we will
make comparisons to the mechanisms of hydrocarbon (HC) formation, since parallels are often
drawn in the literature.

Simply put, particulate matter forms primarily during combustion when carbon-containing
molecules condense or otherwise agglomerate. This form of PM is generally composed of
higher molecular weight hydrocarbon compounds, some of which originate in the fuel/oil and
some of which are formed during combustion. Unlike diesel engines, elemental (molecular)
carbon or soot is not very prevalent with gasoline engines but does form in larger quantities
under relatively rich conditions, e.g., low airfuel ratios. The amount of elemental carbon in PM
varies from vehicle to vehicle (and, even for a given vehicle, varies depending on operating
conditions and state of repair). For gasoline-fueled vehicles,  a typical fraction is about 20% of
PM mass compared to about 70% for a diesel  engine.

Other compounds in the fuel or engine oil, such as trace levels  of sulfur and phosphorus,  form
form sulfates and phosphates during combustion, both of which form particulate.  The sulfur
level in gasoline is now very low, almost eliminating sulfate formation from gasoline sulfur

                                          116

-------
content but motor oil contains higher fractions of sulfur (and phosphorus) compounds. Also,
trace metal constituents in gasoline and oil form PM in the combustion process as metallic
oxides, sulfates, nitrates, or other compounds.  Catalyst attrition products from the substrate and
trace amounts of noble metals also form PM but not in the combustion process.  The catalyst
attrition products are mechanically generated and are usually in larger size ranges compared to
exhaust PM.  Exhaust PM as formed in the engine is generally very small in size (possibly much
of it is nuclei mode PM in the range of 0.05 microns or smaller). In the exhaust system,
including the muffler, some of the PM agglomerates and increases in size.

The wide assortment of technologies used in vehicles can affect PM formation.  These
technologies were mainly developed to control HC, CO and NOx emissions, but most have the
side benefit of reducing PM, since reducing exhaust HC generally also reduces exhaust PM
although not to the same extent. Older engines from the 1980s and earlier that deliver fuel
through a carburetor typically have poorer fuel droplet quality, as well as looser control of fuel
air stoichiometry. These older vehicles are expected to produce more PM  (on average) then their
fuel injected counterparts that followed generally in the late 1980s and early 1990s.

Among fuel injected engines, throttle body fuel injection (TBI) used in earlier engines with fuel
injection typically has poorer fuel atomization quality and airfuel ratio control than the port fuel
injection (PFI) technology that replaced; thus, one might expect older model-year fuel-injected
vehicles to have higher PM emissions (on average) than newer ones. Somewhat before the
widespread adoption of fuel injection, closed-loop control systems were developed in tandem
with oxygen sensors to improve the stoichiometric chemistry of combustion. These closed-loop
controls improved combustion as well as the effectiveness of the after-treatment system.

The after-treatment system on most vehicles consists of a 3-way catalyst. The 3-way catalyst was
designed for simultaneous control of hydrocarbons, carbon monoxide, and nitrogen oxides.
Vehicles with 3-way catalysts would meet more stringent hydrocarbon and carbon monoxide
emission standards while also meeting the first stringent nitrogen oxide standard. In oxidizing
hydrocarbons, these systems are resulted in additional PM control. These systems were utilized
on almost  all  gasoline-fueled vehicles beginning  in the 1981 model year.  On some model-year
vehicles in the 1980s and a few more recently, a secondary air injection system was added
between the engine and oxidation portion of the catalyst in order to add supplementary air to the
oxidation reactions on the catalyst.  These systems also helped oxidize PM (though probably not
to the extent that it oxidizes CO or HC).  The deterioration of these technologies may affect PM
and HC quite differently.  Throughout this chapter, there are parallels drawn between HC and
PM formation as well as control, however it  should be noted that the correlation between these
emissions  is far from perfect.  Many examples of this relation are shown in the 2008 analysis
      ^9
report .

Amounts of PM emitted are very  sensitive to the amount of fuel in combustion as well as the
stoichiometry of the reaction. As mentioned above, over-fueled mixtures result in higher PM
formation  and in some cases, also excess soot formation.  Over-fueling can occur under several
different conditions.  During cold start, engines are often run rich in order to provide sufficient
burnable fuel (i.e. light ends that vaporize at colder temperatures) to start combustion when the
cylinder walls are still cold (which results in increased flame quench). Additionally, under cold


                                           117

-------
conditions (e.g., below 20°F), additional enrichment of the fuel-air mixture is needed to start the
engine and lasts longer while the engine warms up.

When high acceleration rates or loads are encountered (such as a wide-open throttle event), an
extra amount of fuel is often injected for greater power or for catalyst and component
temperature protection.  Emission control systems in the late 1990s are better designed to control
this enrichment.  Finally, engines can run rich when a control sensor (e.g. oxygen, MAP, MAP,
or coolant sensors) or the fuel system fails.

In addition to fuel, lubricating oil can get into the combustion chamber via several pathways.
Some engines may have poor tolerances for pistons and piston rings, thus the negative pressures
(engine intake vacuum) can pull oil through these larger gaps during the intake stroke.
Furthermore, engine components, such as valves, valve seals, piston rings, and turbochargers can
wear and deteriorate resulting in increasing emissions over time. In all gasoline automotive
engines, the crankcase (where the oil bathes the engine components) is vented back into the
combustion chamber through the intake manifold.  This is known as Positive Crankcase
Ventilation (PCV), and  is required in order to remove and burn the excess hydrocarbons in the
hot crankcase.  Unfortunately, it can also introduce PM precursors and oil into the engine
combustion chamber. Because of the relatively small amount of oil consumption compared to
the volume of gasoline burned in a vehicle, the amount of HC from oil is typically small.
However, organic PM from oil consumption can be quite significant because oil is  a high
molecular weight hydrocarbon, and more likely to persist as unburned droplets. Therefore, as
vehicles age, those that  consume more oil will probably have very different emissions behavior
for HC than for PM, compared to when they were new. However, oil consumption can "poison"
the catalyst substrate, reducing the effectiveness of the catalyst at oxidizing HC.

The fuel itself may have properties that exacerbate PM formation, which may be affected by
concentrations of sulfur, lead, aromatics, and impurities.  With the lower levels of lead and sulfur
in fuels, the first two are presumably less of a factor in the Kansas City program than aromatics.
In the calendar years that MOVES models, lead is not a significant portion of the inventory, thus
is largely ignored.

Some of these PM forming mechanisms clearly affect HC emissions. A control technology or a
deterioration path for HC may or may not similarly affect PM depending on the source. It is also
likely that the processes that cause high PM may not be the same processes that cause organic
PM.  Some of the mechanisms also form visible smoke.  Smoke takes on a variety of
characteristics depending on the source, and  can be due to oil consumption or overfueling. The
smoke is visible because of the relative size of the particles compared to the light wavelengths
that are scattered. However, visible smoke is not necessarily a reliable indicator of high PM
emissions.

2.2    New Vehicle  or Zero Mile Level (ZML) Emission Rates

In this section, we develop an approach to extend the PM results from the Kansas City Study to
estimate average emissions across the fleet. The section also compares the new vehicle results
                                          118

-------
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 67), the most significant
challenge is distinguishing model year and age effects. As with most datasets, this issue arises
because the program was conducted over a two-year period, thus ensuring a direct
correspondence between model year and age. 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.

Figure 67. Average Participate Emission Rates from the Kansas City Study, by Model year, shown as cycle
aggregates on the LA92.






I :
ft. 
-------
2.2.1  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 discern model 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 cycles41. 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 HC
(Figure 68).

Figure 68. Hydrocarbon emissions a function of mileage (Gibbs et al., 1979)



_
E
DJ
•___•
0





3.5 -|
3 -
2.5 -

2 -


1.5 -
1 -
0.5 -
0 -

^



i
*

i i
* *
*

+





11111
0 10 20 30 40 50 60
mileage ('1000)
                                                                      42
Hammerle et al. (1992) measured PM from two vehicles over 100,000 miles.  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.
                                           120

-------
Figure 69. Participate emissions as a function of odometer for two Ford vehicles (Hammerle et al., 1992)

   3.5-
                                                                      Explore"
                                                                      Escort
                                                                     - Line EF (Exf lors r)
                2DOCO
                            40000
                                        60000
                                                    ODD 00
                                                               100000
                                                                           130000
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 number of starts, 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)43. 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% 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.
2.2.2  New Vehicle, or ZML Emission Rates and Cycle Effects
                                           121

-------
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 can be done 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 HC 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.

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 38 lists the 15 studies employed for this analysis.
Table 38. Historical gasoline PM studies including new vehicles at time of study.
Program
Gibbsetal41
Cadlee/a/.44
Urban & Garbe45,46
Lang et al.47
Volkswagen48
CARB49
Hammerle et al., 199242
CRC E24-1 (Denver)50
CRC E24-2 (Riverside)51
CRC E24-3 (San
Antonio)52
Chase et al.53
Whitney (SwRI)43
KC (summer)39'40
EPA (MSAT)54
Year of
Study
1979
1979
1979, 1980
1981
1991
1986
1992
1996
1997
1998
2000
1999
2004
2006
No.
vehicles
27
3
8
8
7
5
2
11
20
12
19

13
4
Drive
cycle
FTP
FTP
FTP
FTP
FTP
FTP
FTP
FTP
FTP
FTP
FTP
LA92
LA92
FTP
Before we examine these emissions, we should convince ourselves that the LA92 driving cycle
will not give significantly 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,
Riverside55. 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.

Finally, the California Air Resources Board conducted an extensive measurement program over
several years comparing many different drive cycles. Unfortunately, PM was not measured in
                                           122

-------
this program. However Figure 70 shows the HC emissions compared for the two cycles. The
trends indicate that there is little cycle effect for HC.

Figure 70. Hydrocarbon emissions on the LA92 versus corresponding results on the FTP
                                                                 10
12
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 results will be nearly identical even
if we omit the LA92 data, thus minimizing the significance of this issue.

Figure 71 shows the new-vehicle emission rates from the 11 studies listed in Table 38. 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 PM10 (for the plot),
which is nearly identical (about 97% of total PM is PM10). We also assume that 90% of PM10 is
PM2.5 (EPA, 1981)56. 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 MOBILE6 sulfur emission factors 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.
                                           123

-------
  CO
  E

                                                                      _L
Figure 71. Participate emission rates for new vehicles compiled from 11 independent studies.
        40


        35


        30


  _    25


        20


        15
   all
 • Gibbsetal., 1979
 o CadleetaL 1979
 D Urban&Garbe, 1980
 A Langetal.. 1981
 » VW, 1991
 • CARS. 1986
 « HamrnerleetaL, 1992
 n E24-1 Denver
 » E24-2UC Riverside
 e E24-3 San Antonio
 -' Chrysler/Ford'GM
 o SwRi/NREL
 x KC-Summer LA92
 +• MSAT-Tier2
 • mean for each program
^•Exponential Fit
           0
                                  10          15
                                     Model Year (+1975)
           25
30
To determine 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 determined from the KC 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:
                                              ^-0.814^
                                                                                  Equation 49
where
       y = model year - 1975, and
               = 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 the  SPSS statistical software tool. The averages of
these ratios by model year are shown in Figure 72, in which no clear trend is discernable. The
parameters of the model  are summarized in Table 39.
                                             124

-------
Figure 72. Ratios of hot-running/composite and cold-start/composite, Bag2 and Bagl-Bag3, respectively,
averaged by model year.


7

.
'35 ,-
5r A
o 4
I\
1 -
n -
19










BO
-*coldstart/comp
• bag2/comp





I
	 : : 	 1 	

1970



** *


+ +
***** I **'*•
* ++
• t *•* +

1980 1990 2000 20
model year










10
Table 39. Best-fit parameters for cold-start and hot-running ZML emission rates.
Parameter
LDV hot-running ZML (g/mi)
Exponential slope
Truck/car ratio
Bag-2 coefficient
Cold-start coefficient
Value
0.01558
0.08136
1.42600
0.89761
1.97218
Figure 73 shows the ZML emission rates. The rates are assumed to level off for model years
before 1975 and again after 2005. Elemental and organic carbon fractions are another
modification to the ZML rates.  These fractions are already reported in the analysis report.
                                             125

-------
Figure 73. Participate ZML emission rates (g/mi) for cold-start and hot-running emissions, for LDV and
LDT.
      0.050

      0.045

      0.040

      0.035

      0.030

      0.025

      0.020

      0.015
      0.010
      0.005
      0.000
                                                truck newbag1_3
                                                carnewbag1_3
                                               -truck newbag2
                                               -carnewbag2
         1975
      1930
1935
  1990
model year
1 995
20 00
2005
2.2.3  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 from 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
hindcast the past as well as forecast the future, as required in inventory models.
2.2.2.4
Age Effects or Deterioration Rates
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
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 age Groups. 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+. Having a single age
category for 20 years and older implies that emission rates have stabilized by 20 years of age.
The bag measurements from all of the vehicles measured in Kansas City were first adjusted for
temperature using the equation derived in the analysis report39. The equation used is:
                                           126

-------
                                               -0.03344(72-7)                        _
                                                                                Equation 50

where £pM,72, is the adjusted rate at 72°F for cold-start or hot-running emissions, E?M,T 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 (Chapter
1).

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,
airfuel-ratio control, failed oxygen sensors, worn engine parts, oil leaks, etc.  Figure 74 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 In-linear scale, the deterioration rates
appear approximately linear over this time period, suggesting that the deterioration rates are
exponential over this time interval. 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 y-axis offsets. The
result is a series of ladder-like linear trends in log scale  as shown in Figure 75. The lines fan out
exponentially on a linear scale as shown in Figure 76. 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.
                                           127

-------
Figure 74.  The natural logarithm of THC emissions vs. Age for LDV in the Phoenix (AZ) Inspection and
Maintenance program over a ten-year period (1995-2005).
                                      LDV WEIGHTED
                                 InfTHC) vs. Age (years), LDV
                                        Vehicle age (years)
      Model Year    o o o  1330   o o o  ^81  o o o  1335       -•>  1333  o o o  1334   o o o  1335
                        1986   D D D  1337  D D D  1338   B-B-B  1339  B-B-B  1990   B-B-B  S91
                  B-B-B  1992       '-  1933  A A A  1994   A A A  1995  A A A  1995         1997
                  A A A  I3=e   A A A  1353         2000   -1—I—I-  2001  -1—I—h 2002   -1—'—^  2003
Figure 75.  The Multiplicative deterioration model applied to PM results from Kansas City. The y-axis
offsets represent ZML rates.  The dotted line represents the Kansas-City Data.
     4.5
     4.0
                                                     128

-------
Figure 76.  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% confidence intervals).
   100
                                           15           20
                                          Age (years)
25
30
Because the model is multiplicative, the deterioration factors can be applied directly to trucks,
cold start, hot-running, EC, and OC, since the order of operations does not matter. The start
process requires only a soak time model to estimate remaining rates for starts other than the cold
start (opmodeID=101-107). Because no data is available describing how paniculate start
emissions vary by soak time, we have used the HC soak curves shown previously (see p. 87 ).

Substantial analysis is yet required to fill modal particulate emission rates for
emissionRateByAge table in the MOVES input database. Because the simple multiplicative
model can be applied across the range of VSP, deteriorated rates by operating mode  can be
directly generated, as described in the next section.
2.3    Estimating Elemental and Organic Carbon Fractions

After performing the analyses described above to estimate total particulate (PM2.5), we
partitioned the total into components representing elemental and organic carbon, EC and OC,
                                           129

-------
respectively.  Following this step, the values for EC and OC were loaded into the
emissionRateByAge table, using the pollutant and process codes shown in Table 1.
Table 40. Combinations of pollutants and processes for participate emissions.
pollutantName1
Primary PM2 5 - Organic Carbon
Primary PM2 5 - Elemental
Carbon
pollutantID1
111
112
processName2
Running exhaust
Start exhaust
Running exhaust
Start exhaust
processID2
1
2
1
2
polProcessID3
11101
11102
11201
11202
1 as shown in the database table "pollutant." Note that MOVES will reaggregate the paniculate components to
construct "Primary Exhaust PM10" (pollutantID 100) and "Primary Exhaust PM2 5" (pollutantID 1 10).
2 as shown in the database table "emissionProcess."
3 as shown in the database table "emissionRateByAge."
This discussion in this section is reproduced and adapted from the Kansas-City analysis report39.

Vehicle exhaust particulate matter consists of many different chemical species, including
elemental carbon (EC), organic carbon (OC), sulfates, nitrates, trace metals and elements. The
vast majority of the PM emissions is in the form of EC or OC. Elemental carbon, also known as
soot or black carbon, is produced during combustion when fuel or fuel droplets are pyrolyzed (or
carbonized) under low oxygen levels. In this process hydrogen is stripped from the carbon atoms
in the hydrocarbon, and carbon soot residue remains.  Elemental carbon is formed in gasoline
engines primarily when the fuel air mixture is rich (even in localized portions of the air/fuel
mixture of the engine).  The hot oxygen-starved and fuel rich environment favors pyrolysis
reactions.

We might expect to see higher EC fractions in gasoline engines following engine starts, or when
during enrichment mode such as under heavy engine load. These fine soot particles are generally
non reactive in the  atmosphere, though they may act as agglomerization centers for particle
growth both in the exhaust stream and in the atmosphere.  In other words, other compounds
including organic carbon adsorb onto the surface of the elemental carbon.  In turn, these
adsorbed organic carbon compounds can react in the atmosphere, generally in oxidation
reactions.

Organic carbon forms clusters of organic molecules that agglomerate and grow throughout
combustion, as the exhaust cools, and finally as it disperses into the atmosphere. In gasoline
engines OC can be formed normally during combustion from the fuel or the lubricating oil.
Sulfate emissions have  largely been controlled through fuel sulfur controls, and, previously, by
the closer control of airfuel ratio necessary for the three-way catalyst to effectively function.
We expect the sulfate emissions to be much lower than past studies.  Likewise, we also assume
that nitrates and trace metals and  elements are small on a mass basis by comparison. Therefore,
we spend the remainder of this section discussing EC and OC only.
                                           130

-------
It is important to separate EC and OC in inventory estimation since photochemical models treat
these fractions of paniculate separately. Also, the ratios are helpful for comparing emissions
(and air quality) models to source apportionment studies. Finally, EC is easier to measure and
more stable in the atmosphere than OC, therefore it is useful to track for a variety of purposes.

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 uing the photoacoustic analyzer, which measures EC on a
continuous basis.  More information can be found on these techniques and their calibration and
comparison results in the contractor's report40 and Fujita et al. (2006)57.  The former reference
indicates that the photacoustic 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 77 shows a plot of a comparison of the two
instruments on a natural-log 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, the
photoacoustic analyzer  seems to be shifted by about 2.4 mg/mi (near the origin of the plot).  An
adjustment equation may be appropriate if the TOR is the accepted standard, but since this offset
mainly affects small measurements only, it will probably have little impact on emissions
inventory models.

Figure 77. Comparison of Photoacoustic to TOR EC measurements on a logarithmic scale.
 o
 LU
 O
 O
 O
 ro
 I
 D. -
 »  bag 1
 •  bag 2
 *  bag 3
— Linear (bag 1)
— Linear (bag 2)
— Linear (bag 3)
                                                 = 0.982Jix- 0.2107
                                                   R2 = 0.9417
                                  ln(TOR EC)

We present trends of the ratio of EC to total PM (EC/PM) only.  Since in most cases the sum of
EC+OC = PM, generalizations can be extended to OC/PM as well, accounting, of course, for the
                                           131

-------
inverse relationship between EC and OC. There may be a small amount of non-carbon emission
in the PM, but we assume that it is negligible.

We explore the EC/PM ratio for the four measurement techniques employed in this 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 41 shows the comparison of the 3 different ratio methods using these
instruments. The values were determined from ratios of average values in the numerator and
denominator. The TOR ratios have two major limitations: the ratios are unexpectedly high and,
after eliminating bad data points, there are only 75 valid measurements.  Due to the latter
condition (primarily), the TOR ratios will not be used in subsequent analysis.  The photoacoustic
to dustrak ratios present a reasonable approach, however, since the Dustrak and PM are not
perfectly correlated40, we elected to use the  photacoustic to gravimetric filter ratios for EC/OC
rate estimation.
Table 41. Elemental to total PM ratio for 4 different measurement techniques.

PA/DT
PA/PM
EC/TC TOR
all
0.128
0.197
0.382
start
0.188
0.340
0.540
running
0.105
0.164
0.339
In the next 3 plots, we look for other factors that may affect the EC/PM variability. Temperature,
model year and vehicle weight are all examined. Figure 78 shows the relationship between
EC/PM to test temperature. These values were averaged for all test values within a 10°F bin and
then ratios were calculated between corresponding means.  We conclude from this plot that there
is very little temperature dependence to this ratio (though there may be a very small effect for hot
running bag 2).  Any temperature dependence is miniscule compared to the temperature effects
presented earlier for total PM. One might have  expected cold start EC ratios to be higher in
colder temperatures  due to the potential for extended rich starts,  however the data does not seem
to support this hypothesis.
                                          132

-------
Figure 78. Elemental Carbon to Total PM ratio as a function of test temperature.

n 7

OR
0 *}
0 4
E °'4
_Q.
* 0 3
n 9
n 1
n
c

.
• ec/pm start
• ec/pm running *


* *
•
•
* *
•
•
•
•
D 20 40 60 80 100
temperature
Figure 79 shows the EC/PM ratio within model year groups.  We conclude from this plot that
there seems to be very little model year or age dependence on the EC/PM ratio.
                                           133

-------
Figure 79. Elemental Carbon to Total PM ratio as a function of vehicle model year.
n -, » ec/pm start
0 6
0 5 -
0 4
E U'4
Q.
« 03 -
0 ?
0 1 -
n
• ec/pm running

• »
* « * * *
» *****
* » » *
* * *
* *
* *
••-.••••

•


1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Model year
Figure 80 shows the EC/PM ratio as function of vehicle weight. This plot shows a clear trend of
decreasing EC/PM ratio as weight increases.  This could be a function of engine displacement
(and peak power) as much as vehicle weight (the two tend to be correlated with each other).  The
trend may also be a function of the drive schedule since lighter (and possibly underpowered
vehicles) may be more likely to go into enrichment than more powerful vehicles if driven on
identical drive cycles.  In subsequent modeling (in MOVES), cars and light trucks are modeled
as separate vehicle types, which will capture some of this weight effect.
                                           134

-------
Figure 80. EC/PM ratio as a function of vehicle inertial weight.
0 4

Ooc
n •}

0 ?5
E
^~ 02
u
0)
0 15
n 1
0 05
n
(



• car
A true

A

•
.
A A
" A

D 1000 2000 3000 4000 5000
inertial weight (Ib)



< I







6000
An analysis shows the following statistics, with the breakdown of car vs truck in Table 42:
   *  avg Start EC/PM = 0.337
   *  avg Running EC/PM = 0.132
   *  Composite EC/PM ratio = 0.173
   *  The respective OC ratios can be calculated from the above by subtracting the fraction
       from 1.0.

The markedly higher level of EC during starts is not surprising given the rich fuel conditions that
exist during this mode of operation.

These results are roughly consistent with past studies, which found the OC:PM fraction in
Denver to range from 61-89%50, in the South Coast of California to range from 37-80%51, and in
San Antonio to range from 53-93%52.  For emission rate development, we use the values derived
from the Kansas-City study, summarized in Table 42. Non-carbon PM are included with OC and
is assumed to be small.

Table 42. Elemental and Organic Carbon PM fractions in from vehicles in the KC study.
            EC/PM      EC/PM    OC/PM    OC/PM
 Process     car       Truck       car     truck
   Start     0.345      0.325      0.655    0.675
 Running    0.179      0.068      0.821    0.932
2.4    Modal PM Emission Rates

As mentioned earlier, the continuous emissions measurements from the Kansas City study were
examined at great length, after which we determined that the Dustrak gave the most reliable
                                          135

-------
second-by-second PM time-series data when compared to the quartz-crystal microbalance
(QCM) and the Nephelometer.  In the following sections, we describe some of the trends in
continuous PM for "typical" normal-emitting and high-emitting vehicles.  We conclude by
describing the procedure by which results from the Dustrak were used to develop emission rates
by operating mode.

2.4.1  Typical behavior in particulate emissions as measured by the Dustrak and
Photoacoustic Analyzer

After looking at over 500 second-by-second traces, it became apparent that most of the vehicles
fell into certain general patterns. The most common behavior involved a highly non-linear PM
emissions increase as engine load increased. This pattern led to a monolithic "spike" in
emissions during the most aggressive acceleration event in the LA92 drive cycle during the 2nd
(hot-running) bag, at around 850 seconds.  This peak is captured in Figure 81, which includes
two plots. The higher emissions prior to 300 seconds can be attributed to cold start, during
which the engine is still cold and the fuel:air mixture tends to be on the rich side.  The plot on the
bottom confirms this supposition since it indicates that elemental carbon is relatively high during
the start. The hydrocarbons are overlaid on the bottom plot merely for comparison, and provide
a loose and qualitative comparison to organic PM emissions.  Some vehicles had variations on
this spike where it was much larger than even the cold start emissions, but this pattern is more
typical of the newer vehicles tested on the warmer days.

On the following series of plots the dustrak (most prominent), nephelometer and QCM are
overlaid on the top chart, while the photoacoustic analyzer, hydrocarbon and speed are overlaid
on the bottom chart.  Ordinate values are all relative and not absolute. "Shifted"  means time-
aligned, "Temp" means ambient temperature and the filter measurements as well  as vehicle type
and model year are written above the figures.

Figure 81. A typical time-series plot of continous particulate emissions as measured by several instruments.
                                          136

-------
            xlo* Test [84714] Model [STATION WAGON] MY [1994] Bag1 PM [B2.09 mg/m] Bag2 PM [6.69 mg/m] Temp [51.5 F]
                                                       L. J      >  i    .1
                     200        400        6

                   QCM (raw) 	DustTrak (shifted, n
  Time, second
rmalized)
                                                  DataRAM (shifted, normalized)	PA (g, shifted)
                            - PA (g, shifted)
                                                 GOO        1000       120i
                                           Time, seconds
                                         Speed (mph)	HC (g, normalized to bag 2 PA)
                                                                              1400
The next series of two figures shows how in some cases, the cold-start emissions appear to be
persist into the "hot-running" phase of the cycle (bag 2).  Figure 82 shows an older MY1976
vehicle tested at 54°F, for which one might expect the cold start emissions to have a longer
duration than a newer vehicle.  In this case, the cold start emissions seem to end at around 550
seconds (based on the HC trace).  However, such cases where large portions of the cold start
emissions leak occur during bag 2 were rare in the dataset, and thus they were not "corrected".
This step can be considered for future study.
                                              137

-------
Figure 82. Continous particulate emissions from a 1976 Nova measured at 54°F.

              x 10"'     Test [84712] Model [NOVA] MY [1976] Bag1 PM [354.43 mg/m] Bag2 PM [13.71 mg/m] Temp [54.3 F]
          o-
  3

| 2.5

f  2

i 1.5

  1

 0.5

  0
                                         J    L      ,
                       200
                                  400
                                                                  1000
                                                                             1200
                                             600         800
                                                Time, seconds
                      QCM (raw) 	DustTrak (shifted, normalized)	DataRAM (shifted, normalized)	PA (g, shifted)
              x10
                                  400
                                                                  1000
                                             600         800
                                                Time, seconds
                                • PA (g, shifted)      Speed (mph)     HC (g, normalized to bag 2 PA)
                                                                             1200
                                                                                        1400
Figure 83 shows a similar but slightly more commonly seen effect for a newer vehicle. The
difference is that the cold  start seems to end at around 250 seconds in bag 1, but then is high
again when bag 2 starts at around 350 seconds. Here the HC is low, but the EC (as indicated by
the PA) is relatively high hinting at a slightly fuel rich mixture.  It is uncertain at this time, why
these vehicles need to go into enrichment during this relatively mild acceleration.

Figure 83. Measured Particulate time series for a recent model year vehicle.
             x 10-'   Test [B4628] Model [TRAIL BLAZER] MY [2002] Bag1 PM [64.21 mg/m] Bag2 PM [12.17 mg/m] Temp [50 F]
          I 3
          «
          £ t
                      31
                    JMJLj
                       200
                                 400
                                                                 1000
                                                                            1200
                                            600         800
                                               Time, seconds
                     QCM (raw) 	DustTrak (shifted, normalized)	DataRAM (shifted, normalized)	PA (g, shifted)
                                                                                       1400
                                 400
                                                                 1000
                                            600         800
                                               Time, seconds
                               • PA (g, shifted)      Speed (mph)     HC (g, normalized to bag 2 PA)
                                                   138

-------
The traces shown so far have been "normal emitters" during hot running operation, i.e. they did
not have unusually high emissions during bag 2. These vehicles represent the bulk of the data.
However, some vehicles do exhibit higher or otherwise unusual hot-running PM emissions.
Examples are shown in the following series of figures.

Figure 84 shows a large "hump" of PM emissions starting at the beginning of bag 2 that lasts for
nearly 600 seconds.  The dustrak, nephelometer and the QCM all register this hump to varying
degrees, so it's unlikely that it is a mere instrument artifact.  The bulk of the bag 2 PM emissions
lies in this "hump," which does not coincide with a high load event. It is interesting that the PA
is not detecting a broad EC portion, so this hump is most likely organic carbon (OC), which
leads us to deduce that this hump probably represents OC particulate due to oil consumption.
Because these humps are not load based events, they don't suit themselves well to
characterization by VSP as correlation to power should not be high during the event.  Moreover,
it is interesting to note that the broad hump does not repeat.  Some vehicles have the hump at
different locations in the cycle (or throughout the whole cycle in rare cases), thus making this
effect impossible to model  physically using only a power-based approach. Therefore, the effect
can only be captured on an aggregate level by simply averaging with the normal emitters
described earlier.  It follows logically that if the recruitment of these "high emitters" was
representative in Kansas City, and these high emissions humps are not load dependent, then this
effect on the inventory should be captured by normalizing the modal rates to the filter
measurements; i.e. they are captured in the base emission rates.

Figure 84. Particulate time-series for a 1988 Dynasty.
                  Test[B414B] Model [DYNASTY] MY[19BB] Bag1 PM [52 mg/m] Bag2 PM [47.61 mg/m] Temp [87.5 f]
  0.01

I 0.008
Ci
g" 0.006

1 0.004
In

°- 0.002

    0
LkJkJU
                                               BOO
                   200        400        600
                                         Time, seconds
                  QCM (raw) 	DustTrak (shifted, normalized)
           x10
        5 0.5
                                                    JL-  i  A.
t Li-
                                                Dat
                                                 1000       1200       1400

                                          taRAM (shifted, normalized)	PA (g, shifted)
                             400
                                                        1000
                                      600       BOO
                                         Time, seconds
                           • PA (g, shifted)     Speed (mph)    HC (g. normalized to bag 2 PA)
                                                                 1200
                                                                           1400
Figure 85 shows another likely candidate for designation as an oil burner.  The emissions humps
are much broader, though the absolute emissions are similar to the Dynasty.  Note again that the
dustrak, nepholometer, and the QCM all register the hump, while the PA shows very little EC,
                                            139

-------
one of the "fingerprints" of oil-based particulate.  In one of the repeat test vehicles in the study,
one test exhibited a hump in emissions and the repeat test did not. The inconsistency and non-
repeatability of some of these humps arising from oil  consumption explains how some vehicles
can flip from "high" to "normal" emissions or vice-versa in replicate measurements. These
observations have ramifications for future PM research, in that sample sizes should be large and
fleets properly representative.

Figure 85. Continuous particulate time series for a 1995 Lincoln Continental.
            „ 1r/'  Test P43BO] Model [CONTINENTAL] MY [1995] Bagl PM [12.24 mg/m] Bag2 PM [41.35 mg/m] Temp [62.1 F]
          1.5-
          0.5
                     Ji
 200

QCM (raw)
                              400
                                                           1000
                                        600       800
                                           Time, seconds
                             " DustTrak (shifted, normalized)    DataRAM (shifted, normalized) -
 1200       1400

	PA (g, shifted)
                                                 800
                                           Time, seconds
                                                                              1400
                            • PA (g, shifted)
                                         Speed (mph)
                                                    • HC (3, normalized to bag 2 PA)
The next figure (Figure 86) shows a more typical high PM emitter, where the bag 2 emission rate
is 266 g/mi.  Here the EC does mirror the high emissions seen in the other instruments.  Even the
HC measurements are saturated.  This trace, representing a 1978 MG, is an indicator of poor fuel
control, as might be expected with an older (1978) carbureted engine.
                                              140

-------
Figure 86. Continuous particulate (and HC) time series for a 1978 MG.

                  Test 184277] Model [MG] MY[197B] Bagl PMP51.11 mg/m] Bag2 PM [266.15 mg/m] Temp [69.7 F]
                                                      1000
                                                               1200
                                     600        BOO
                                       Time, seconds
                  QCM (raw) - DustTrak (shifted, normalized) -- DataRAM (shifted, normalized) - PA (g, shifted)
                            400

                          - PA (g, shifted)
00       BOO       1000       12(
  Time, seconds
 Speed (mph)    HC (g, normalized to bag 2 PA)
We are now ready to classify the emission rates into operating modes based on speed,
acceleration and vehicle-specific power (VSP) (Table 5). The following two figures show
Dustrak PM emissions binned by VSP and classified by model year Groups. Figure 87 shows
this relationship on a linear scale and Figure 88 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.
                                              141

-------
Figure 87. Particulate emissions, as measured by the Dustrak, averaged by VSP and model-year
Group (LINEAR SCALE).
                                  Cars
         x10
                                                           1983
                                                           1989
                                                           1996
                                                           2000
                                                           2001
                                                           2004
                    K     15    20   25    30    35   40    45    50
                              VSP, kw/tonne
Figure 88. Particulate emissions, as measured by the Dustrak, averaged by VSP and model-year
Group (LOGARITHMIC SCALE).
                                   Cars
        -51	1	1	1	1	.	.	.	1
-6

-7
     «  -9
     I
     1-10
     o
       -11
       -12
       -13
                                                            1983
                                                           •1989
                                                            1996
                                                           "2000
                                                           •2001
                                                           •2004
                    10    15    20    25   30    35   40    45   50
                               VSP, kw/tonne
                                           142

-------
In order to determine the actual MOVES VSP based rates, followed seven steps:

    1.  The LA92 equivalent hot-running emission rate (g/mi) is determined for every model
       year and age group from the model described in section 2.2.
    2.  The gram per second (g/s) emission rate is determined from the dustrak for cars and
       trucks based on the KC data.  These trends are then extrapolated to the higher VSP bin
       levels where data is missing.
    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 modal rates (Step 2) are then combined with the operating-mode distribution and
       summed to give a total bag 2 emission factor that must match the aggregate LA92
       emission rates in step 1 (as calculated from the filter measurements).
    5.  The emission rates  are normalized to match the filter values through a normalization
       factor that is applied to every combination of model year and age group.
    6.  The rates from step 5 are then multiplied by the corresponding EC and OC factors to
       give rates for the hot-running process.
    7.  Steps above are repeated for all ages and model years.

The output from step 3 (operating-mode distribution) for cars and light trucks is shown in Figure
89.  For operating-mode definitions,  see Table 5.

Figure 89. Operating-Mode distribution for cars and light trucks representing the hot-running
phase (Bag 2) of the LA92 cycle.
   160
   140 -•
   120 -•
 •£ 100
 in
 T3
 c
 o
 o>  80 i
  0)
  c  60
    40 -•
    20
                                                                Dear
                                                                Dliqht truck
                                                  rfl
        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
The output of step 5 for each model year ZML (0-3 year age Group) is shown in Figure 90.
                                           143

-------
Figure 90. Particulate emissions for passenger cars (LDV) from Kansas City results, by model year
Group, normalized to filter mass measurements.
 IS
 cc
dC -,


in .




o:
c
•


.
.
• Jt
₯
• - I
. ,
B illlci i-i-ii-l •_!_* S 5 sJ-ii
•1960-1930
1961-1982
•1953-1934
t198S
* 1986-19 87
H 9 88-1 9 89
-1990
-1991-1993
1994
1995
1996
1997
1993
*1999
2000
-2001
-2002
+ 2003
2004

5 10 15 20 25 30 35 40 45
                                     VSP bin

After the rates were calculated, a quality check was performed to ensure that the aged rates in
any particular bin 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 bins in operating mode 30, representing values for cars and trucks in
the 1975 model-year Group.  In these cases, the value from operating mode 29 was copied into
mode 30.
2.5    Conclusions

The previous discussion describes analyses of particulate-matter emissions designed to develop
operating-mode based emission rates for use in the MOVES emissionRateByAge table,
incorporating the effects of temperature, model year and age. These rates include organic and
elemental carbon for cold-start and hot-running emissions from cars and light trucks (e.g., LDV
and LDT).  This analysis is crucial for understanding how PM emissions have changed over the
years and how new vehicle PM rates are projected to deteriorate over time.  The new vehicle
(zero mile level) PM emissions are estimated by analyzing the new-vehicle emissions rates from
historical PM studies.  The trends indicate that emissions have been decreasing exponentially
with model year as the engine and fuel controls have improved and after-treatment devices have
been installed.  The new truck rates are found to be larger than the car rates.  The deterioration
effect of age is determined by comparing the new vehicle  rates to the Kansas City data.
Based on patterns observed for the gaseous emissions, we have assumed that emissions
deteriorate exponentially with the age of the vehicle, but remain constant after about 20 years.
We also found that PM emission increase exponentially with VSP (or road or engine load).
                                           144

-------
There is still much analysis that can be conducted with these data. In the future, it would be
important to examine trends in the speciated hydrocarbons and organic PM from the standpoint
of toxic emissions and also quantifying the PM emissions due to oil consumption. These
analyses are likely to expand the scientific understanding of PM formation and why certain
gasoline fueled vehicles emit more PM than others under certain conditions.  It would also be
useful to explicitly capture the non carbon portion of particulate emissions.
                                           145

-------
3.     Gaseous and Particulate Emissions from Light-Duty Diesel
Vehicles (THC, CO, NOx, PM)

In MOVES, emission rates for running emissions 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
in relation to vehicle specific power. Therefore, we used aggregate results (in grams per mile)
from the Federal Test Procedure (FTP) to estimate aggregate rates, which we then translated into
corresponding modal rates (in grams per hour).
3.1    Gaseous Emissions: MY2009 and earlier, Particulate Emissions:
MY2003 and earlier.

The analyses in this section pertain to development of rates representing vehicles manufactured
prior to introduction of Tier-2 standards. For gaseous emissions, this grouping is represented by
MY 2009 and earlier.  For particulate emissions, the grouping represents MY 2003 and
earlier.3.1.1   Estimating Zero-Mile FTP Emissions:

We identified FTP results on the Annual Certification Test Results & Data website
(http://www.epa.gov/otaq/crttst.htm) and on the Test Car List Report Files Website
(http://www.epa.gov/otaq/tclrep.htm) for 513 diesel-powered LDV and 187 LDT from the 1978
through 2008 model years. These vehicles had been measured for purposes of engine
certification or generation of fuel economy estimates. These vehicles were new (age = zero
years), with each vehicle having accumulated about 4,000 miles. These data were used to
calculate mean (composite) FTP emissions (grams per mile of HC, CO, NOx, and PM10) for
each model year group. (We examined, but did not include data on European diesels since those
vehicles might not be representative of those sold in the U.S.) The sample sizes (by model year
group) and the mean composite FTP  emissions are given in Table 43 for cars and Table 44 for
trucks:
                                         146

-------
Table 43. Mean Composite FTP Emissions (g/mile) for diesel-fueled Cars (LDV).
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
Sample
Size
104
114
116
73
79
13
3
5
6
HC
0.4883
0.2508
0.2006
0.2178
0.2075
0.2123
0.2273
0.1364
0.0196
CO
.3425
.0861
0.9809
.1386
.3581
.6854
.2233
0.4140
0.5367
NOx
1.4126
1.1859
1.0517
0.8436
0.5952
0.5685
0.8567
0.8180
0.3925
PM1

—
0.2999
0.2881
0.2751
0.5668
0.4990
0.1747
0.0848
—
1 Measurements of PM emissions were not performed for the Pre-1981 model
year cars (or trucks). For this analysis, we applied the (later) 1982 standard
of 0.6 grams per mile to those earlier model years.
Table 44. Mean Composite FTP Emissions (g/mile) for diesel-fueled light-duty trucks (LPT).
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
Sample
size
13
45
56
11
20
5
17
14
6
HC
0.6900
0.3478
0.2578
0.2297
0.2364
0.3020
0.2213
0.1526
0.0181
CO
.7923
.3277
.0302
.1200
0.9985
.7000
.6256
.6179
0.2767
NOx
1.6577
1.3748
1.3052
0.9473
1.4435
1.2600
1.3814
1.4629
0.4583
PM1
—
0.3296
0.2700
0.2673
0.2790
0.1280
0.1114
0.0960
—
Because measurements of PM emissions were not performed for the Pre-
1981 model year cars (or trucks), we applied the (later) 1982 standard of 0.6
grams per mile to those earlier model years. Due to questionable PM results
for the 2006-2008 LOT, we used the LDV average PM value (0.03 12
grams/mile).
3.1.1.2
Estimating Bag Emissions:
The 700 certification (car and truck) test results were composite FTP results (HC, CO, NOx, and
PM), not differentiated by test phase (bag). Therefore, the first task was to estimate the
individual bag results based on the composite results.

A smaller sample (151 tests) of FTPs from other data sets had emission results by bag.  These
FTPs of in-use vehicles (of various ages from various model years) were used only to develop
correlations between the composite FTP emissions and the corresponding emissions of each of
the three bags/modes. The sources of these data are summarized in Table 45.
                                           147

-------
We regressed the Bag-2 emissions (in grams per hour) against the corresponding composite FTP
emissions (in grams per mile) to obtain an estimate of running emissions. For these regressions,
we used a piecewise linear approach rather than a polynomial regression to account for slight
curvature in the relationships.  Similar analyses were performed regressing Bag-1 emissions and
Bag-3 emissions (in total grams) each against the corresponding composite FTP emissions (in
grams per mile). Each of the 14 regressions produces an equation, such as the following
example, which correlates the Bag-1 "cold-start" HC emissions (Enc.Bagi, g) to the corresponding
composite FTP HC emission rate (EHc,composite,g/mile):
                          E-
                            HC.Bagl
                  = -0.6433 + 4.702885 E,
                                        HC, composite
                                                                                Equation 51
Graphing this equation along with the 146 FTP test results, as shown in Figure 91 below,
illustrates the relationship between the individual bag HC emission and the composite HC
emission.

Table 45.  Data Sources used to distinguish emissions by phase (bag) on the FTP for light-duty diesels.
Source
Norbeck et al., (1998a)M
Norbeck et al., (1998b)5!i
USEPA In-Use Verification Program
Mobile-Source Observation Database (MSOD)30
Total
No. Tests
19
15
12
105
151
Figure 91. Example: Bag-1 HC (g) versus Composite FTP HC (g/mile)
   16
   12
     0.0
0.5
1.0         1.5        2.0
FTP HCRate (grams/  mile)
2.5
3.0
                                           148

-------
We then applied those 14 equations (derived from the regression analyses) to the corresponding
composite FTP emissions shown in Table 43 and Table 44. This step yielded (for each model
year group in Tables 1 and 2) estimates of the emissions rate (in grams per hour) for Bag-2 as
well as the total emissions (in grams) for each of Bag-1 and Bag-3.

We then assumed that the running emission rates (in grams per hour) on Bag-2 were comparable
to the rates on the running portion of the Bag-1 (and Bag-3). Subtracting the total emissions
associated with those running rates from the estimated total emissions of Bag-1 (based on the
regressions of Bag-1 versus composite FTP) yielded estimates of the cold-start emissions (by
model year).  Similarly, subtracting the estimated running emissions from the estimated total
Bag-3 emissions produced estimates of hot-start emissions. Those estimated emission rates
(running, cold-start, and hot-start) are  summarized in the four following tables (Table  46 to Table
49), one table for each of the four pollutants.
Table 46
. Estimated A
Model Year
Group

Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
ggregate HC Emission Rates.
Diesel-Fueled Passenger Cars
Running
(g/hr)
8.0991
4.0262
3.1838
3.4727
3.2992
3.3802
3.6322
2.1069
0.1477
Cold-Start
(g)
1.0961
0.5505
0.4325
0.4729
0.4486
0.4600
0.4953
0.2816
0.0071
Hot-Start
(g)
0.1688
0.1626
0.1349
0.1444
0.1387
0.1414
0.1496
0.0995
0.0351











Diesel-Fueled Light-Trucks
Running
(g/hr)
11.2131
5.6533
4.1427
3.6724
3.7835
4.8847
3.5308
2.3782
0.1226
Cold-Start
(g)
1.6077
0.7784
0.5668
0.5009
0.5165
0.6707
0.4811
0.3196
0.0036
Hot-Start
(g)
0.3280
0.2161
0.1664
0.1510
0.1546
0.1908
0.1463
0.1084
0.0342
Table 47
. Estimated A
Model Year
Group

Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
?gregate CO Emission Rates.
Diesel-Fueled Passenger Cars
Running
(g/hr)
21.3626
17.1121
15.3696
17.9833
21.6212
27.0463
19.3873
5.9718
8.0052
Cold-Start
(g)rt
3.0900
2.5146
2.2787
2.6326
3.1250
3.8594
2.8226
1.0066
1.2818
Hot-Start
(g)
1.0957
0.8647
0.7700
0.9121
1.1098
1.4046
0.9884
0.2592
0.3698











Diesel-Fueled Light-Trucks
Running
(g/hr)
28.8186
21.1168
16.1856
17.6745
15.6605
27.2886
26.0552
25.9270
3.6954
Cold-Start
(g)rt
4.0993
3.0567
2.3892
2.5908
2.3181
3.8922
3.7252
3.7079
0.6984
Hot-Start
(g)
1.5010
1.0824
0.8144
0.8953
0.7858
1.4178
1.3508
1.3438
0.1355
                                          149

-------
Table 48. Estimated Aggregate NOx Emission Rates.
Model Year
Group

Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
Diesel-Fueled Passenger Cars
Running
(g/hr)
23.4257
19.5462
17.2503
13.6886
9.4389
8.9815
13.9128
13.2512
5.5883
Cold-Start
(g)rt
1.6481
1.4573
1.3444
1.1692
0.9602
0.9377
1.1802
1.1477
0.8433
Hot-Start
(g)
1.5561
1.3466
1.2227
1.0304
0.8009
0.7762
1.0425
1.0067
0.6673











Diesel-Fueled Light-Trucks
Running
(g/hr)
27.6186
22.7786
21.5870
15.4631
23.9537
20.8139
22.8916
24.2849
6.4738
Cold-Start
(g)rt
1.8543
1.6162
1.5576
1.2565
1.6740
1.5196
1.6218
1.6903
0.9325
Hot-Start
(g)
1.7824
1.5211
1.4568
1.1262
1.5846
1.4151
1.5272
1.6025
0.7619
Table 49. Estimated Aggregate PM Emission Rates.
Model Year
Group

Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
Diesel-Fueled Passenger Cars
Running
(g/hr)
7.0131
3.3778
3.2356
3.0774
6.6108
5.7897
2.4073
1.1338
Cold-Start
(g)rt
2.4362
1.2427
1.1960
1.1441
2.3041
2.0346
0.6682
0.3368
Hot-Start
(g)
1.2789
0.6436
0.6188
0.5911
1.2086
1.0651
0.3020
0.1378










Diesel-Fueled Light-Trucks
Running
(g/hr)
7.0131
3.7378
3.0160
2.9830
3.1250
1.7460
1.5101
1.2931
Cold-Start
(g)rt
2.4362
1.3609
1.1239
1.1131
1.1597
0.4961
0.4347
0.3782
Hot-Start
(g)
1.2789
0.7065
0.5804
0.5746
0.5995
0.2167
0.1863
0.1583
The PM rates in the preceding table represent the PM10 rates for all paniculate matter on the
collection filter (i.e., elemental carbon (EC), organic carbon (OC), sulfates, etc.).  Disaggregating
the PM estimates to obtain rates separately for EC and for OC, will be described in 3.3 below.
3.1.1.3 Assigning Operating Modes for Starts (Adjustment for Soak Time)

MOVES has start emission rates for eight operating modes (opModes), each based on the length
of the soak time prior to engine start. One mode corresponds to the 12 hour cold-soak
(opmodelD = 108).  The remaining seven modes have soak times ranging from three minutes up
to nine hours (opModelD  = 101-107).

Assuming that the start emissions change as functions of the temperature of the engine, and
assuming that the engine temperature decreases (cools) exponentially with the soak period (i.e.,
length of time the engine is shut off), then we should be able to approximate the start emissions
(following a soak EopModeio) by exponential functions  of the form:
                                                                              Equation 52
                                          150

-------
       where Ew% = cold-start emissions (g) and t = soak time (min),  in minutes.

(Note that the factor of 1.001 (rather than 1.0) in the preceding equation allows the exponential
curve to pass through the cold-start value at 720 minutes rather than simply approaching it.)

Using the estimated cold-start (CS) emissions i.e., emissions following a soak of at least 720
minutes (Eios) and the hot-start emissions i.e., the emissions following a soak of only 10 minutes
(Eid) from the preceding four tables, we solved algebraically for both the a and ft coefficients,
specifically:

                                _ 720y8+ln0.001
                                         F
                                   In 1-^M-lnO.OOl                         Equation53
                               Q -
                                           710

This approach yielded a unique start emission curve (as a function of soak time) for each
pollutant and for each model year group.

The effect of this exponential approach is illustrated in the following example (Figure 92) which
was created using the estimated cold-start THC emissions of 0.281593 grams for the  1995-2005
model year diesel-fueled passenger cars and the estimated hot-start THC emissions of 0.099486
grams from the preceding table.
                                           151

-------
Figure 92. Estimated THC Start Emissions (g) in terms of Soak Time (1995-2005 LDV).


              THC Start Emissions (grams per start)
                         1995-2005 LDDVs
   0.3
   0.2
   0.1
              120      240      360      480
                         Soak Time (minutes)
600
720
This continuous concave curve is broadly comparable to the piecewise approach that the
California Air Resources Board used in its analysis of the effect of soak time on the start
                                                           	    T/^
emissions of gasoline-fueled vehicles and that EPA used in MOBILE6 .
3.1.2  Running Emissions by Operating Mode

In MOVES, running emission rates are estimated for a set of operating modes defined in terms of
vehicle-specific power, speed and acceleration (see Table 5, page 14).  However, we lacked the
requisite second-by-second data for the diesel-fueled cars and light-trucks to perform those
calculations. Therefore, we developed modal rates for LDT from corresponding rates for light
heavy-duty diesel-fueled trucks (LHD<=14K) (i.e., from trucks with gross vehicle weight ratings
between 8,500 and 14,000 pounds).

To adapt the LHDDT operating modes for application to LDDs, we developed operating mode
frequencies in each mode for the 1,372-second LA-4 drive cycle (the first two phases of the FTP
run sequentially). Due to differences in vehicle weight, we obtained separate (slightly different)
distributions for passenger cars and and light-trucks, as shown in Table 50.
                                          152

-------
Table 50. Operating-Mode Distribution for the LA-4 Drive Cycle.
opModelD
0
1
11
12
13
14
15
16
21
22
23
24
25
27
28
29
30
33
35
37
38
39
40
LDV
164
255
93
142
99
69
34
20
68
149
123
35
21
14
8
2
0
25
35
13
3
0
0
LOT
164
255
96
139
103
66
33
20
70
164
110
33
19
15
7
2
0
29
33
11
o
J
0
0
Applying the appropriate distribution to the modal emission rates for the LHDDVs, we obtained
estimates of the emission rates (in grams per hour) over a simulated LA-4 driving cycle.
Dividing those rates into the hour running rates for the light-duty diesels (Table 46 through
Table 49), by model-year group, yielded ratios of the light-duty emission rates to the light heavy-
duty rates.  The resulting ratios are then used as adjustment factors to scale the modal LHD rates
to give estimated modal LDD rates. For example, applying the LA-4 operating-mode
distribution to the NOx modal rates for the 1999-2002 model year LHDDVs produces an
estimated NOx rate of 143.66993 grams per hour compared to the actual passenger car average
rate of 13.2512 grams per hour.  Dividing yields a ratio of 0.092234.  Therefore, we used that
ratio (0.092234) as a scaling factor to multiply all of the modal LHDDV rates for that model-
year group to produce the corresponding VSP bins for the 1999-2002 model year diesel-fueled
passenger cars.  Thus, summing all of the LA-4 modal rates will exactly match the total estimate
LA-4 (running) emissions.

Not all of the operating modes are represented by the LA-4 driving cycle.  Specifically, modes
30, 39, and 40 do not occur during the LA-4.  For this analysis, we applied the same adjustment
factor to all operating modes.

This approach is illustrated by the following plots (Figure 93 to Figure 96) of the estimated
modal emission rates for 1995-1998 model year diesel-fueled passenger cars and trucks.
                                          153

-------
Figure 93.  Modal Emission Rates for THC, for MY1995-98 diesel-fueled Cars and Trucks.

       5.0

       4.5    ...


   "C1  4.0
   -E

   "55  3.5


   Si  3.0
   (B
   re  2.0
   00

   §  L5                                                          Cars
   E  i.o
                                                                   Trucks
       0.5

       0.0

           0   1  11  12  13  14 15 16 21 22 23  24  25  27 28 29 30 33 35  37  38  39 40

                                      Operating Mode



Figure 94.  Modal Emission Rates for CO, for MY1995-98 diesel-fueled Cars and Trucks.

      90.0
      80.0
                 Cars
                 Trucks
      60.0
  01
  re  50.0
  cc

  8{  40.0
  re

  C  30.0
  re
      20.0
      0.0

           0   1  11  12  13  14 15 16 21 22 23  24  25  27 28 29 30 33 35  37  38  39 40

                                      Operating Mode
                                                  154

-------
Figure 95.
    160.0
    140.0
 ^ 120.0
    Modal Emission Rates for NOx, for MY1995-98 diesel-fueled Cars and Trucks.

         • Cars
           Trucks

    100.0
  to
  0£
  0)
80.0
 £  60.0
  c
  to
  Q)  40.0
  E
     20.0
      0.0
           0  1 11 12 13 14 15 16 21 22 23 24 25  27 28  29  30  33  35  37 38 39 40
                                    Operating Mode
Figure 96. Modal Emission Rates for PM, for MY1998 diesel-fueled Cars.
       90	-	•   •
       80	• •  •   •
   •C1  70	      -     	     	      •   •
   .22  so •   •   •           	     	      •   •
    Ol
    re  50	
    8  40 •   •  •        -   	-   	      •   •
    re
   00
    =  30	
    s                                                 *
    E  20	
       10
        0
      *,»,».»„»„»,»,,»,» . »„»,»,»,  .   .    *,*.*,.   .
      0  1  11 12 13  14 15 16 21 22  23 24 25 27 28 29  30  33 35 37 38 39  40
                              Operating Mode
                                              155

-------
3.2    Gaseous Emissions: MY2010 and Later, Particulate Emissions: MY2004
and Later.
3.2.1   Gaseous Emissions

For model years 2010 and later, we did not apply the analyses described above in 3.1. Start and
running rates for light-duty diesels in model years 2010 and later were assumed to equal those
for light-duty gasoline vehicles, as vehicles running on both fuels would be certified to the same
standards. See Table 36 and Table 37 (dataSourcelD 4910).
3.2.2  Particulate Emissions

To achieve substantially lower PM emissions, manufacturers are now equipping their diesel-
fueled vehicles (cars and trucks) with paniculate traps.

Similarly to the gaseous emissions, for MY 2004 and later, particulate emissions for light-duty
diesels were assumed to equal those for light-duty gasoline vehicles.  Thus, for these model
years, corresponding gasoline rates, as described in Chapter 2.0 above, were replicated to
represent diesel vehicles.
3.3    Particulate Emissions: Estimating Elemental and Organic Carbon
Components (EC, OC)
3.3.1   Group 1: MY 2003 and earlier

For these model years, total PM was partitioned into EC and OC components using ratios
developed for application to heavy-duty diesels. Figure 97 below, which is reproduced from the
Heavy-Duty Emissions Report59, shows the ratios.
                                         156

-------
Figure 97.  Elemental Carbon (EC) fractions running-exhaust particulate emissions, for Heavy-heavy-duty
and medium-heavy-duty diesel vehicles, by operating mode.










n -
• • *



•
m

• • 1



4 * 2 • " "
1



"

*
V




• I




»HHD
• MHD


        0  1   11  12  13  14  15  16  21 22  23 24 25 27 28 29 30 33 35 37 38 39  40
                                  Operating mode bin
3.3.2  Group 2: MY 2004 and later

For these model years, total PM was partitioned into EC and OC components using ratios
developed for application to light-duty gasoline rates.  See 2.3 And Table 42 above.
                                             157

-------
4.0   Crankcase Emissions

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 virtually all crankcase emissions from escaping to the atmosphere.

PCV valve systems have been mandated in all gasoline vehicles, both  light-duty and heavy-duty,
since model year 1969. Diesel vehicles  with turbocharged  engines , both light-  and heavy-duty
have only been required to have PCV valves since model year 2008. Thus, MOVES emission
inputs assume that all 1968 and earlier gasoline vehicles, and 2007 and earlier diesel vehicles do
not have PCV valves.

The MOBILE series of models included crankcase emission factors solely for gasoline
hydrocarbons. For purposes of MOVES, we have developed additional emission factors, as
explained below.

Crankcase emissions are calculated in MOVES by chaining the emission calculators which
calculate start, running, or extended idling emissions to a crankcase emission ratio.  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 particulate fractions organic carbon PM2 5,
elemental  carbon PM2.s, sulfate PM2.s, and sulfate  PMi0.  For each of these pollutants, the
crankcase emissions are calculated from the start, running exhaust, or extended idling emissions
of the same pollutant and then multiplying by the appropriate ratio in the
CrankcaseEmissionRatio table.

For vehicles with working PCV valves, we assume that emissions are zero.  Based on EPA
tampering surveys, MOVES assumes a 4% PCV valve failure rate.60  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% of the emission
fractions assumed for uncontrolled emissions.  While this 4% 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.

Very little information is available on crankcase emissions, especially those for gasoline
vehicles. A literature review was conducted in order to identify available data sources for
emission fractions (Table 51).
                                          158

-------
Table 51.  Selected Sources of published data on crankcase emissions from gasoline and diesel vehicles (light- and heavy-duty).
Authors
Hare and Baines61
Heinen and Bennett62
Bowditch63
Montalvo and Hare64
Williamson65
Kittelson66
Hill67
Ireson68
Zielinska69

Year
1973
1960
1968
1985
1995
1998
2005
2005
2008
x = no data
Type
Diesel
Gasoline
Gasoline
Gasoline
Diesel
Diesel
Diesel
Diesel
Diesel

# Vehicles
1
5
X
9
1
1
9
12
2

HC
0.2-4.1
33
70
1.21-1.92
50
X
X
X
X

PM(all
species)
0.9-2.9
X
X
X
35
0.038
100
25-28
20-70

CO
0.005-0.43
X
X
X
X
X
X
X
X

NOX
0.005-0.43
X
X
X
X
0.005
X
X
X

Units
% of exhaust
% of exhaust
% of exhaust
g/mi
% of exhaust
g/hp-hr
% of exhaust
% of exhaust
% of exhaust

                                                                    159

-------
Based on these sources, we estimated emission fractions for model years without mandated PCV
valves (Table 52). 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 here are largely heavy duty, and most gasoline vehicles are light duty, there is a potential
mismatch between the data sources, which is necessitated by a paucity of data. As noted
previously, model years with PCV valves were assigned emission fractions calculated as 4% of
the fractions  shown in the table.

Table 52.  Emission Fractions for Vehicles without PCV systems (percent of exhaust emissions)
Emission Type
HC
NOX
CO
PM (all species)
Gasoline
33%'
0.03%
0.005%
20%
Diesel
2%
0.03%
0.005%
20%
        The gasoline HC fraction is substantially larger than the diesel ratio. This result may be driven by
       differences between the Otto and diesel cycles, wherein the Otto cycle potentially allows a
       significantly greater proportion of combustion gases to escape to the crankcase.

The crankcase emission fractions for HC, CO and NOxmay underestimate emissions.  These
percentages of exhaust emissions are generally based on [engine- out] uncontrolled exhaust,
which is not calculated 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.

A 1995 study by Williamson65 estimated a significantly higher proportion of HC, CO, and NOX
exhaust due to crankcase than earlier works. However, Williamson tested only a single engine.
In absence of more consistent or compelling evidence, the emission fractions in MOVES rely on
the older set of data and maintain consistency with those emission factors used in the
NONROAD model. However, we note the wide range in the data sources.

Emission fractions for other fuels (LPG, methanol, etc) were set equivalent to diesel emission
factors. Emission factors for electric vehicles were set to zero.

Generally, the contributions of crankcase emissions to the overall emission inventory are
expected to decrease as additional diesel vehicles acquire PCV systems.
                                            160

-------
Appendix A: Peer-Review Comments and Response: Reviewer 1
Reviewer 1:  John M. German, International Council on Clean Transportation, Washington,
D.C.
Mr. German is a Senior Fellow and Program Director with the International Council on Clean
Transportation. He is a highly qualified expert in the areas of automotive engineering and
emissions control, whose career includes experience with both the industry and the USEPA. His
experience with the EPA includes managing the development of the US06 and SC03 test cycles
used to implement the Supplemental Federal Test Procedure, oversight of the development of the
Comprehensive Modal Emissions Model, a project conducted by engineers at the University of
California at Riverside, management of the cold-temperature CO Rule, and development of
facility cycles used in the MOBILE6 model. His industry experience includes power-train
engineering at both  Chrysler and Honda over a period of 19 years.

This document contains comments received from Mr. German following conclusion of his
review of the draft report. Following each comment, I have included our specific response,
describing whether we have accepted the comment and made corresponding revisions in the final
report, or whether we have offered a rebuttal or otherwise declined to make revisions.

Note that page and paragraph numbers listed in the comments refer to the draft document: Development of
Emission Rates for Light-Duty Vehicles in the Motor Vehicle Emissions Simulator (MOVES2009): Draft
Report. A copy of this document is included in the peer-review records, and is also available at
http://www.epa.gov/otaq/models/moves/techdocs/420p09002.pdf.
                                         161

-------
Comment Summary

In general, the MOVES draft did an excellent job of assessing emissions.  This is a very difficult
task, especially considering the lack of data in many areas.  Of course, given the lack of data and
the multitude of assumptions that have to be made, there were a number of places where
different approaches may yield better results, as discussed in my comments.

My detailed comments were written as annotations on the draft report itself. I did not have the
time or facilities to print out all 124 pages of the draft report with my annotated comments.
Thus, the detailed comments are submitted only in electronic  form. It is perfectly OK if EPA
wants to print out these detailed comments.
PEER REVIEW CHARGE QUESTIONS:

1.  For the most part, the report provides adequate description of selected data sources to allow
   the reader to form a general view of the quantity and representativeness of data used in
   development of emission rates. In some cases information about the sources of data used in
   the report were lacking. My detailed comments note these places and asks for additional
   description of the data.
2.  The description of the analytic methods and procedures was generally excellent and  allowed
   the reader to understand the steps taken and assumptions made.  There were a few cases
   where additional explanation would be helpful, as noted in my detailed comments. The
   examples chosen for tables and figures were also generally excellent.
3.  Most of the methods and procedures employed were technically appropriate and reasonable.
   However, there were some areas where alternative approaches might better achieve the goal
   of developing accurate and representative model inputs.  These are listed in my detailed
   comments.  The major areas of concern are as follows (these duplicate the summary  at the
   beginning of my detailed comments):
      a.  I  did not see anything on ambient temperature adjustments.  This is a very large
          factor.
      b.  The use of EVI data for pre 2000 vehicles needs to be validated. If I recall correctly,
          there are offsets between EVI data and the FTP and the correlation is not all that good.
          While the EVI data is needed to determine deterioration rates over time, it may not be
          a good idea to use it directly for baseline emissions.
      c.  Diesels used yet another source of data - FTP data instead of IUVP or EVI data.
          Should establish a correlation between FTP and IUVP data and apply this as  an offset
          to the FTP data.
      d.  The use of bag 2 for running emissions is not appropriate. Running emissions plus
          start emissions should equal the FTP.  Using only bag 2 completely ignores the
          running emissions from the 505 (bag 1/3). Running emissions should be determined
          by subtracting start emissions from total FTP emissions and dividing by 7.5 miles.
4.  In most cases,  EPA's assumptions when applicable data is meager or unavailable were
   reasonable.  My detailed comments note areas where different assumptions might be better.
   The major area of concern is that the modeling of PM deterioration implicitly assumed that
   PM correlated with HC (this is also in the summary at the beginning of my detailed


                                           162

-------
   comments).  However, CO is a much better predictor of air/fuel ratio than HC (reasons
   explained in my detailed comments on page 88). I would investigate how well your
   individual PM test results correlate with CO, or with a combination of HC and CO, instead of
   just assuming they correlate with HC.  If a reasonable correlation can be established, this
   would be a much better way to establish PM cold start and running emissions and to assess
   PM deterioration.
   In general, the model inputs were appropriate and are reasonably consistent with physical and
   chemical processes involved in exhaust emissions formation and control. Cases where better
   inputs could be used are listed in my detailed comments.
My most important recommendation is outside the scope of reviewing the draft report. EPA
desperately needs to have better data upon which to base the MOVES model.  Collection of
consistent data across the variety of vehicles and operation conditions would allow creation of a
much better model and help avoid all the assumptions needed in the current version.
                                          163

-------
Overall comments.

   1.  I did not see anything on ambient temperature adjustments. This is a very large factor.

RESPONSE:

    The base rates described in this report represent the temperature range of 68-86 °F, i.e., the
    "FTP temperature range ".  They are not designed to represent the effect of temperature.
    The revised report mentions this fact and refers readers to the appropriate report describing
    adjustments for temperature (and other factors).


   2.  The use of IM data needs to be validated. If I recall correctly, there are offsets between
   EVI data and FTP and the correlation is not all that good.  While the EVI data is needed to
   determine deterioration rates over time, it may not be a good idea to use it directly for
   baseline emissions.

RESPONSE:

    The I/M data used were not used as cycle aggregates. Rather, second-by-second  data were
    used, after being classified into operating modes on basis of vehicle-specific power and
    speed, as described in 1.3. Breaking down the cycle in this way neutralizes the differences
    that would be expected had we used cycle aggregate values, as inMOBILE6.


The potential offset is made worse by the use of FTP data for diesels.  Need to establish a correlation
between FTP and IUVP data and apply this as an offset to the FTP data.

RESPONSE:
    As with the data for gasoline vehicles, the FTP data for diesels was not used directly, as in
    MOBILE,  but rather to develop scaling factors that were applied to modal emission rates for
    light-heavy-duty diesels so as to represent light-duty-diesels.


   3. The use  of bag 2 for running emissions is not appropriate.  Running emissions plus start
   emissions should equal the FTP.  Using only bag 2  completely ignores the running
   emissions from the 505 (bag 1/3).  Running emissions should be  set by subtracting start
   emissions from total FTP emissions  and dividing by 7.5 miles.

RESPONSE:
                                            164

-------
    Again, we did not use FTP bag 2 emissions directly to assign emission rates for running
    operation. Rather, we used them to derive relative changes in hot-running emissions for
    vehicles in different standard levels relative to Tier 1. These relative changes, or ratios, were
    used to scale down Tier 1 modal emissions appropriately to represent vehicles certified to
    NLEV and Tier 2 standards.  We used Bag-2 emissions for this purpose because we are
    confident that the  engine is conditioned before Bag 2 commences, and because we lack a
    way to readily separate start and running emissions in Bags 1 and 3, i.e., none of the data
    available to use included Bag 1 run under hot stabilized conditions ("hot-running505 ").  In
    the revised report, this process is described in 1.3.4.2.
    We explicitly avoided the use of FTP composites, which include both start as well as running
    emissions. While the FTP standard represents the effects of control of both start and running
    emissions, the relative levels of control for start and running differ both from each other and
    from that for the composite. Generally, start emissions decline less than the standard would
    suggest, and running emissions decline more. We elected to treat start and running
    separately to account for these differences in levels of control. Thus, to represent start
    emissions, we followed the common practice of estimating the cold start as the difference
    between Bagsl and 3.  We selected Bag 2 to represent hot-running emissions because unlike
    Bags 1 and 3, it does not contain a "start increment". We did not use Bags 1 or 3 to
    represent running because it is not possible to isolate the "running component" from the
    "start increment" in these bags, except for the cold start, as described.


   4. Modeling of PM deterioration implicitly assumed that PM correlated with HC.  However,
   CO is a much better predictor of air/fuel ratio than HC (reasons explained in my comments
   on page 88). I would investigate how well your individual PM test results correlate with
   CO, or with a combination of HC and CO, instead of just assuming they correlate with HC.
   If a reasonable correlation can be established, this would be a much better way to establish
   PM cold start and  running emissions and to assess PM deterioration.

RESPONSE:

    We agree that PM, as well as CO, respond to enrichment. More generally, though, a major
    component ofPMis HC consisting of higher molecular weight compounds (about C10-12)
    including semi-volatiles and non-volatile compounds, both of which are emitted in the
    particulate phase, with particulate being formed from unburned and partially burned fuel
    components. In the cylinder, processes such as wall quenching, particularly during cold
    starts, tends to create both HC and particulate.  In addition, since the introduction of NLEV
    or LEV-I standards, targeted reductions in HC have yielded associated reductions in PM,
    but have not driven compliance for CO, which the manufacturers have easily achieved.
    Taking these factors in combination,  we suggest that while a relationship between CO and
    PM exists, the corresponding relationship between HC and PM is primary.
Chapter 1.  Light-Duty Gasoline Criteria Exhaust Emissions (HC/CO/NOx)

1.  PageS, 1.5.1 :
   Are you considering only EVI data? If so, you should state so.  If not, should state what the
   data sources are and why FTP data was not included.

                                             165

-------
   RESPONSE:


    We did not restrict consideration to I/Mdata as such, although we did require that data was
    measured on vehicles subject to I/M requirements. For example, data from the New York
    Instrumentation Protocol Assessment (NYIPA) received serious attention. These data are
    not I/M program data, but were measured on vehicles subject to the I/M program in New
    York City between 1998 and 2002. Nor did we exclude the FTP as such.  Had datasets
    measured on the FTP been available and met all requirements, we would have considered
    using them.


2.  Page 9, 1.5.1.1.1, Re: Eq 1-2:
   How were these equations derived?   Should describe or include a reference. Also, are A, B,
   and C derived directly from these equations? As A is proportional to v, B to v sq., and C to
   v cubed, it seems like these  equations are missing a step.

RESPONSE:
    We have cited a reference for these equations in the final report. They are taken from the
    "IM240 andEvap Technical Guidance", April, 2000, EPA420-R-00-007. Note that the
    squared and cubic exponents in the denominators of the equations for B and C lead to
    correct proportionality in the resulting units.


3.  Page 10,  1.5.1.3, 2nd para:
   Table 1-6  only lists about 80,000 vehicles. If the large majority of these "several million"
   are RSD or poor data, this statement probably should be revised.

RESPONSE:
    The data  sources listed in Table 1-5 in the draft report represent data determined to be
    available and potentially suitable, before we began examination to verify suitability and
    quality. Table 1-6 lists datasets that did received detailed scrutiny, and with the exception of
    the St. Louis I/M data, were also confirmed to be suitable.  The corresponding Tables in the
    final report are  Tables 8 and 9.


4.  Page 11,  1.5.1.3, Table 1-6:
   What about British Columbia, Denver, Indiana, Ohio, and Wisconsin listed in Table 1-5?
   Why were they not discussed and/or excluded?

RESPONSE:
    These data were not considered or discussed due  to a combination of quality issues or lack
    of time and people needed to process them. In the final report, we have removed references
    to these datasets in  Table 8 (formerly Table 1-5).
5. Page 18, 1.5.4.2.2, first para:
   To help the reader, might want to mention here that extrapolation to high VSP bins was
   modified based on actual high VSP data in section  1.5.5.

                                             166

-------
RESPONSE:


    We have added a sentence in this paragraph to refer the reader to the description of the
    adjustments in 1.3.3.5. (formerly 1.5.5).


6.  Page 26, Table 1-11:
   Emission rates for 81-82 are very different than for 1980 and previous.  Probably not
   appropriate to use 81-82 to represent older vehicles.

RESPONSE:

    While acknowledging the differences between the two groups of vehicles, our difficulty with
    the 1980 and older vehicles is that we lacked sufficient data to backcast their emissions to
    young ages.  Given this difficulty, and the negligible influence of the 1980 and older model-
    year group on inventory, we considered the substitution reasonable.


1.  Page 26, 1.5.5, 2nd para:
   Should provide a reference for this program [NCHRP].
RESPONSE:
    In the final report, we have cited a reference for this program: "Development of a
    Comprehensive Modal Emissions Model: Final Report", NCHRP Project 25-11, April,
    2000.


8.  Page 26, 1.5.5, 2nd para:
   Should say something about why it [MEC cycle] was developed and by whom.

RESPONSE:


    In the final report, we have added a brief description of the context and purposes for
    development of the MEC cycles.


9.  Page 27, 1st para, bottom of page:
   Typo here [incomplete sentence].

RESPONSE:
    We have corrected the incomplete sentence.


10.  1.5.6, Page 35,  1st para:
   Were the vehicle compositions the same for the migrating and local I/M vehicles? Need to
   demonstrate that the comparison did not include effects of different mix of vehicle types
   (car, light trucks)  and sizes within vehicle type. If there are effects, this should be corrected.
   This is especially  important, considering that the migrating vehicles had lower HC and NOx
   than the local vehicles.

                                            167

-------
RESPONSE:
     At the time lacked a means to verify standard levels or vehicle class, without information on
    engine family for measured vehicles. Since release of the draft, we have made progress in
    this area. It may be possible to revisit the analysis in terms of truck classes for revisions to
    be considered during the next 24 months.


11.  Section 1.5.6, Page 35, 2nd para:
   Can this data be sorted by the same age groups as the new data?  This would help the
   comparisons.

RESPONSE:
    Unfortunately, we are unable to distinguish age groups in these data, which we acquired
    from a published source.


12. Section 1.5.6, Page 35, 3rd para:
   Were the locations controlled so that they produced similar vehicle speeds and accelerations
   at the point of [remote]  sensing?  Some discussion of why these data are compatible should
   be included.  Also, were the fleet mixes  similar?

RESPONSE:
    For the data collected in the Atlanta area, the researchers attempted to select multiple sites
    with similar driving characteristics, as described in the 2004 Biennial Evaluation Report.  "
    RSD sampling sites are selected to ensure physically consistent but demographically diverse
    characteristics. Single straight lines of traffic with an average 35 mile-per-hour velocity are sought to
    facilitate single vehicle measurements and speeds that maximize measurement opportunities. Driver
    behavior and driving maneuvers are also observed at each site to ensure that remote sensing
    measurements would not be biased high by acceleration or low by coasting. " (Reference 10 in the
    Final Report). In the 2004 comparisons, the areas used were geographically continguous,
    to account for the existence of a new low-sulfur fuel requirement in the 2 5-county Atlanta
    area (131/M and 12 non-I/M counties), which suggests broad similarity in the composition
    and age of the two fleets. In addition,  before generating aggregate fleet-level results, we
    compared the I/M and non-I/M remote-sensing data  on a model-year basis, to control for
    potential differences in fleet age.

13. Page 37, Figure 1-13:
   It is troubling that the migratory vehicles had lower emissions than the local vehicles in
   Phoenix for the 0-4 age group.  Need to  explore possible biases in the data, such as vehicle
   composition, and correct if possible.

RESPONSE:
    It is possible that fleet composition could contribute  bias. As mentioned above, we lacked a
    way of assigning vehicle class and standard level while performing these analyses. Recent
    developments in this area may allow to reevaluation of this question in the future. In any
    case, the differences shown in the figure are not statistically significant for HC or CO, and
    perhaps marginally soforNOx.
                                              168

-------
14.  Page 38, 1st para:
   Does this mean that you ignored the data from the 0-4 year age group? If so, should explain
   why and if not, should explain how the 0-4 year age data was used.

   Also, how were the three data sets combined to determine the value at 7.5 years?

   Also, why is the midpoint of the 4-5 MOVES age group 5 years instead of 4.5?

RESPONSE:
     We did not combine the three datasets, but rather assigned the ratios on the basis of our
    analysis of the Phoenix data. Accordingly, we used the other two datasets for verification.
     We did tend to discount the 0-4 yr age group in the case of Phoenix. Because this program
    had afour-yr exemption period, it did not make sense  to assume that the program was
    achieving benefits for vehicles that were exempt from  testing. For Phoenix we assumed a
    ratio of 1.0 during the exemption period. However, the development of the non-I/M reference
    rates must allow for the fact that many programs have exemption periods shorter than those
    in Phoenix. For this reason we did not think it reasonable to assign no I/M difference in the
    MOVES 0-3 year ageGroup,  and performed the interpolation to estimate a difference for this
    group. In the interpolation, the value at 7.5 years was taken as the value for the 5-9 year
    group from the Phoenix analysis.
     When the midpoint of the 4-5 year ageGroup was set at 5 rather than 4.5 years is because
    this age group spans two full years. Vehicles enter this group when they turn 4, and leave it
    when they turn 6.  When they turn 5 they have been in the group for one year, and will
    remain in it for another year.  When they turn 4.5, they have been in the group only 0.5 year,
    but will remain in the group for another 1.5 years.  On this basis, we concluded that 5,
    rather than 4.5, is the actual midpoint of the ageGroup.


15.  Page 38, 2nd para:
   This implies that the 10+ data was also ignored.

RESPONSE:
    No. The statement reflects the fact that the ratios in Figure  1-13 show that the ratios in the
    10+ age class are very similar to those in the 5-9 year age class, suggesting that the ratio
    has stabilized by 10 years of age.


   You need to add explanations of how the three data sets were combined and how the data
   was translated into the lines in Figure 1-14 and the  bars in 1-15.

RESPONSE:
    As mentioned above, the datasets were not combined;  the ratios were assigned based on the
    Phoenix data, as modified by the interpolation, with the remote-sensing data used for
    verification.


   It is potentially troubling that the non-I/M ratios stabilized at only 6 years of age.  I would
   expect that emissions from vehicles in I/M areas would stabilize after a certain age due to

                                             169

-------
   being required to be fixed, but I would NOT expect emissions from vehicles in non-I/M
   areas to stabilize with age.  So, the ratios should continue to increase beyond 6 years.

RESPONSE:
    Note that the fact that the ratio stabilizes does not mean that the absolute emissions in the
    non-I/M area stabilize at six years, because the absolute emissions in the I/M area continue
    to increase until after 10 years.  Therefore, the absolute emissions in the non-I/M areas also
    continue to increase.

16. Page 40, 3rd para:
   I don't understand  this statement. The non-I/M % increases appear to be the  same for both
   modes.  Mode 27 simply has much higher emissions whether the vehicle is from an I/M or
   non-I/M area.   The I/M factor has the same effect on both.

RESPONSE:
    In percentage terms the effect is the same,  but in absolute terms the effect is greater in
    opMode 27 than in 11, simply because the emission rate is correspondingly higher.

17. Page 43, 1st para:
   Again, this seems to be reasonable for vehicles in I/M areas, due to requirement to maintain
   emissions. But is it also true in non-I/M areas?  Do you have data supporting this?

RESPONSE:
    We revised assumptions for non-I/M areas between the draft and final releases. The revised
    assumptions are described in the final report and in the response to comment #20 below.

18. Page 46, 1st para:
   For NOx,  the average emissions for the 15-19 age group were lower than for the 10-14 age
   group for your 1986-89 model year control group.  Is it appropriate to lower NOx emissions
   for higher age vehicles for model years after 1990? Probably better to assume that they
   don't change after  10-14 years.

RESPONSE:
    The apparent decline with increasing age can be seen in various datasets.  It is uncertain
    whether it is a real effect, or due to erratic behavior in sub-samples of decreasing size.  We
    have thus assumed that rates do not decline after stabilizing.

19. Page 46, Table 1-14:
   These ratios don't match the data in graphs  1-15, which shows that the NOx for 15-19 year
   old vehicles was lower than 10-14. However, the ratios in Table 1-14 are the reverse.

RESPONSE:
    Table 1-14 has been obviated by revisions since release of the draft. Its counterpart in the
    final report is Table 17.  The approach used to stabilize emissions has been modified, as
    described in 1.3.3.7 of the final report.

                                            170

-------
20.  Page 48, 2nd para:
   In section 1.5.6, you presented extensive analyses for non-I/M areas based upon Phoenix
   I/M data.  You need to explain what that data doesn't work for this section.

RESPONSE:

    The migrating vehicle sample used to develop the non-I/M reference rates was not sufficient
    in itself to allow assessment of age trends in non-I/M areas past about 10 years. For this
    reason, it was not useful to inform modeling of emissions stabilization in non-I/M areas.


   I think this is a poor assumption. Catalyst problems identified by HC and CO monitors will
   also lower NOx emissions.  On the other hand, fixing air/fuel ratio problems may well
   increase NOx emissions. Many malfunctions are being identified and repaired - and these
   repairs have impacts on NOx emissions. This is not similar to a non-I/M area. For I/M areas,
   the emission increase was not the same for all pollutants. Why would it be the same for non-I/M
   areas?

RESPONSE:
    After some consideration, we have come to agree with this comment.  We have accordingly
    revised the assumptions used to represent trends in emissions for vehicles over 15 years of
    age in the non-I/M reference rates.  In the final report, the revised assumptions are
    described in 1.3.3.7.2.  In the revisions, we assume that the relative trend observed between
    the 10-14 and 15-19 year ageGroups will persist from the 15-19 year ageGroup to the 20+
    ageGroups.  Thus, in non-I/M areas, rates continue to rise after ten years, but at lower rates
    than before ten years.


21.  Page 49, 1.5.9, 1st para:
   How are start emissions calculated?  I don't see any discussion or reference.

RESPONSE:


    Discussion of start emissions, especially for vehicles manufactured prior to 1996, was
    inadequate in the draft report.  In the final document material has been added; the expanded
    discussion has been Inserted In the new section 1.4.


22.  Page 49, 2nd para:
   Where? [is the definition of start rates?].

RESPONSE:

    In the final report, start emissions are defined in 1.3.4, and in 1.4.1.1.2.


23.  page 50, para 1:
   What simulated FTPs? All you have talked about is bag 2. Also, what relationship does bag
   2 have to start emissions? This doesn't make sense.
                                            171

-------
RESPONSE:

    In this paragraph, the "simulated FTPs" would have been more accurately referred to as
     "simulated FTP Bag 2. " The Bag 2s were simulated to describe the relative deterioration
    rate for running emissions, as described in [draft] Table 1-16. The relative deterioration
    trend for starts was then assessed in relation to the relative deterioration trend for running
    emissions, as described in the follow ing paragraphs.


24. page 50, para 2:
Are your "rates" multiplicative or additive?

RESPONSE:

    The rates are multiplicative. For light-duty gaseous emissions, deterioration is applied
    multiplicatively, and any adjustments or modifications are also multiplicative.


25.  page 52,  1.6,  1st para:
   How does the FTP data from the IUVP program correlate with the IM data used for pre-
   2000 vehicles?  If there is an offset between the IM and FTP data, this will lead to
   discontinuities in your assessment.

   Not to mention the simple correlation between IM data and real world data. You need to
   validate that EVI second-by-second data correlates with FTP data, or apply an offset factor to
   the EVI data.

RESPONSE:
    We address this question in our responses to overall comments 2 and 3 above.


26.  Page 53,  5th para (step 4):
   This may be true for vehicles certified before SFTP phase-in.  But for vehicles certified to
   the SFTP, it may not be a valid assumption.

RESPONSE:

    Based on supplementary analyses performed since release of the draft, we believe that the
    assumption holds, although uncertainty remains in estimating differing degrees of control.


27.  Page 53,  1.6.2.1, 1st para:
   This means that you are throwing away running emissions on bags  1 and 3.  Would be more
   appropriate to subtract cold start emissions from total FTP emissions, then calculated the
   running emissions over the entire drive cycle, not just bag 2.

RESPONSE:

    While it would be a desirable step, we are unaware of away to distinguish running from start
    emissions in Bags 1 and 3, to allow calculation of running emissions over the total FTP as
    you suggest.  See our response to overall comment 3 above.
                                            172

-------
28.  Page 53, 1.6.2.1, 2nd para:
   It's not HC control, its NOx control. Manufacturers have found that catalysts are most
   efficient when the air/fuel control is precisely at stoich, instead of cycling from slightly rich
   to slightly lean.  The elimination of the slightly rich events has reduced CO, as well as fast
   catalyst lightoff.

RESPONSE:
     While we don't disagree that prevention of rich events and promotion of fast lightoff would
    reduce CO, as well as NOx, we would suggest that as CO and exhaust hydrocarbons are
    both products of incomplete hydrocarbon combustion, and that the overall control strategy
    is to drive oxidation towards completion, CO control is more fundamentally linked to HC
    control than to NOx control.

29.  Page 56, 1st para:
   Or there are running emissions in bags 1  and 3, which were not evaluated.

RESPONSE:
    There are running emissions in Bags 1 and 3, of course. Incorporating them would not
    fundamentally change the relationships shown in the figure 1-23, as most of the mass of Bag
    1 is attributable to the start increment, with the imputed running component making a
    relatively small contribution.

30.  Page 56, 1.6.2.2, 3rd para:
   This is not correct. LDT3/4 had interim  standards that had the same phase-in as
   LDV/LDT1/LDV2. However, the phase in to the final Tier 2 standards for LDT3  and 4 was
   50% in 2008 and 100% in 2009.

RESPONSE:
    Since release of the draft model, we have fundamentally revised the phase-in assumptions
    based on certification records and sales figures, as described in 1.3.4.2.2 in the final report.

31.  Page 58, 1.6.2.3, 1st para:
   LDTls are rapidly disappearing; although it probably doesn't matter, as they are becoming
   LDT2s which have the  same standards.

RESPONSE:
    Revised phase-in assumptions project low fractions ofLDTl, on basis of certification
    records and sales.

32.    Page 64,  1st para:
   Your "FTP region" INCLUDES bagl/3 driving, but your running emissions don't. This
   mismatch needs to be fixed.

RESPONSE:
                                           173

-------
    We agree that the text at this point is unclear, and not reflective of what the rates represent.
    In the final report we redesignate the "FTP region" as the "hot-running-FTP" region, and
    the "SFTP region" as the "US06 region."  Under this designation, the "hot-running-FTP"
    region includes the speed and power ranges covered by the FTP Bag 2 (or the
    IM240/IM147). It is not intended to include the somewhat more aggressive driving
    represented in Bags 1/3. The revised discussion is in 1.3.3.2.4 in the final report.

33.    Page 64, 3rd para:
   You can justify this - you don't have to just assume it. The SFTP standards were calibrated
   to the Tier 1 and NLEV FTP standards. When the FTP standards were increased in
   stringency with Tier 2,  the SFTP standards were NOT increased correspondingly. Instead,
   the SFTP standards are still calibrated to NLEV levels.  Thus, the SFTP  standards are not as
   stringent and the expected reductions are less.

RESPONSE:
    We agree on this point.

34.    Page 64, 5th para:
   You have SFTP data for 2001-2003?  If so, you should discuss the data,  similar to how you
   discussed the data for vehicles prior to 2000 in section 1.5.5 and Table 1-12. Is the vehicle
   composition for 2001-3 similar to 1998-2000?

RESPONSE:
    In revisions to the rates for MOVES2010, SFTP results from the IUVP program were
    applied for the NOx rates, but not for the HC or CO rates. For each pollutant, we adopted
    the approach that appeared to  most improve verification against external data.
35.    Page 65, 2nd para:
   If I follow this correctly, this means that SFTP emissions are the same for 2010 vehicles as
   for 2005 vehicles. If this is accurate, should state this explicitly.

RESPONSE:
    This is correct. In the final report, this point is made clear in Figure 34.

   While I agree that the reduction in SFTP emissions should not track FTP reductions, as
   SFTP standards were not reduced in conjunction with Tier 2 FTP reductions, it may not be
   reasonable to assume that there is no reduction in SFTP emissions for Tier 2 vehicles.

    In the rates released with MOVES2010, we assume some reduction in Tier 2 vehicles for
    NOx, but not for HC or CO. Revisions made to the draft rates are described in 1.3.4.2.4.1.

36.    Page 69, Table 1-19:
                                           174

-------
   You have a minor problem here.  You define cold start emissions as > 720 min soak period
   minus a 10 minute soak period. But you then apply start emissions for a 10 minute soak.
   This implies that your cold start emissions are understated and should be corrected by
   adding the 10 minute soak start emissions to the emissions from bag 1 minus bag 3.


    We have neglected the component associated with the 10-min soak in the cold start rates. We
    have assumed this was reasonable in that for Tier-1 and later vehicles, we lacked data to
    estimate the start for a "0-min " soak.  We preferred not to apply the soak curve relationships
    (Figure 39) for this purpose as to use the relationships to  estimate the cold start and then to
    reuse them to estimate warm starts would introduce circular reasoning into the process.


37.    Page 69, 2nd para:
Some information on the vehicles used to derive these soak fractions would be very useful.  The NOx
start fractions suggests that these are from older vehicles - I would not expect to see higher NOx for an
intermediate start on an NLEV or Tier 2 vehicle.

    It is true that these relationships were derived for older vehicles, manufactured before 1995.
    In the final report we cite the study, released by CARB, from which they were derived:
    "Methodology for Calculating and redefining Cold and Hot Start Emissions ", California Air
    Resources Board, El Monte, CA, March, 1996 (Reference 22).


38.    Page 69, 3rd para:
   How do you handle ambient temperatures other than 75 degrees? Colder temperatures will
   dramatically change the soak fractions as a function of soak time.  For example, at 20 F a
   soak time of 240 to 360 minutes should result in a complete cold start and even a 30 minute
   soak will have a completely cold catalyst.


    Emissions for temperatures outside the "FTP range " of 68-86 F are estimated through the
    application of temperature adjustments. Adjustments are described in a separate report:
    "MOVES2010 Highway Vehicle Temperature, Humidity, Air Conditioning, and Inspection
    and Maintenance Adjustments " EPA-420-R-10-027.
    (http://www.epa.gov/otaq/models/moves/420rl0027.pdf). At present, the model does not
    have the sophistication to apply interactions of soak time and temperature. Soak
    relationships and temperature relationships are assumed  to be independent.
39.    Page 74, 1.7, 2nd para:
   Replicating gasoline data for ethanol and advanced gasoline technologies, including hybrids,
   is reasonable.  However, it is NOT reasonable to replicate the gasoline data for diesels.
   Diesels are inherently lean-burn, which changes both the emissions from new vehicles and
   how they deteriorate.
    Lacking better data on light-duty diesel vehicles under Tier 2, we have retained this
    assumption at the present.
                                             175

-------
Chapter 2  Light-Duty Gasoline Particulate Exhaust Emissions

1.      Page 84, para 3:
   Should note that at colder temperatures, additional enrichment is needed and the enrichment
   lasts longer.

    We have added a sentence to this effect.
2.      Page 88, 2nd para:
   Not a good assumption. LA92 is a higher load cycle which induces more enrichment. The
   additional enrichment should be expected to cause additional PM emissions, so there might
   not be any deterioration effects.

   The effects of enrichment can be analyzed using CO emissions.  There is a direct
   relationship between enrichment and CO from the engine - virtually all of the excess fuel is
   emitted as CO, not HC. In fact,  air/fuel ratio can be calculated from engine-out CO
   emissions. Of course, the catalyst interferes with this relationship, but during rich operation
   CO conversion efficiency drops  faster than HC conversion efficiency.  Thus, tailpipe CO is
   still a reasonable predictor of the amount of enrichment.

   So, for example, you could see if the single vehicle with significantly increased PM also had
   large increases in CO emissions. If there were large increases in CO, then it could have
   been an enrichment effect caused by the additional load on the LA92 - especially if the
   vehicle had a lower power to weight ratio.  If the CO did not increase dramatically, then you
   can be more confident that this is actually deterioration of PM.

    Notwithstanding the aggressiveness of the LA92 with respect to the FTP, the data we
    reviewed showed reasonably good and direct correlation between the two, when working
    with cycle composites, that would probably not be expected if comparing individual bags.
    Perhaps the inclusion of the longer start bag in the FTP compensates for the aggressiveness
    of the hot-running bag of the LA92.
3.      Page 89, para 2:
   Not necessarily.  Again, if the LA92 caused additional enrichment, the higher PM could be
   from the enrichment. Comparing the CO emission rates, not the HC emission rates, will
   give you a better handle on this.

    See our response to overall comment #4 above.

4.      Page 89, para 3:
   Again, you shouldn't compare just HC.  Enrichment is reflected in CO emissions, not HC
   emissions. The CO comparison is more important.
                                           176

-------
    See our response to overall comment #4 above.

5.      Page 90, para  1:
   You can't conclude this just looking at HC.  It is more important to compare CO.

    See our response to overall comment #4 above.

6.      Page 91, para  2:
   Instead of this general PM methodology, you should evaluate how well your PM test results
   correlate with CO  and/or HC emissions, after adjusting for fuel effects.  If you can establish
   a reasonable correlation, then you can adjust PM rates based on HC and CO emissions.  This
   would be especially useful in estimating PM deterioration.

    At the outset, it is not clear to us how attempting to analyze PM through correlations with
    HC and CO, and introducing associated uncertainty, would improve the resulting model
    inputs when PM measurements are available. In addition, adjustment for fuel effects simply
    not possible with the older data, as the needed fuel parameter information is not available.

1.      Page 91, para  2:
   Again, bag 2 is not representative of hot running emissions, which should be the average of
   bag 2 and bag 3.

    Bag 3 in the LA92 includes a hot-start component. As with the gaseous emissions, we lack a
    way to separate the start and running components in Bag 3. For this reason, we have
    focused on Bag 2,  rather than Bag 3, to represent hot-running emissions.

8.      Page 94, 2nd para:
   You discuss this later, but it would help the reader if you would state here that the "aged"
   data are affected both by vehicle age and model year.

    We have added text to clarify this point:  "the rates in each model-year group represent
    emissions for that group at the age of measurement ..."
9.      Page 96, 3rd para:
   Again, CO correlates better with air/fuel ratio than HC and air/fuel ratio has a strong impact
   onPM.

    See our response to overall comment #4 above.
                                            Ill

-------
10.    Page 98, Figure 2-15:
   How does a multiplicative model based on CO instead of HC look?  One based on both HC
   and CO?

    Multiplicative patterns for CO look very similar to those for HC. For reference, see Figures
    1-18, 1-19 in the draft report (Figures 18 and 19 in the final report).

11.    Page  ,2.3.1, 1st para:
   What about CO (the air/fuel surrogate)? How does this compare with PM emissions?

    See our response to overall comment #4 above.

12.    Page 100, 1st para:
   Again, HC is not the proper metric to determine when the cold start ends. CO is directly
   proportional to the air/fuel ratio and is a better predictor of closed-loop air/fuel control.

    See our response to overall comment #4 above.

13.    Page 101, 1st para:
   Again, CO will tell you exactly when the vehicle runs rich and by how much.

    See our response to overall comment #4 above.
14.    Page 106, Figure 2-23:
   Note that the PM emissions by VSP look a lot more like the CO emissions by VSP on page
   28 than the HC emissions by VSP on page 29. An indication that PM may correlate more
   strongly with air/fuel ratio (CO) than with HC.

    See our response to overall comment #4 above.

15.    Page 109, 1st para:
   This wasn't "determined" for PM. It was determined for other pollutants and assumed to be
   the same for PM.

    The comment is well taken. We have revised the text to reflect the assumption that
    deterioration trends for PM would show patterns similar to those for the gaseous emissions.
Chapter 3  Light-Duty Diesel Criteria Exhaust Emissions (HC, CO, NOx)

1.      Page 110, 1st para:


                                           178

-------
   How do the FTP test results compare to the IUVP tests used for 2000+ gasoline vehicles and
   IM data used for pre 2000 gasoline vehicles?  If the use of the FTP for in-use emissions is
   OK, then why wasn't it used for gasoline vehicles?
    It would not be relevant to compare the IUVP or I/Mdata, used to develop rates for
    gasoline vehicles, to the FTP data used to develop rates for diesel vehicles. For gasoline
    vehicles, we did make use of FTP results (by bag) when second-by-second data was not
    available (for MY 2001 +), as described above. Had second-by-second data for light-duty
    diesels been available, we would have applied it, had it been measured on the FTP or
    another cycle.

       You need to establish a correlation between FTP and IUVP data and adjust the FTP data for the
offset.
     Given the way we used the data, correlating FTP and other datasets is unnecessary, as
    previously described.


2.     Page 111, bottom of page:
   Where were these other data from? Why aren't they suitable  for determining baseline data?
   A  short explanation (even if in a footnote) would be appreciated.
    We have added a Table in the final report (Table 44) summarizing the sources of these data.


3.     Page 112, 1st para:
   Again, should subtract start emissions from total emissions to determine running emissions.
   Then regress start emissions and running  emissions against total emissions. More accurate
   and much simpler than the method in this section.
    It is unclear why the suggested approach would be either more accurate or simpler. Without
    a matchedHR505, we cannot cleanly separate the hot-start and running components, as you
    suggest.


4.     Page 112, bottom of page:
   Bad assumption -and unnecessary.
     We agree that this assumption is not appropriate, and leads to error in the resulting start
     emissions. The errors are relatively small due to the great difference between start and
     running components, particularly in Bag 1. Nonetheless, this assumption requires
     reexamination when the rates are evaluated for revision.


5.     Page 115, 1st para:
       Why is this procedure different than the soak adjustments for gasoline vehicles? Also note that
my comment on gasoline vehicle soak time also applies here - determination of start emissions is bag 3
bag 1, which implicitly assumes zero soak emissions at 10 minutes.
                                             179

-------
    The soak/start relationships for diesel engines was assessed independently from those for
    gasoline engines, given that we lacked corresponding soak/start data for diesel engines .
       Page 117, last para:
   What about CO, NOx, and PM?
   It would also be helpful to have graphs of HC, CO, NOx, and PM emissions versus VSP.
   This will help the reader compare the impacts of high load on diesel emissions to those on
   gasoline emissions on pages 28-29.
     We have added similar plots for CO, NOx andPM (Figures 90 - 92 in the final report).
Chapter 4 Crankcase Emissions

1.     Page 119, 5th para:
   Should note that the emissions are actually a percentage of engine-out emissions, but that
   vehicles before 1969 did not have catalysts, so the tailpipe correlation works.
     We have added text to clarify this point.


2.     Page 119, 6th para:
   Is it appropriate to assume the rate is 4% for all vehicles?  Should be virtually zero for newer
   vehicles, especially those still under warranty, and be higher than 4% for old vehicles, with
   some function in between.
Text has been added to clarify this point. While this 4% estimate may be pessimistic for new vehicles, and
optimistic for old vehicles, current data does not support a more detailed estimate. As older vehicles have higher
overall emissions due to deterioration effects, this may understate the impacts of crankcase emissions.  Should
additional data become available, this may be a candidate for future updates.

3.     Page  121, 1st para:
Probably not a good assumption. Diesels have much higher compression ratios and are likely to have
higher blow-by rates.

Text has been added to clarify this point. Diesel engines have both higher compression ratios and require
a tighter seal in order to operate.  Otto cycle engines potentially allow a greater proportion of combustion
gases to escape to the crankcase. As a result, it is difficult to predict whether diesel engines have higher
or lower crankcase emissions.

That being said, we agree with the commenter that it would be preferable to have data on gasoline
engines.

4.     Page 121, Table 4-2:
   Why is HC crankcase emissions 16.5 times larger for gasoline than diesel? Deserves
   explanation.

                                              180

-------
Text has been added to clarify this point.  See previous response.
                                          181

-------
Appendix B:  Peer-Review Comments and Response: Reviewer 2
Reviewer 2:  Robert A. Harley, PhD., University of California at Berkeley.

Dr. Harley is a professor in the Department of Civil and Environmental Engineering.  He has
conducted research and published extensively in the field of automotive emissions measurement
and control.

This appendix contains comments received from Dr. Harley following conclusion of his review
of the draft report. Following each comment, we have included our specific response, describing
whether we have accepted the comment and made corresponding revisions in the final report, or
whether we have offered a rebuttal or otherwise declined to make revisions.

Note that page and paragraph numbers listed in the comments refer to the draft document: Development of
Emission Rates for Light-Duty Vehicles in the Motor Vehicle Emissions Simulator (MOVES2009): Draft
Report. A copy of this document is available at
http://www.epa.gov/otaq/models/moves/techdocs/420p09002.pdf
                                         182

-------
                 Review of Draft Report

Development of Emission Rates for Light-Duty Vehicles in the
     Motor Vehicle Emissions Simulator (MOVES2009)
                    June 2009 version
                      Reviewed by
                   Robert Harley, Ph.D
     Department of Civil and Environmental Engineering
     University of California, Berkeley, CA 94720-1710
                      Prepared for


     U.S. EPA Office of Transportation and Air Quality
                     September 2009
                          183

-------
1. The draft report is missing introductory and concluding text.

Add Introduction. An introduction should be included, briefly describing motivation for
developing MOVES. It may not be obvious or known to all readers that MOVES is intended to
estimate emissions from both on-road and some off-road mobile sources, replacing the existing
MOBILE and OFFROAD modeling tools. A discussion of why a new modeling approach is
needed should be added (this might be simply a reference to other documents where more details
are available). Where should one look for other MOVES-related documents and information?
Clearly the present report is part of a larger effort, but that context is missing here.

RESPONSE:

    We have added material to orient a reader to the broader context of MOVES and its
    development, and to refer them to available sources of more detailed information.

The importance of light-duty (LD) vehicles as a source of air pollution should be summarized.
Readers may not be  aware of an earlier related report on development of MOVES2004 for
estimating greenhouse gas emissions from LD vehicles. Some relevant background and
findings on mobile source emission trends from recent studies are that:

       (a) LD vehicle  emissions of carbon monoxide (CO), hydrocarbons (HC), nitrogen oxides
(NOx), and paniculate matter (PM) have declined substantially in recent years [Harley et al.,
2006; Bishop and Stedman, 2008; Ban-Weiss et al., 2008]
       (b) in-use deterioration rates have declined for newer vehicles  [Bishop & Stedman 2008]
       (c) the effect of variations in engine load/driving conditions on emissions is not as large
as in the past. For example, Bishop and Stedman [2008] show that plots of exhaust emission
factors versus vehicle specific power (VSP) are flatter  (i.e., there is less variation in emissions
with changes in engine load) for newer vehicles.
       (d) the relative importance of other mobile sources of air pollution has increased as LD
vehicle emission control efforts have progressed. For example for NOx, diesel trucks now
dominate total on-road emissions, and there are also significant contributions to NOx from off-
road diesel-powered equipment [Harley et al., 2005; Ban-Weiss et al., 2008].

RESPONSE:

    We have added a brief summary of light-duty exhaust emissions and their control. Given the
    length of the document, we did not attempt a lengthy or comprehensive discussion, but cite
    several sources in the peer-review literature.

Provide Methodology  Overview Key features of the modeling approach for LD vehicles
emissions that should be summarized at the outset include use of g/hr (rather than g/km or g/kg)
emission factors, binning of emission factors with vehicle specific power (VSP) serving as a
master variable, changes in how I/M vs. non-I/M areas are modeled, and reliance on emissions
data from Phoenix for  pre-2000 vehicles, and manufacturer-conducted in-use vehicle emissions
testing for post-2000 vehicles.

                                           184

-------
RESPONSE:
    We have added an overview describing the content of sections and subsections in Chapter 1,
    and briefly discussing important differences between MOVES and MOBILE, including
    changes with respect to time-based rather than distance-based emission rates, and changes
    in the approach to modeling I/M.

Add Recommendations/Conclusions What are the data needs, and how will new data be
incorporated into the MOVES model to update it? What are key areas of uncertainty that would
benefit from additional study?
Most  importantly, how will MOVES be verified against independent data that were not used in
model development? In addition to doing overall comparisons of MOVES against older
MOBILE/OFFROAD models, I recommend efforts to evaluate/verify each component of
MOVES as part of the model development process. Documenting the modeling approach and
input  data can serve as a starting point for this,  but the evaluation itself (or at least plans for such
an evaluation) is missing from the present LD vehicle emissions report. If this were a manuscript
being reviewed for publication in a scientific journal, verification of the model predictions using
independent observations would be an essential component. Given the potential future
importance of MOVES predictions to national air pollution control policy, I believe that
similarly high standards should apply with respect to including model evaluation/verification in
this report.

RESPONSE:

    Since the release of the draft model and database, we have made substantial efforts to verify
    several aspects of MOVES with respect to independent data. Of course, it is very difficult
    and labor intensive to verify evaporative and start emissions, which is unfortunate, given
    their importance. Nonetheless, we have verified running emissions for light-duty (and heavy-
    duty) emissions and have identified several areas that merit attention and improvement. The
    results of these analyses will be made available in a separate report, within which data and
    research needs will be discussed.
2. There is unnecessary/imprecise use of acronyms and jargon where plain language would
be clearer and more accurate.

Light-duty vehicles (LDV) include cars as well as light trucks. The report often uses the LDV
acronym in text, figures, and tables when referring only to cars. In those situations, I
recommend using the term "cars". Use of LDV when you mean cars is confusing: for me LDV
includes light trucks as well. The report title uses the phrase "Light-Duty Vehicles" which
emphasizes the point that both cars and LD trucks are included under the heading LDV.

RESPONSE:
    We have added a paragraph describing the technical designations for "LDV" and "LDT. "
    Throughout the rest of the document, we have simplified the discussion by substituting "car "

                                           185

-------
    and "truck" for "LDV" and "LDT. "

I recommend changing the first chapter title to be "Gaseous" rather than "Criteria" pollutant
emissions. Criteria pollutant is confusing jargon. Isn't PM also a criteria pollutant? Also I
suggest merging the discussion of gaseous emissions from LD diesel vehicles (currently
section 3) together with section 1. This would provide a more consistent parallel report
structure between gaseous and PM emissions, and between gasoline and diesel engines.

RESPONSE:
    The comment is well taken, given that particulate matter is also a criteria pollutant.
    Accordingly, we have revised the document to refer to HC, CO andNOx as the "gaseous
    exhaust pollutants. "

3. Discussion of evaporative emissions and weather effects on all emissions is missing.

I did not find any discussion of evaporative emissions. If a separate report is planned to discuss
methods for representing evaporative emissions, that should be mentioned. Also climate
variables such as temperature and humidity affect vehicle emissions through increases in cold
start emissions as ambient temperature decreases, increased evaporative emissions with
increasing diurnal temperature range, and increased use of vehicle  air conditioning on hot days. I
did not find any discussion of how changes in weather affect vehicle emissions, or how such
effects are modeled in MOVES. Ambient temperature affects gaseous as well as PM emissions,
including cold start effects and gas/particle partitioning of semi-volatile organics present in the
exhaust.

RESPONSE:

    Both evaporative emissions and adjustments to exhaust emissions are discussed in separate
    reports. The revised report refers readers to these additional documents.
A potential problem with using emissions data from Phoenix to represent all pre-2000 model
year vehicles nationwide is that mild winters in Phoenix may extend vehicle lifetimes and
reduce in-use emission deterioration rates relative to other parts of the country that
experience more severe weather.

RESPONSE:

    777/5 issue was raised and considered in the FACA MOVES Review workgroup. It is difficult
    to address because datasets broad and deep enough to assess deterioration are difficult to
    locate, and where they do exist (such as remote-sensing), observed differences potentially
    attributable to "climate " are likely to be confounded by several other factors. Prominent
    confounders can include measurement differences related to instrumentation and
    calibration, differences in fuel composition, differences in I/M requirements,  differences in
    the degree of representativeness, and random error. Nonetheless, we attempted to evaluate

                                            186

-------
    the issue by comparing the Phoenix data to evaluation data from the Chicago I/M program,
    as well as data collected during the New York Instrumentation Protocol Assessment
    (NYIPA). Both the Chicago and New York data represent fleets operating in colder climates
    with harsher winters than in Phoenix. Aside from differences ostensibly attributable to I/M
    requirements, e.g., the lackofaNOx requirement in Chicago, the three programs appeared
    comparable enough to suggest that "climate " per se was not a major issue.


4. Reliance on IUVP Data Raises Concerns

Going forward, the pre-2000 vehicle model year data from Phoenix will play a minor role in
determining LD vehicle emissions, as those vehicles are 10+ years old already, and they will
constitute a declining fraction of the in-use vehicle fleet in assessments of future year emissions.

For emissions from 2000 model year and newer vehicles, the MOVES model relies on data
from the In-Use Verification Program (IUVP), a program that started in 2003 which is
administered by EPA and run by the vehicle manufacturers. Relying on IUVP data to model
vehicle emissions is a questionable approach.  The IUVP appears to have been instituted as a
regulatory compliance program, and as such may not be well-suited to capturing the full range
of vehicles, operating conditions, and emissions that are relevant to the MOVES model and
developing emission inventory estimates. For example, will the vehicle sample in IUVP be
large and random enough to ensure that major emission contributions from malfunctioning/
high-emitting vehicles will be captured? How will high-emitting vehicles be represented in
MOVES? As fleet-average emissions decline, the remaining emissions  are increasingly
dominated by contributions from high-emitting vehicles (Bishop and Stedman, 2008).

RESPONSE:

    We agree that the IUVP is not necessarily designed to obtain representative data for the
    entire "real world" fleet. But it is important to remember that similar issues apply equally
    to most sources of emissions data, including high-quality laboratory studies reported in the
    peer-reviewed literature, which very frequently use relatively small vehicle samples. This
    situation is entirely understandable, given the difficulty and expense of measurement using
    dynamometers or portable instruments combined with the difficulty of acquiring
    representative samples. Despite these questions, we found the IUVP to be a very valuable
    source, in that it provided information allowing assignment of standard level to individual
    vehicles, which we found indispensable in projecting NLEV and Tier-2 emissions. However,
    we did not take the representativeness of IUVP entirely for granted, and the approaches we
    adopted compensated in three ways. (1) We used IUVP only to estimate rates for "young"
    vehicles, aged 0-3 years. (2)  we used the IUVP to develop scaling factors that we applied to
    results from Phoenix I/M that represented Tier-1 rates. Thus the rates developed using the
    IUVP data incorporate a direct link to  "real-world" results, that probably more effectively
    represent "high-emitting " or "mal-functioning " vehicles. (3) In projecting deterioration
    from NLEV and Tier-2 vehicles, we used logarithmic variances "borrowed" from the
    Phoenix I/M data,  which are higher than those obtained directly from the IUVP.  Because
    this parameter represents the degree of skew in the distribution, increasing it effectively
    represents an increase in the fraction of "high-emitting" vehicles, with associated increases
    in mean emission rates.
                                            187

-------
5. PMEmissions

MOVES relies on data from the 2005 Kansas City PM study which focused on LD gasoline
vehicles. This was a comprehensive and well-conducted study and the resulting emissions data
are relatively current. Cold start as well as running emissions were measured. A limitation of
this study is the lack of information on how vehicle emissions for a given model year increase
with vehicle age/odometer reading.

On pp. 91-92, following equation 2-1, there is excessive precision and no associated uncertainty
reported for the multiplicative adjustment factors (0.898 and 1.972) used to give hot running and
cold start emission rates  for LD trucks. Excessive precision and lack of uncertainty estimates
concern also applies to values presented in Table 2-2.

There is no discussion of PM emission rates from LD diesel vehicles; the Kansas City study
was for LD gasoline vehicles only. Chapter 3 on LD diesel vehicle emissions covers only the
gaseous pollutant emissions, not exhaust PM. Also non-exhaust PM emissions (e.g., tire wear,
brake wear) from LD vehicles are not discussed in this report.
RESPONSE:

    Chapter 3 does cover PM emissions for light-duty diesel vehicles, as it does for
    HC/CO/NOx. See tables 42-47 in the final report.

As a model evaluation case study, EPA staff may wish to consider long-term LD vehicle PM
emission trends reported by Ban-Weiss et al. (2008) at the Caldecott tunnel in California. LD
vehicle fleet-average PM2.5 mass emission rates were measured to have decreased by 36±17%
over a 9-year time period between 1997 and 2006, due to model year effects on zero mile levels
and/or deterioration rates. Both VSP and average vehicle age (i.e., calendar year-average model
year) were similar between the two field campaigns. Cold start emissions were not measured by
Ban-Weiss et al.

RESPONSE:
    We thank the reviewer for pointing out the availability of this dataset as an opportunity to
    verify PM predictions. It could make a good candidate for a future verification effort.

References

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.

Bishop, G.A.; Stedman, D.H. (2008). A Decade of On-Road Emissions Measurements.
Environmental Science & Technology 42, 1651-1656.

                                          188

-------
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.
Environmental Science & Technology 39, 5356-5362.

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. Environmental Science &
Technology 40, 5084-5088.
                                          189

-------
References
1  USEPA Mobile Source Emissions Modeling Workgroup.  EPA's New Generation Mobile Source Emissions
Model: Initial Proposal and Issues. EPA420-R-01-007.  Office of Air and Radiation, Office of Research and
Development, Region 4. April, 2001. (http://www.epa.gov/otaq/models/ngm/r01007.pdf).

2 Koupal, I;  Cumberworth, M.; Michaels, H.; Beardsley, M.; Brzezinski, D.  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, USEPA Office of Transportation and Air Quality, Ann Arbor, MI. October,
2002. (http://www.epa.gov/otaq/models/ngm/p02006.pdf).

3 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.
(http://www.epa.gov/otaq/models/ngm/r02027.pdf).

4 Hart,C.; Koupal, J.; Giannelli, R. EPA's Onboard Emissions Analysis Shootout: Overview and Results. EPA420-
R-02-026. Assessment and Standards Division, Office of Transportation and Air Quality, Ann Arbor, MI. October,
2002. (http://www.epa.gov/otaq/models/ngm/r02026.pdf).

5 Cite latest version of SDRM.

6 Mobile-Source Emissions, Past, Present and Future: Pollutants.
(http://www.epa.gov/otaq/invntorv/overview/pollutants/index.htm').

7 Mobile-Source Emissions, Past, Present and Future: Solutions.
(http://www.epa.gov/otaq/invntory/overview/solutions/index.htm).

8 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.

9 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.

10  Harley, R.A.; Marr, L.C.; Lehner, J.K.; Giddings, S.N. Changes in motor vehicle emissions on diurnal to decadal
time scales  and effects on atmospheric composition. Environ. Sci. Technol.  39:5356-5362.

11 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.

12 Bishop, G.A.; Stedman, D.H. 2008. A decade of on-road emissions measurements. Environ. Sci. Technol.
42:1651-1656.

13  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 Paniculate Matter
from On-Road Gasoline and Diesel Vehicles. Atmospheric Environment. 42: 220-232.

14 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.

15 Gururaja, P.; Hawkins, D. Development of Evaporative Emissions  Calculations for the Motor Vehicle Emissions
Simulator (Draft MOVES2010). USEPA Office of Transportation and Air Quality, Assessment and Standards
Division, Ann Arbor, MI. July, 2011.

                                                  190

-------
16 Nam, E.K.; Giannelli, R. Fuel Consumption Modeling of Conventional and Advanced Technology Vehicles in the
Physical Emission Rate Estimator (PERE). USEPA Office of Transportation and Air Quality, Ann Arbor, MI.
EPA420-P-05-001, February, 2005. (http://www.epa.gov/otaq/models/ngm/420p05001.pdf).

17 USEPA Office of Air and Radiation, Office of Transportation and Air Quality. MOVES2010 Highway Vehicle
Population and Activity Data. EPA-420-R-10-026. Assessment and Standards Division, Ann Arbor, MI.
November, 2010. (http://www.epa.gov/otaq/models/moves/420rl0026.pdf).

18 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.
(http://www.epa.gov/oms/regs/im/r00007.pdf). (page 106).

19 Heirigs, P.; Dulla, R.; Crawford, R. W.  Processing 0/IM240 Data for Use in MOVES. SR007-05-02.  Sierra
Research, Inc., Sacramento, CA.  May, 2007.

20 Singer, B.C.; Harley, R.A.; Littlejohn, D.; Ho, I; Vo, T. 1998. Scaling of infrared remote sensor hydrocarbon
measurements for motor vehicle emission inventory calculations.  Environ. Sci. Technol. 32(21) 3241-3248.

21 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.

22
   Kish, L. 1965. Survey Sampling. John Wiley & Sons, New York.

23
   Neter, J.; Kutner, M.H.; Nachtsheim, C.J.; Wasserman, W. Applied Linear Statistical Models.  Irwin,  Chicago.
Fourth Edition. 1996.

24 Earth, 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.
(http://www.cert.ucr.edu/cmem/docs/NCHRP_Final_Report.pdf).

   McClement, D.; Dulla, R.G Identification ofNon-I/M Vehicles in I/MProgram Vehicle Emission Datasets.
Draft Report, Sierra Research, Sacramento, CA., October, 2007.

26 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.

27 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.

28  Annual Certification Test Results & Data: Cars and Light Trucks, http://www.epa.gov/otaq/crttst.htm.

29 Engine and Vehicle Compliance System (VERIFY):  http://www.epa.gov/otaq/verify/index.htm.

30 USEPA, Office of Transportation and Air Quality. Mobile Source Observation Database (MSOD): User Guide
and Reference. Assessment and Standards Division, Ann Arbor, MI. EPA420-B-08-017. December, 2008.
(http://www.epa.gov/otaq/models/msod/420b08017.pdf)

31  Enns, P.; Brzezinski, D.  Comparison of Start Emissions in the LA92 andSTOl  Test Cycles. EPA420-R-01-
025. (M6.STE.001), USEPA Office of Transportation and Air Quality, Ann Arbor, MI. April 2001.


                                                 191

-------
32 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.

33 Glover, E.; Brzezinski, D. Exhaust Emission Temperature Correction Factors for MOBILE6: Adjustments for
Engine Start and Running LA4 Emissions for Gasoline Vehicles. EPA420-R-01-029 (M6.STE.004). USEPA Office
of Transportation and Air Quality, Ann Arbor, MI. April, 2001.

34 Assessment & Standards Division. MOVES2010 Highway Vehicle Temperature, Humidity, Air Conditioning,
and Inspection and Maintenance Adjustments. EPA-420-R-10-027. USEPA Office of Transportation and Air
Quality, Ann Arbor, MI, December, 2010  (http://www.epa.gov/otaq/models/moves/420rl0027.pdf).

35 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.

37 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).

39 NamE.; Fulper, C.; Warila,  J.E.; Somers, I; Michaels, H.; Baldauf, R.; Rykowski, R.; Scarbro, C.  Analysis of
F'articulate 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.

40 Kishan, S.; Burnette, A.; Fincher, S.; Sabisch, M.; Crews, W.; Snow, R.; Zmud, M.; Santos, R.; Bricka, S.;
Fujita, E.; Campbell, D.; Arnott, P. Kansas City PM Characterization Study: Final Report. EPA420-R-08-009.
USEPA Office of Transportation and Air Quality, Ann Arbor, MI. April, 2008.

41  Gibbs, R.E.; Wotzak, G.P.; Byer, S.M.; Kolak, N.P. Sulfates and P articulate Emissions from In-Use Catalyst
Vehicles. Regulated/Unregulated Emissions and Fuel Economy. EPA-600/9-79-047. US Environmental Protection
Agency, 1979.

42 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.

43 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.

44  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.

45 Urban, C.M.; Garbe, RJ. Regulated and Unregulated Exhaust Emissions from Malfunctioning Automobiles.
SAE790696. Society of Automotive Engineers.   1979.

   Urban, C.M.; Garbe, RJ. Exhaust Emissions from Malfunctioning Three-Way Catalyst Equipped Automobiles.
SAE800511.  Society of Automotive Engineers. 1980.

47 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.
                                                 192

-------
48  Volkswagen AG.  Unregulated Motor Vehicle Exhaust Emission Gas Components. Volkswagen Aktien
Gesellschaft, Forschung und Entwicklung [Research and Development], Wolfsburg, Germany. 1991

49
    California Air Resources Board. Particulate Exhaust Emissions: Gasoline Powered Vehicles. Emissions
Studies Section, El Monte, CA. Project No. 2R8618.  1986.

50  Cadle, S.H.; Mulawa, P.; Hunsanger, E.G.; 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.

51  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.

    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.

53  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.

54  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.

     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.

56 U.S. Environmental Protection Agency. Study of Particulate Matter Emissions from Motor Vehicles.  1981.

57 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.

58 Norbeck, J.M.; Durbin, T.D.; Truex, T.D.; Matthew R. Smith, M.R.;  Characterizing Particulate Emissions
from Medium- and Light Heavy-Duty Diesel-Fueled Vehicles: Final Report.  98-VE-RT2E-004-FR.  Center for
Environmental Research and Technology, College of Engineering, University of California, Riverside, CA.
September, 1998.

59 USEPA Office of Air and Radiation, Office of Transportation and Air Quality.  Development of Emission Rates
for Heavy-Duty Vehicles in the Motor Vehicle Emissions Simulator (DRAFTREPORT). EPA-420-P-09-005.
Assessment  and Standards Division, Ann Arbor, MI. August, 2009.
(http://www.epa.gov/otaq/models/moves/techdocs/420p09005.pdf).

60 US Environmental Protection Agency. EPA Motor Vehicle Tampering Survey.  EPA 420-A-90-001. 1990.

   Hare, C. T.;  Baines, T. M.; Characterization of Diesel Crankcase Emissions. Society of Automotive Engineers,
Off-Highway Vehicle Meeting and Exhibition. MECA, Milwaukee. 1977.

62 Bennet, P.A. et al.
142A, January 1960.
Bennet, P.A. et al. Reduction of Air Pollution by Control of Emissions from Automotive Crankscase. SAE Paper
                                                  193

-------
63 Bowditch, F.W.  The Automobile and Air Pollution. SAE Paper No. 680242. Presented at SAE Mid-Year
Meeting, Milwaukee, WI. May, 1968.

64 Montalvo and Hare, 1985. EPA 460/3-84-011 Crankcase Emissions with disabledPCVsystems.  Southwest
Research Inst, San Antonio, TX. ;Environmental Protection Agency, Ann Arbor, MI. Emission Control Technology Div.

    Williamson, J.  1995.  Exhaust emissions and fuel economy evaluation of SV technologies power valve
crankcase ventilation and oil recovery system. SV Technologies report, Lake Elsinore, CA

  Kittelson, D.B.; Spears, M.; Taubert, T.; Waytulonis, B. Measurement and  Characterization of Crankcase Blow-
by Emissions: Phase I Report to Caterpillar, Inc. Center for Diesel Research,  Department of Mechanical
Engineering, University of Minnesota, Minneapolis. February 25, 1998.

  Hill, Zimmerman, Gooch.  A Multi-City Investigation of the Effectiveness of Retrofit Emissions Controls in
Reducing Exposures to Particulate Matter in School Buses. Clean Air Task Force, Boston, MA .  January, 2005.

  Ireson, R. 2008. Intentional Tracer Measurements of Tailpipe and Crankcase Particulate Matter Concentrations
Inside School Buses: Results, Refinements, and Effects of Bus Operating  Conditions . 18th CRC On-Road Vehicle
Emissions Workshop, San Diego, CA, March 31-April 2, 2008.

69 Zielinska, B.; Campbell, D.; Lawson, D. R.; Ireson, R. G.; Weaver, C.  S.; Hesterberg, T. W.; Larson, T.; Davey,
M.; Liu, L.-J. S.  2008.  Detailed characterization and profiles of crankcase and diesel paniculate matter exhaust
emissions using speciated organics Environ. Sci. Technol. 42(15): 5661-5666.
                                                   194

-------