Development of Emission Rates for
Light-Duty Vehicles in the Motor
Vehicle Emissions Simulator
(MOVES2009)
Draft Report
Protection
-------
Development of Emission Rates for
Light-Duty Vehicles in the Motor
Vehicle Emissions Simulator
(MOVES2009)
Draft 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.
,nnq
August 2009
-------
Table of Contents
1. Criteria Pollutant Emissions from Light-Duty Gasoline Vehicles (THC, CO, NOx) 1
1.1 Emissions Sources (sourceBinID) 1
1.2 Age Groups (ageGroupID) 2
1.3 Operating Modes (opModelD) 2
1.4 Scope 5
1.5. Emission-Rate development: Subgroup 1 (Model years through 2000) 5
1.5.1 Data Sources 5
1.5.1.1 Vehicle Descriptors 5
1.5.1.1.1 Track Road-Load Coefficients: Light-Duty Vehicles 6
1.5.1.2 Test Descriptors 6
1.5.1.3 Candidate Data Sources 7
1.5.2 Data Processing and Quality-assurance 9
1.5.2.1 Sample-design reconstruction (Phoenix only) 10
1.5.3 Source selection 11
1.5.4 Methods 12
1.5.4.1 Data Driven Rates 12
1.5.4.1.1 Rates: Calculation of weighted means 12
1.5.4.1.2 Estimation of Uncertainties for Cell Means: 13
1.5.4.2 Model-generated Rates (hole-filling) 14
1.5.4.2.2 Rates 15
1.5.4.2.2.1 Coast/Cruise/Acceleration 16
Means model 16
Model application 17
1.5.4.2.2.2 Braking/Deceleration 19
Means model 19
Variances model 20
Model application 20
1.5.4.2.3 Estimation of Model Uncertainties 21
1.5.4.2.4 Reverse transformation 21
1.5.4.3 Table Construction 22
1.5.5 Verification and Adjustment for High-Power Operating modes 23
1.5.6 Estimating Rates for non-I/M Areas 29
1.5.7 Stabilization of Emissions with Age 40
1.5.7.1 I/M Reference Rates 40
1.5.7.2 non-I/M Reference Rates 45
1.5.8 Deterioration for Start Emissions 46
1.6. Emission-Rate Development: Subgroup 2 (MY 2001 and later) 49
1.6.1 Data Sources 49
1.6.1.1 Vehicle Descriptors 49
1.6.2 Estimating I/M Reference Rates 50
1.6.2.1 Averaging IUVP Results 50
1.6.2.2 Develop Phase-In Assumptions 53
1.6.2.3 Merge FTP results and phase-in Assumptions 55
1.6.2.4 Estimating Emissions by Operating Mode 60
-------
1.6.2.4.1 Running Emissions 60
1.6.2.4.2 Start emissions 65
1.6.2.5 Apply Deterioration 66
1.6.2.5.1 Recalculate the logarithmic mean 67
1.6.2.5.2 Apply a logarithmic Age slope 67
1.6.2.5.3 Apply the reverse transformation 68
1.6.2.6 Estimate non-I/M References 70
1.7 Replication and Data-Source Identification 71
2. Particulate-Matter Emissions from Light-Duty Vehicles 76
2.1 Introduction and Background 76
2.1.1 Particulate Measurement in the Kansas City Study 77
2.1.2 Causes of Gasoline PM Emissions 80
2.2 New Vehicle or Zero Mile Level (ZML) Emission Rates 82
2.2.1 Longitudinal Studies 83
2.2.2 New Vehicle, or ZML Emission Rates and Cycle Effects 85
2.2.3 Aging or Deterioration in Emission Rates 90
2.2.2.4 Age Effects or Deterioration Rates 90
2.3 Modal PM Emission Rates 96
2.2.1 Typical behavior in particulate emissions as measured by the Dustrak and
Photoacoustic Analyzer 96
2.4 Conclusions 105
3. Criteria Pollutant Emissions from Light-Duty Diesel Vehicles (THC, CO, NOx) 107
3.1. Estimating Zero-Mile FTP Emissions: 107
3.1.2 Estimating Bag Emissions: 108
3.1.2 Assigning Operating Modes for Starts (Adjustment for Soak Time) Ill
3.2 Running Emissions by Operating Mode 113
4. Crankcase Emissions 116
References 119
11
-------
1. Criteria Pollutant Emissions from Light-Duty Gasoline
Vehicles (THC, CO, NOx)
This chapter describes the technical development of emission rates for criteria pollutants (HC,
CO, NOx) from light-duty vehicles for use in the draft MOVES model .
Section 1 describes the structure of the MOVES emissionRateByAge table, as it applies to
criteria-pollutant emissions from gasoline-fueled light-duty vehicles. Section 1.5 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 1.5.3) or by development and application of
statistical "hole-filling" models (subsection 1.5.4). 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 program.
Because steps 1.5.3 and 1.5.4 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
IM cycles. The development and application of adjustments to rates in operating modes at high
power is discussed in sub-section 1.5.5.
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 processes are exhaust emissions of total hydrocarbons (THC), carbon monoxide
(CO) and oxides of nitrogen (NOx), during running operation (running exhaust) (pollutantID =
1,2,3, respectively). The pollutant process is running exhaust emissions (process 01). Thus, the
pertinent values of polProcessID are 101, 201 and 301, respectively.
For these pollutant processes, the meanBaseRate is expressed in units of g/SHO, where SHO
denotes "source-hours operating."
1.1 Emissions Sources (sourceBinID)
For these pollutant processes emissions sources include light-duty vehicles (cars and trucks).
The corresponding sourceBins are defined as shown in Table 1 - 1. Note that the engine-size and
weight-class attributes are not used, as they were for energy consumption. Unlike fuel or energy
consumption, these parameters are assumed not to influence emissions, since light-duty vehicles
are required to meet applicable standards irrespective of size and weight.
-------
Table 1-1. Construction of sourceBins for Running-Exhaust Emissions for running-
exhaust emissions from light-duty vehicles
Parameter
Fuel type
Engine Technology
Regulatory Class
Model-Year group
Engine Size Class
Vehicle Test Weight
MOVES Database Attribute
fuelTypelD
engtechid
regClassID
shortModYrGroupID
engSizelD
weightClassID
Values
Gasoline = 01
Ethanol = 05
Conventional = 01
LDV = 20
LOT = 30
1.2 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 10 years.
1.3 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 passengers1. It is estimated in terms of a vehicle's speed and weight,
as shown in Equation 1 - 1.
+mvtat
1-1
m
In this form, VSP (P\,t, kW/tonne) is estimated in terms of vehicles':
speed at time t (vt, m/sec),
acceleration at (m/sec2),
- mass m (tonne) (usually referred to as "weight, "),
- track-road load coefficients A, B and C3, representing rolling resistance, rotational
00 11
resistance and aerodynamic drag, in units of kW-sec/m, kW-sec /m and kW-sec /m ,
respectively.
Note that 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.
On the basis of VSP, speed and acceleration, a total of 23 operating modes are defined for
running-exhaust processes (Table 1 - 2). 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/tonne. For reference, each mode is identified by a numeric label, the "opModelD."
-------
Table 1 - 2. Definition of the MOVES Operating Mode Attribute for Motor Vehicles
(opModelD).
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
(VSPt, kW/tonne)
VSP,< 0
0
-------
1.4 Scope
In estimation of energy consumption for MOVES2004, it was possible to combine data from
various sources without regard for the places of residence for various vehicles. In contrast, when
turning attention to the criteria pollutants, it was clear that it would be essential to know with a
high degree of confidence whether vehicles had been subject to inspect!on-and-maintenance
(I/M) requirements at the time of measurement. After reviewing data sources, it became clear
that the amounts 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 a large and valuable
source 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 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 MOBILE, in which "non-I/M" is that "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.
1.5. Emission-Rate development: Subgroup 1 (Model years through 2000)
1.5.1 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.
1.5.1.1 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 1-3.
-------
Table 1-3. 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 sourceBinID, calculate age-at-test
Assign sourceBinID
Distinguish trucks from LDV
Calculate track road-load coefficients A, B and C
1.5.1. LI
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 1 - 2.
,LHP-C,
1-2
TRLHP-c,
where:
PF.4 = default power fraction for coefficient^ at 50 mi/hr (0.35),
PF5 = 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),
ci = 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.2
1.5.1.2
Test Descriptors
-------
In addition, a set of descriptive information was required for sets of emissions measurements on
specific vehicles. Essential items for use in MOVES are listed in Table 1-4.
Table 1-4. 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 to correct result stratum
Calculate vehicle-specific power
1.5.1.3
Candidate Data Sources
In addition to the parameters listed in Table 1-3 and Table 1-4, datasets with historic depth and
large sample sizes were highly desirable, to characterize the high variability typical of exhaust
emissions as well as trends against age.
Various datasets were available, representing several million vehicles when taken together
(Table 1 - 5). In some cases they could be combined as broadly comparable pairs representing
I/M and non-I/M conditions. Likely candidates were subjected to a high degree of scrutiny and
quality-assurance, after which some were excluded from further consideration for specific
reasons.
Table 1-5. Datasets considered for use in Estimating Light-duty Runnning Emissions.
Dynamometer
I/M
AZ (Phoenix)
IL (Chicago)
MO (St. Louis)
British Columbia
CO (Denver)
Indiana
Ohio
Wisconsin
NY (New York)
non-I/M
Other MSOD
Remote-Sensing (RSD)
I/M
AZ Phoenix
IL (Chicago)
MO St. Louis
Maryland/N Virginia
CA (Los Angeles)
TX (Houston)
GA (Atlanta)
non-I/M
VA (Richmond)
GA (Augusta/Macon)
NE (Omaha)
OK (Tulsa)
Several remote-sensing datasets received consideration. However, we elected not to use remote-
sensing data directly to estimate 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 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 VOCs
relative to 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.3 (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 data4, mass rates
cannot be calculated without an independently estimated CO2 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, decel/braking or idle modes. For these reasons we reserved the RSD for secondary
roles, such as verification of results obtained from dynamometer data.
Table 1-6. 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
1995-99
2002-2004
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
Dynamometer datasets that received serious consideration are described below and summarized
in Table 1-6.
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. 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 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-screen bias" proved
impractical, 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
IM240 and an alternative transient test. It is not entirely clear whether it can be considered a
random sample, in part because estimation of representative averages was not a primary goal of
the study. All data that we accessed and used was measured on full-duration IM240s during a
four-year period. There was a high degree of replication in the conduction of tests, allowing
fully-conditioned operation to be isolated by exclusion of the initial test in a series of replicates.
While these data played a prominent role in development of energy consumption rates for
MOVES2004, the four-year duration of the program limits its usefulness in analysis of age
trends for criteria pollutants.
1.5.2 Data Processing and Quality-assurance
We performed several quality-assurance steps to avoid known biases and issues in using I/M data
to estimate mean emissions. One source of error, "inadequate conditioning" can occur when
vehicles idle for long periods while waiting in line. To ensure that measurements used reflected
fully-conditioned vehicles we excluded either portions of tests or entire tests, depending on test
type and the availability of replicates. If back-to-back replication was performed, we discarded
the first test in a series of replicates. If replication was not performed, we excluded the first 120
seconds of tests (for IM240s only).
-------
Another problem occurs when calculation of fuel economy for tests yields values implausible
enough to indicate that measurements of one or more exhaust constituents are invalid. 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 CC>2 and
NOX, and non-parametric Spearman coefficients for CO and THC.
1.5.2.1 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 1 - 1 . Within both strata, sample
vehicles then received three replicate IM147 tests.
Based on test records, reconstructing sampling rates simply involved dividing the numbers of
sampled vehicles by the total numbers of vehicles tested, by model year and calendar year, for
failing (f) and passing (p) strata, as shown in Equation 1-3.
/ _ ^f.MY.CY / _ ^p.MY.CY
/f.MY.CY ~~ ,^ /p.MY.CY ~ ir 1-3
-''f.MY.CY -'VMY.CY
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 1-4.
1 1
1-4
p..
/f.MY.CY /p.MY.CY
10
-------
Figure 1-1. 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 previous: Loaded-mode + Idle
MY 1981- 1995: IM147
MY 1996 and later: OBD II
Failing Stratum
Oversampled
"higher" sampling rate
Passing Stratum
"lower" sampling rate
Triplicate IM147
1.5.3 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 involved several difficult technical issues.
Table 1-6 shows that the datasets were of strongly differing sizes. Thus, if the datasets were
pooled without any type of relative weighting, Phoenix would have exerted much stronger
influence than the others in most shared calendar years. To rectify disparities in influence by
assigning the different datasets similar or proportional influence would have required
development of some sort of a weighting scheme, but a rational basis for such relative weighting
is not immediately apparent.
The question of pooling is further complicated by the fact that use of the Phoenix data collected
in CY 2002 to 2005 requires use of sampling weights for passing and failing tests (as discussed),
11
-------
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.
1.5.4 Methods
1.5.4.1
Data Driven Rates
Where data was present, the approach was simple. We calculated means and other summary
statistics for each combination of source and operating mode (i.e., table cell). We classified the
data by regulatory class (LDV, LDT), model-year group, age group and operating mode (Table
1). The model-year groups used are shown in Table 7, along with corresponding samples of
passing and failing tests.
Table 1-7. Test sample sizes for the Phoenix random-evaluation sample.
Model-year
groupb
1981-82
1983-85
1980-89
1990-93
1994-95
1996-98
1996
1997-98
1999-2000
Total
LDV
fail"
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
LDT
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
a Note that 'failure' can indicate failure for CO, HC or NOx, as applicable.
b Note that these are the model-year groups used for analysis; NOT the
model-year groups used in the MOVES database.
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 "ce//," to be denoted by label'//'.
1.5.4. LI
Rates: Calculation of weighted means
The emission rate (meanBaseRate) in each cell is a (£/,) simple weighted mean
12
-------
ix
where wt is a sampling weight for each vehicle in the cell, as described above, and Ri:t is the
"second-by-second" emission rate in the cell for a given vehicle at a given second t.
1.5.4.1.2 Estimation of Uncertainties for Cell Means:
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-vehicle variability5. 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 vehicle tests represented in a cell.
Measurement sample size (n,): the number of measurements in a cell representing an individual
test on an individual vehicle.
Cell sample size («&,;): the number of individual measurements 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 vehicle test represented in a cell,
calculated as the average squared deviation of measurements for a test about the mean for that
test. Thus, we calculated a separate 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
13
-------
Weighted Within-Test Variance Component (s^ ): the variance component due to variability
within tests, or the variance of measurements within individual tests (Ri>t) about their respective
test means, calculated in terms of the test variances, weighted and summed over all tests in the
cell:
1-7
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.
1-5
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
fc
E- 1-9
Note that the term CV^ is synonymous with the term "relative standard error" (RSE).
1.5.4.2 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 1
- 2). In the figure, "age holes" are represented by unshaded areas. Filling in these un-shaded
areas required "hind-casting" emissions for younger vehicles for older model years, 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 IM240, meaning operating
modes with power greater than about 24 kW/tonne.
14
-------
Figure 1-2. Model-year by Age Structure of the Phoenix I/M Random Evaluation
Sample.
MY Vehicle Age at Test (years)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
1.5.4.2.2
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.
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/tonne ( e.g., 0-3, 3-6, ... 27-30, 30+). Speed
classes followed those used for the MOVES operating modes (e.g., 1-25 mph, 25-50 mph, 50+
mph). The resulting dataset had a single mean for each test in each 6-way cell. The purpose for
this averaging was to give the resulting statistical model an appropriate number of degrees of
freedom for each of the class variables, i.e., the d.f would be determined by the number of tests
rather than the number of individual "second-by-second" measurements. Note that the matrix
used for this purpose was finer than that represented in Table 1-1.
We fit separate models in three groups of operating modes. For all operating modes except
brake/deceleration and idle, we fit one model that incorporates 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 terms of VSP (Table 1). Overall, we fit
three models for each combination of LDV and LDT, for the model-year groups shown in Table
7, 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 8). This sampling was performed to fit model on smaller amount of data that a
standard desktop computer could handle. The sample was stratified by test result and age with
allocation proportional to that of the total sample.
15
-------
Table 1 - 8. Sample sizes for statistical modeling, by regulatory class and test result
Model-year
group
1 980 & earlier
1981-82
1983-85
1980-89
1990-93
1994-95
1996-98
1996
1997-8
1999-2000
LDV
fail
663
247
pass
1,738
954
LOT
fail
346
671
pass
854
1,730
Each model included two sub-models, one to estimate means and one to estimate variances, as
described below.
1.5.4.2.2.1 Coast/Cruise/Acceleration
Means model
For the means sub-model, the dependent variable was the natural logarithm of emissions
where :
\t\Eh = natural -logarithm transform of emissions (in cell //),
Pv, Py2, Py3 = first-, second- and third-order terms for vehicle-specific power
(kW/tonne),
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/*, a 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
16
-------
speed class, whereas the power/speed-class interaction allows slightly different power slopes in
each speed class. The random factor term for test fits a random intercept for each test, which
does not strongly affect the mean estimates but does affect the estimation of uncertainties in the
coefficients.
After fitting models, we performed basic diagnostics. We plotted residuals against the two
continuous predictors, VSP and age. We checked the normality of residuals across the range of
VSP and age, and we plotted predicted vs. actual values.
Variances model
The purpose of this sub-model was to model the variance of 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 1 - 8). 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
+ s
1-11
where/? 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 1-9.
Table 1-9. 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 groups1
Operating modes
Age Groups
Pollutant processes (running HC,
CO, NOx)
TOTAL cells
MOVES Database attribute
fuelTypeID = 01
regClassID = 20, 30
opModelD = 11-16, 21-30, 33-40
ageGroupID = 3, 405, 607, 809,
1014, 1519,2099
polProcessID = 101, 201, 301
17
-------
Next, we constructed a vector of coefficients for the means sub-model (P) and merged it into the
cell matrix.
p = \j30
1-12
Then, for each table cell, we constructed a vector of predictors (X/,). Equation 13 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 10.
3al 00 Pv\
1-13
Table 1 - 10. 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
32.0
0.5
9.0
The final step was to multiply coefficient and predictor vectors, which gives an estimated
logarithmic mean (InE/j) for each cell h.
1-14
18
-------
The application of the variances model is similar, expect that the vectors have four rather than
nine terms
a = [a0 ala2a3] i- is
X, = [l Pv a Pva] i -16
Thus, the modeled logarithmic variance in each cell is given by
sih=XA« i -17
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 LDV, and the 1997-98 model-year group for LDT.
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 a slope from the available data would give 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 were recalculated by rearranging Equation 1 - 10 to evaluate the model in operating
mode 24, using the age slope from the previous model-year group (/?4*) 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/tonne and the speed class is 25-50 mph.
A* = \nEat -7.5A -7.52jB2 -7.53/?3 -/?4a*-/?5(25_50) -7.5/?6 1 - 18
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.
1.5.4.2.2.2 Braking/Deceleration
Means model
We derived models similar to those one used for coast/cruise/acceleration. For these operating
modes, 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:
19
-------
In Eh=0Q +f3la + y1tj+s l -19
Variances model
In addition, we fit variances models for these operating modes, which were also simple functions
of age.
+ aa + s 1-20
Model_application
In these operating modes, rates were to be modeled for a total of 840 cells. This total is
calculated as in Table 1-9, except that the number of operating modes is 2, rather than 21. We
set up coefficient and predictor vectors, as before.
For the means model the vectors are
and
X,=[l a] 1-22
respectively.
For the variances model the coefficients vector is
a = [a0 a^ ] 1-23
and the predictor vector is identical to that for the means model.
As with CCA 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
InEh calculated from the sample data for the youngest available age class. In this case, equation
24 is a rearrangement of Equation 1-19.
-/7a* 1-24
20
-------
After these steps, the imputed values of InE/, were calculated, as in Equations 1-14 and 1 - 17.
1.5.4.2.3
Estimation of Model Uncertainties
We estimated the uncertainty for each estimated InE1/, in each cell. During each model run, we
saved the covariance matrix of the model coefficients (sp2). This matrix contains covariances of
each of the nine coefficients in relation to the others, with the diagonal containing variances for
each coefficient.
'0,4
°"5(0-25)
°"5(25-50)
'6,4
°"5(50+)
1-25
Using the parameter vectors X/, and the covariance matrix sp2, the standard of error of estimation
for each cell was calculated as
3lnE,
1-26
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 cell6. 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 covariances represent the uncertainties in each coefficient in relation
to itself and the others.
1.5.4.2.4
Reverse transformation
To obtain an estimated emission rate Eh in each cell, the modeled means and variances are
exponentiated as follows
F - e
^h ~ e
1-27
21
-------
The two exponential terms use the results of the means and variances sub-models, respectively
(Equations 1-6 and 1 - 7). The left-hand "means" term represents the geometric mean, or the
center of the implied log-normal distribution, whereas the right-hand "variance" term reflects the
influence of the "high-emitting" vehicles representing the tail of the distribution.
The estimate of 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 1
- 19 to solve for s2. If the mean of emissions data is xa and mean of In-transformed data is x/,
then the logarithmic variance can be estimated as
2 (x ^
sf=2\n -f- 1-28
In practice one of these options was selected based which most successfully provided model
estimates that matched corresponding means calculated from the data sample.
The uncertainties mentioned above represent uncertainties in InE1/,. 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 1-28
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(s2E ), as in Equation 1-8.
Finally, we calculated the CV-of-the-mean (CV^) for each modeled emission rate, as in
Equation 1-9.
1.5.4.3 Table Construction
After compilation of the modeling results, the subset of results obtained directly from the data
(Equations 1 - 4 to 1 - 9), shaded area in Figure 1-1) and the complete set generated through
modeling (Equations 1 - 10 to 1 - 28) 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 («/, > 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 11.
22
-------
Table 1-11 Mapping 'Analytic" Model-Year Groups onto MOVES database Model-Year
Groups.
"Analytic"
LDV
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
1999-2000
1999-2000
LOT
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
1999-2000
1999-2000
"MOVES database"
1980 and previous
1981-82
1983-84
1985
1986-87
1988-89
1990
1991-1993
1994
1995
1996
1997
1998
1999
2000
1.5.5 Verification and Adjustment for High-Power Operating modes
The rates described were derived from data measured on the IM240 or IM147, 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/tonne 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/tonne.
Based on comments from members of the FACA MOVES Review Workgroup, 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 (NCFtRP) 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 be representative of any specific driving pattern but rather to exercise vehicles
through a wide range of speed and acceleration. Driving traces for both cycles are shown in
Figures 3 and 4. Both cycles range in speed up to over 70 mph and in VSP up to and over 30
kW/tonne.
23
-------
Figure 1 - 3. Example Speed Traces for the US06 and MEC Drive Cycles.
80 f
200 400 600
800 1000 1200
Time (sec)
Drive CyolQ rrec us06
1400 1600 1800 2000
Figure 1 - 4. Example VSP Traces for the US06 and MEC Cycles
-30
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time (sec)
Drive Cycle msc us06
Table 12 summarizes the numbers of available tests by regulatory class, model-year group and
drive cycle, had a certain number of tests, as shown in Table 12. Samples were somewhat larger
for LDV for both cycles, which represented a broad range of model-years.
24
-------
Table 1 -12. Sample Sizes for US06 and MEC Samples (No.
Model-year group
1 980 & earlier
1981-85
1986-89
1990-93
1994-95
1996-99
Total
LDV
US06
4
15
21
54
49
58
201
MEC
14
23
24
57
45
28
191
LDT
US06
8
13
22
22
56
121
MEC
6
19
31
36
30
17
139
Total
24
65
89
169
146
159
652
tests)
Figure 1 - 5 to Figure 1-7 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 1-5. CO emissions (g/ssec) on Aggressive Cycles vs. VSP, by Regulatory Class and Model-
year Group.
£.-
i
5-
(H
-10
Reg/MYG
LDT-(X>80
LDT-9495
LDV-8689
20
VSP (kW/tonne)
1 LDT-8185
1 LDT-9699
LDV-9093
LDT-8589
LDV-0080
LDV-9495
LDT-90S3
LDV-8185
LDV-9699
Figure 1-6 THC Emissions (g/sec) on aggressive Cycles vs. VSP, by Regulatory Class and Model-
year Group.
25
-------
0.15:
0.14-
0.13-
0.12-
0,11-
0.10-
0.09-
0.03-
0.07-
0.06-
0.05-
0.04-
0.03-
0.02-
0.01-
0.00;
10 20 30
VSP (kW/tonne)
Reg/MYG
LDT-OCeO
LDT-9495
LEW-8689
1 LOT-8185
1 LOT-9699
LCW-9033
LDT-8689
LDV-O:«)
1 U3V-9495
LDT-9:83
LCV-8185
LW-9699
Figure 1-7. 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-
0.14-
0.13-
0.12-
0.11-
0.10-
0,09-
0,09:
0,07-
0,06-
0,05:
0,04-
0,03-
0,02-
0.01-
0,00-
10 20
VSP (kW/tnnne)
Reg/MYG
LDT-0080
LOT9495
LOW-6689
LOT-8185
LOT96S9
LCV-9093
1 LDT-868J
LDV00«J
LDV-9495
LDT-9093
LD\/8185
LCW-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 1 - 12. After averaging, we
calculated ratios firm 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/tonne. With mode 27 as a reference, we calculated ratios to
modes 28, 29 and 30
26
-------
77
,,27=--, for/ =28,29,30 1-29
and with mode 37 as a reference, we calculated ratios to modes 38, 39 and 40.
^,37=-^-, for/ =38,39,40 1-30
Calculating uncertainties in the ratios was an important step. If the ratio R is calculated as a
numerator divided by a denominator (MD), the variance in the ratio is propagated by summing
the products of the squared partial derivatives of R to N and D and the variances of their
respective means.
+
dN N (dD
Note that this expression contains only variances and neglects potential covariances between N
and D. In the case ofN/D the partial derivatives, which express the sensitivity of the ratio to
each, are simply calculated as
dR 1 , dR -N
= and = - 1-32
dN D dD D2
To apply the ratios to operating modes 28-30 and 38-40, we calculated ratio-based emissions
estimates (ER) as the products of their respective ratios and the initial rate for modes 27 or 37
pR - n ^initial pR _n ^initial
^h,i - ^i:27-^h,27 ' UI ^h,i ~ ^i-37^h,37 1 - JJ
respectively, where Ehmtial is the initial data-driven or model-generated rate calculated as
previously described.
We used the variances of the ratios to calculate upper and lower confidence limits on the ratio-
based rates.
1-34
R _ 77^ , ^
In applying the confidence band as an evaluation criterion, each of the high-power operating
modes /', the initial value E'mtml was a candidate for replacement by ER if it fell outside the 80%
confidence band of ER, or
EMtial UCL* 1-35
27
-------
Due to volatility in the ratios, the confidence limits were quite wide in some cases. After initially
calculating and evaluating a 95% confidence band, we settled on using a somewhat narrower
80% band, for the reason that it was more sensitive in identifying implausible values ofE'mtml,
whether high or low.
We present some examples below. In the THC example (Figure 1 - 8), 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. The example
for CO is different (Figure 1 - 9). The initial values for modes 28-30 all fall within the
confidence intervals 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. Finally, in the NOx example (Figure 1-10),
the initial rates are replaced in five out of six cases. The initial values for 28-30 and 40 all fall
below the LCL, whereas that for 30 falls above the UCL.
Figure 1 - 8. THC emission rates (g/hr) vs. Operating Mode for MY-1998 LDV at ages 6-7: initial
(data or statistical model) and calculated by ratio based on aggressive cycles.
70 -
fin
en
Af\
on -
Tfl _
1 n
n -
c
/
2
T
**,.*»**«***'
T
'>*
...**
0 11112 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
Figure 1-9. CO emission rates (g/hr) vs. operating mode for MY-1998 LOT at ages 6-7: initial
(data or statistical model) and calculated by ratio based on aggressive cycles.
3,UUU
n
1
i
i
p
T
I
r
J*
/+
« tf
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
28
-------
Figure 1 - 10. NOx Emission Rates (g/hr) vs. Operating Mode, for MY1995 LDV at ages 8-9: initial
(data + statistical model) and calculated by ratio based on aggressive cycles.
n
^
i
[.
*I
/"
1 /
* *
»** »»**
Initial
D ratio
0 1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
1.5.6 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 inspection 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" (meanBaseRatelM) and the other represents
rates under "non-I/M conditions" (meanBaseRate). The first set, representing vehicles subject to
I/M requirements, we call the "I/M reference rates". The second, representing vehicles not
subject to I/M requirements, we call the "non-I/M reference rates."
For the I/M reference rates, the term "reference" is used because the rates represent a particular
program, with a specific design characteristics, against which other programs with differing
characteristics can be modeled. Thus, the I/M references are, strictly speaking, regional rates, and
not intended to be (necessarily) nationally representative. Development of the I/M reference rates
is discussed above in sections 1.5.2 and 1.5.3. 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.
29
-------
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 1 - 1 l(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 1 - 1 l(b)).
Figure 1 -11. Basic 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
Area
For convenience, we refer to the first approach as the "between-area" approach, and the second
as the "within-area" approach. Neither approach attempts to measure the incremental difference
attributable to a program from one cycle to the next.
The approach we adopted emphasizes the "within-area' approach, based on a sample of vehicles
"migrating" into Phoenix. 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
30
-------
had not yet been influenced by the program. The specific criteria to identify particular migrating
vehicles are presented in Table 13.
Table 1 - 13. 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
inAZ
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
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 program7.
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 1-12 Geographic Distribution of Vehicles Migrating into the Phoenix I/M Area, 1995 -
2005.
West East
South Central' South Central
-------
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 36. 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.
We propagated uncertainty for these ratios as in Equations 1-31 and 1 - 32.
Ratio = - i . 36
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, by T. Wenzel, 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 his results for our
purposes, we converted averages for migrating and program fleets into ratios as in Equation 1 -
36.
A another valuable source for comparison was remote-sensing data collected in the course of the
Continuous Atlanta Fleet Evaluation (CAFE) Program8'9. 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 1-13. 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
32
-------
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-EVI 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. A 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.
33
-------
Figure 1 -13. 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).
0-4
5-9
Age Class
10+
I AZ I/M '//,. GA RSD (CY04) = VA RSD (CY04)
1.80
lAZI/M BGARSD(CY04) = VA RSD (CY04)
0-4 5-9 10+
Age Class
i AZ I/M GA RSD (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 have shorter exemptions; for
example, both the Atlanta and N. VA programs have two-year exemptions.
34
-------
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 age Group to a value of 1.0 and 0 years of age, as shown in Figure 1 - 14.
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 1 - 14. 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.
MOVES AgeGroup
Analysis AgeGrou p
\geGrou
o
0-3 years 4-5 years 6-7 years 8-9 years
0 - 4 years Nj; 5 - 9 years i
Figure 1-15 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 6-10 years.
35
-------
Figure 1 - 15. 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
i
0.60
0.40
0.20
0.00
(a) CO
0-3
4-5
6-7
8-9
Age Class
10-14
15-19
20+
0.00
(b) THC
0-3
4-5
6-7
8-9
ge Class
10-14
15-19
20+
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
Age Class
10-14
15-19
20+
The ratios shown in Figure 17 are applied to the I/M reference rates to derive non-I/M reference
rates.
1-37
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 EM/M.
1 ^8
1-38
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 given the additional assumptions involved
in developing the non-I/M reference rate.
Figure 1-16 shows an example of the reference rates by operating mode, for all three pollutants,
with error bars representing 95% confidence intervals. Note that not all the modes are shown, to
allow examination of differences between non-I/M and I/M rates at lower VSP. It is
immediately apparent that uncertainties are considerably larger for the non-I/M rates, which
reflects the uncertainties in the non-I/M:I/M ratios, in relation to the relatively small uncertainties
in the I/M references derived directly from data. The very large uncertainties in high-power
operating modes (28-30, 39) reflect the combined uncertainties in the high operating mode ratios
(see 1.5.5) and the non-I/M ratios.
Figure 1-17 shows corresponding trends by age for two operating modes. The first is opmode
11, (speed = 1-25 mph, VSP <0 kW/tonne) and 27 (speed = 25-50 mph, VSP = 12-18
kW/tonne). As before, the uncertainties are visibly larger for the non-I/M rates. Trends level off
at in the 10-14 year age Group. An obvious observation is that the I/M difference is much larger
in the more aggressive mode (27) than in the less aggressive one (1 1), with the inference that I/M
differences will be more strongly expressed for more aggressive than less aggressive driving.
37
-------
Figure 1 -16. Non-I/M and I/M Reference Rates by Operating Mode (Example: LDV, MY 1994, at 8-9 years
of age)
4T1
3C _
35
= -^n
aj0
5 n0
!ft 20
: 11;
i lj
c _
n -
/0\ T\-\(~*
(a) mu
j
f *
V '
*
1
9f *
1
4
1 ]
1 "
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
n -
_. . _
non-I/M Reference
^ T
f| * ;
»fi »P*
1
^
i
7
i
i
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
n -
.
T "
if
u *
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
38
-------
Figure 1 -17. Non-I/M and I/M Reference Rates vs. Age for Two Operating Modes (Example:
LDV, MYG 1994).
10 15
Operating Mode
200 :
-O--IM Ref:
-D --non-yp
IM Ref:
non-I/M
/l->\ /"*/~\
(D) U(J
J=*
B B
0 5
Mode 11
Ref: Mode 11
Mode 27
Ref; Mode 27
y
T -11
T ^^^^
jS ^4
fr
^
- - --H
= e-- - =
.=-&----&''"'"
10 15 20 2
Operating Mode
200
150
100
50
O--IMRef: Mode 11
a- non-I/M Ref: Mode 11
-IMRef: Mode 27
-non-I/M Ref: Mode 27
(c) NOx
I I
10 15
Operating Mode
20
25
39
-------
1.5.7 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 around 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 studies10. Figure 1-18 and
Figure 1-19 show age trends by model year for LDV and LDT, respectively. The values shown
are aggregate mass rates over the IM147 expressed as g/sec for CO, THC and NOx.
1.5.7.1 I/M Reference Rates
From Figures 11 and 12, as well as Figure 1, it is clear that no data was 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 10-12 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 the down-turn as the trend levels off and stabilizes. To compensate for this limitation,
we employed a simple approach to represent the decline and stabilization of the rates.
40
-------
Figure 1 -18. Aggregate IM147 Emissions (g/sec) for LDV, by Model year and Age, for the
Phoenix Random Sample.
LDV WEIGHTED
CO vs. Age (years)
Vehicle age (years)
LDV WEIGHTED
THC vs. Age (years), LDV
Vehicle age (years)
LDV WEIGHTED
NOx vs. Age (years), LDV
\fehicle age (years)
o o Q ises
41
-------
Figure 1 -19. Aggregate IM147 Emissions (g/sec) for LOT, by Model year and Age, for the
Phoenix Random Sample.
LOT WEIGHTED
CO us, Age (years)
Vehicle age (years)
R ?1 S3 8 W
1984 000 1935
1930 1991
2000 A A A 2001
LOT WEIGHTED
THC vs. Age (years), LOT
(2 - JU f2 ;
\fehicle age (years)
LDT WEIGHTED
ISOx vs. Age (years), LOT
(c) NQx
Vehicle age (years)
1983 '. 1984 O O O 1935
1989 ODD 1990 LJ £3 -_3 1991
1&54 &-A-4 1995 A A A 1995 A A A 1997
20CO - 2001 I I I 2002 I I I 2003
42
-------
We calculated ratios of means between the 15-19 and 10-14 year age Groups for the model-year
group 1986-89, which contains data for vehicles as old as 19 years. For this purpose we used
Phoenix data aggregated by MOVES model-year and age groups, as shown in Table 1-15.
For this purpose, we used aggregated tests (g/mi). After averaging by model-year group and
ageGroup, we calculated ratios of means between the 15-19 and 10-14 ageGroups.
n _
~
'86-89,15-19
1-39
We calculated modified rates for the 15-19 and 20+ age Groups as the product of the rate for the
10-14 ageGroup and the resulting ratio (.Rage, Error! Reference source not found.). The
resulting rate was the same for 15-19 and 20+. We calculated variances for the ratios, but did not
propagate the uncertainty through to the final result.
Table 1 -14. Age-Group Ratios (Rage) between the 15-19 and 10-14 ageGroups (MYG
1986-89)
Regulatory Class
LOT
LDV
CO
1.22
1.15
THC
1.19
1.14
NOx
1.08
1.00
43
-------
Table 1 - 15. Aggregate IM147 Emissions (g/mi) by Model-Year Group and Age Group.
35.0 ^
10 15
Age (years)
o.o
pre-80
81-82
83-85
86-89
90-93
94-95
96-98
99-00
10 15 20
Age (years)
25
pre-80
81-82
83-85
86-89
90-93
94-95
96-98
99-00
10 15
Age (years)
20
25
44
-------
1.5.7.2
non-I/M Reference Rates
The ratios developed in 1.5.7.1 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 specific to 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 the program in some way. However, in the
absence of a program, high-emitting vehicles are not identified and owners have little 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
are collected within I/M areas. But one possibility exists. As mentioned, we analyzed and
considered data fromt the Chicago program for this project. A characteristic of this program is
that vehicles were evaluated only on results for HC and CO; NOx was measured in some lanes
but did not inform test results. Thus, with respect to NOx, we considered the Chicago data as a
rough surrogate for a non-I/M area. It was therefore helpful to compare NOx results in Chicago
to those in Phoenix, on an aggregate basis, as shown in Figure 1 - 20. In the figure, NOx
emissions for the 1990-93 model-year grouip appear to be increasing at a higher rate in Chicago
than in Phoenix between the ages 8 to 12.5 years. Based on this increment (15%), we assumed
that emissions for all pollutants would increase by 15% between the 15-19 and 20+ year
ageGroups. Note that the effects of this adjustment can be seen in the non-I/M series in Error!
Reference source not found, above
Figure 1 - 20. Aggregate IM147 NOx Emissions for LDV in Phoenix (unlabeled) and Chicago (IL)
10
Age (years)
45
-------
1.5.8 Deterioration for Start Emissions
Because MOBILE assumed that start emissions would deteriorate, based on analyses performed
at the time of its development, we thought it reasonable to include start deterioration in draft
MOVES. Due to the complete lack of data in this area, we elected to model start deterioration as
a function of running deterioration.
As described, start rates are defined in terms of the FTP, but the rates for running are represented
by operating mode. However, the modal approach enables simulation of multiple driving
patterns, as represented by test cycles, such as the Federal Test Procedure (FTP), or the US06
cycle, which represents high-speed driving in the Supplemental Federal Test Procedure (SFTP).
Any driving cycle can be represented as a weighted average of the MOVES emission rates and
the "operating mode distribution" for that cycle. In this case we developed an operating-mode
distribution for the "hot-running" phase of the FTP (Bag 2). This phase is an 860 second long
trace that represents urban driving over a 3.9 mile route after the engine has stabilized at its
normal operating temperature. We estimated an operating-mode distribution using the "Physical
Emission-Rate Estimator" (PERE)2. This distribution, shown in Error! Reference source not
found., represents a typical LDV, with an engine displacement of 2.73 L and test weight of
3,350 Ib. Estimating the Bag 2 emissions simply involves calculation of averages of the emission
rates weighted by the mode distribution.
Table 1 -16. Operating-mode distribution for a typical LDV on the hot-stabilized phase of
the FTP (Bag 2)
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
Total
Time in Mode
(seconds)
97
155
77
121
83
59
22
4
42
111
62
18
7
2
1
861
Time in Mode (%)
11.3
18.0
8.9
14.1
9.6
6.9
2.6
0.50
4.9
12.9
7.2
2.1
0.80
0.2
0.1
100
46
-------
After calculation, the simulated FTPs were converted to ratios by dividing the value for each age
Group by the value for the 0-3 year age Group.
A major uncertainty in our knowledge of start emissions is whether it is reasonable to assume
that start emissions would deteriorate at the same (relative) rate as running emissions. Given the
lack of data in this area, we adapted assumptions applied in the MOBILE model. In MOBILE we
assumed that start emissions for NOx would deteriorate at same relative rate as those for
running,11 but that HC and CO emissions would deteriorate at lower relative rates.12 Thus, we
adapted the equations for "normal" and "high" emitters applied in MOBILE and developed a set
of ratios for each age Group that reduce HC and CO start deterioration relative to the running
deterioration, as shown in Error! Reference source not found.. Final ratios after application of
the relative reductions are shown in Figure 1-21. Deteriorated start rates are developed (in all
operating modes) by multiplying the rate in the 0-3 age Group by the ratio for each successive
age Group.
NOTE: if possible, these assumptions will be reviewed, and considered for revision before
release of the final model.
Table 1 -17. Ratios Expressing Relative Reduction in Start Deterioration, relative to
Running Deterioration, by Age Group.
ageGroup
0-3
4-5
6-7
8-9
10-14
15-19
20+
HC
1.00
0.58
0.47
0.41
0.36
0.36
0.36
CO
1.00
0.57
0.46
0.39
0.33
0.33
0.33
47
-------
Figure 1 - 21. Example: Start Deterioration Ratios Applied to MY 2001-2021.
10 15
AgeGroup Midpoint (years)
20
25
10 15
AgeGroup Midpoint (years)
10 15
AgeGroup Midpoint (years)
48
-------
1.6. Emission-Rate Development: Subgroup 2 (MY 2001 and later)
1.6.1 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.
1.6.1.1
Vehicle Descriptors
In addition to the parameters listed above in Table 1-3. Required Vehicle Parameters., the
IUVP data provides engine-family information. Using engine family, the IUVP files can be
merged with certification logs by model year. The certification logs provide information on Tier
level and specific emissions standards applicable to each vehicle. The Tier 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 Tier
Level.
Table 14. Vehicle Descriptors Available in IUVP files and Certification Logs
Parameter
VIN
Fuel type
Make
Model
Model year
Engine Family
Tier
Emissions Standard
Units
Source
IUVP
Y
Y
Y
Y
Y
Y
Cert. Log
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 properly associated with
the correct Tier Level and emissions standard, which allows inference of the correct vehicle class
(LDV, LDT1, LDT2, LDT3, LDT4).
49
-------
1.6.2 Estimating I/M Reference Rates
The goal of this process is to represent I/M reference rates for young vehicles, i.e., the first age
Group (0-3 years). The rates are estimated by Tier, model year and regulatory class. The process
involves six steps.
1. Average IUVP results by Tier and vehicle class
2. Develop phase-in assumptions for MY 2001 - 2021, by Tier, 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. For start emissions, use weighted average FTP starts
directly.
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 remaining six age Groups. We assume that
NLEV 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.
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 16).
Each of these steps is described in greater detail in the sub-sections below.
1.6.2.1 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 1-22
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 effect
of HC control on CO emissions, whereas CO standards were not stringent enough in themselves
to substantially reduce emissions.
50
-------
Figure 1 - 22. Composite FTP Results for Tierl, NLEV and Tier 2 Vehicles, as measured by IUVP,
in relation to certification and useful-life standards.
0.350
0.300
0.250
0.200
0.150
0.100
0.050
0.000
n
A.
(a) THC
'\n
q\ n.
\e->. /;/
^v*- O-'
Xp^*
- D - - Useful Li fe Stan dards
.n .n
1 =
In In
m -) s
r
iD i 5
8
0 5 -
nn
n -a- p
/ \
n -n- p . ; q \
(b)CO \
\n-' '-b a a
. b b D n
^ *v
N^ ^
^*^^x ±
Tl TLEV LEV ULEV bin5 bin4 bin3 bin2
0.70
- - o - Certification
--E3-- Useful Life
*Composite
0.00
51
-------
Figure 1 - 23. 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).
0.40
0.35
0.30-
0.25-
0.20
0.15-
0.10-
0.05
0.00
(a) THC
s
15
Tl TLEV LEV ULEV bin5 bin4 bin3 bin2
0.30
0.25 -
0.20
0.15
0.10
0.05
0.00
52
-------
Figure 1-23 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, expressed 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. 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 not as markedly so. In any case, the results show that use of the
composite results could give misleading results in projecting either start or running emissions.
1.6.2.2 Develop Phase-In Assumptions
For rates stored as MOVES defaults, we developed assumptions intended to apply to vehicles
sold in states where Federal, rather than California standards applied. Thus, the phase-in
represents the phase-in of NLEV and Tier-2 standards. To construct the default scenarios, we
divided the vehicle classes into two groups. The first group includes LDV, LDT1 and LDT2, to
which NLEV standards applied; the second includes LDT3 and LDT4, which transitioned
directly from Tier 1 to Tier 2.
Assumptions for LDV, LDT1 and LDT2 are shown in Figure 1 - 24. For these classes, The
transition between Tier 1 and NLEV is abrupt, occurring between 2000 and 2001. For the first
two model years, the fleet is a mixture of "transitional LEV" (TLEV) and LEV, and entirely
LEV for the following four model years. However, a three-year phase-in of Tier 2 began in 2003,
and was complete by 2007, after which the Federal fleet is entirely Tier 2. The breakdown of
Tier 2 Bins during the transition is shown in Figure 1 - 25. This scenario reflects a tendency for
Bin 5 to dominate the fleet by 2007, followed by Bin 3, and with Bins 4 and 2 playing minor
roles.
For the second group, LDT34, the transition from Tier 1 to Tier 2 lasts three years, as shown in
Figure 1-26 and Figure 1 - 27. Like the LDV group, the transition to Tier 2 is complete in MY
2007. Tier 2 vehicles are primarily represented by Bin 8 for the first several years, after which
the heavier LDTs also move into Bin 5. For both groups, we assume that all phase-in fractions
are stable between MY2010 and 2021.
53
-------
Figure 1
1.00
0.90
0.80
0.70
0.60
| 0.50
"" 0.40
0.30
0.20
0.10
0.00
24. Phase-in Assumptions for Tier 1, NLEV, and Tier 2, for LDV, LDT1 and LDT2
X X X X X X X X X X X X X )( Xi
cMCNi
Model Year
Figure 1 - 25. Phase-in Assumptions for Tier 2, by Bin, for LDV, LDT1 and LDT2.
illiiiliiiiiiiiiiiiili
Model Year
54
-------
Figure 1 - 26. Phase-in Assumptions for Tier 1 and Tier 2 Vehicles, for LDT3 and LDT4.
1.20
1.00
0.80
o
re
0.60
0.40
0.20
0.00
AAAAAAAAAAAAAA
ModelYear
Figure 1 - 27. Phase-in Assumptions for Tier 2 Vehicles, By Bin, for LDT3 and LDT4.
1.200
1.000
O CM
ModelYear
1.6.2.3
Merge FTP results and phase-in Assumptions
For running emissions, the goal of this step is to calculate a weighted average of the FTP results
for different tiers 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 constant fractions adapted from the MOBILE model (0.15, 0.60, 0.15,
0.10 for LDT1, 2, 3, and 4, respectively). We did not vary these fractions by model year.
55
-------
Figure 1-28 shows an example of the Phase-in calculation for LDV NOx in four model years.
The tables 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. For running emissions, however, the
averages are not used directly; rather, each is expressed as a ratio to the corresponding Tier-1
value.
Figure 1 - 28. Example of Phase-in Calculation, LDV NOx, for four Model years.
Tier Standard Cold Start Running Phase-In by MY
(g)
(g/mi)
2000
2002
2005
2010
Tier 1 rieri
TLEV
NLEV LEV
ULEV
BinS
Bin 7
Bin 6
Tier 2 Bm 5
Bin 4
Bin 3
Bin 2
0.983
0.827
0.569
0.453
0.481
0.378
0.275
0.172
0.098
0.036
0.049
0.149
0.123
0.052
0.048
0.032
0.025
0.018
0.011
0.007
0.005
0.00005
Start (g)
Running (g/mile)
RATIO to Tier 1
1
0
0
0
0
0
0
0
0
0
0
0.98
0.14895
1.00
0
0.2
0.8
0
0
0
0
0
0
0
0
0.62
0.06577
0.44
0
0
0.5
0
0
0
0
0.45
0.03
0.01
0.01
0.37
0.0312
0.21
0
0
0
0
0
0
0
0.595
0.035
0.368
0.002
0.12
0.00887
0.06
Table 1-18 shows weighted average values for model-years 2001-2010 for simulated FTP
composites, start and running emissions. The Start values, expressed as the 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 1 - 28, and in Figure 1 - 29,Figure 1-30 and Figure 1-31 below.
NOTE: We are aware of the anomalous increasing trends in running CO and HC shown in
Figures 1-29 and 1-30, which are due to discontinuities in the underlying FTP results. We plan to
revisit and correct this ratios before formulation of the final set of rates.
56
-------
Table 1-18 Weighted Average FTP Values Projected for LDT and LDV for MY 2001-
2010
regClass
LDT
LDV
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)
2.17
1.81
1.76
1.72
1.34
1.06
0.77
0.47
0.46
0.46
0.45
1.604
1.173
1.170
1.167
1.004
0.857
0.693
0.430
0.420
0.418
0.402
Start
(g)
14.72
14.15
13.81
13.48
10.73
8.74
6.63
4.27
4.20
4.18
4.07
10.24
8.86
8.38
7.91
7.11
6.46
5.68
3.92
3.83
3.80
3.65
Running
(g/mi)
1.23
0.848
0.846
0.843
0.647
0.497
0.342
0.193
0.190
0.189
0.185
0.983
0.579
0.613
0.647
0.539
0.435
0.325
0.180
0.176
0.175
0.169
HC
Composite
(g/mi)
0.171
0.140
0.135
0.129
0.097
0.074
0.051
0.032
0.031
0.031
0.029
0.112
0.0764
0.0702
0.064
0.0557
0.0482
0.0404
0.0286
0.0283
0.0282
0.0277
Start
(g)
1.77
1.45
1.37
1.28
1.03
0.85
0.67
0.49
0.48
0.47
0.45
1.294
0.932
0.835
0.738
0.671
0.616
0.556
0.437
0.432
0.432
0.423
Running
(g/mi)
0.0587
0.0486
0.0497
0.0508
0.0337
0.0220
0.0102
0.0022
0.00209
0.00208
0.00197
0.027
0.016
0.017
0.018
0.014
0.0101
0.00615
0.00193
0.00191
0.00190
0.00187
NOx
Composite
(g/mi)
0.298
0.248
0.222
0.197
0.144
0.104
0.0632
0.0286
0.0244
0.0243
0.0186
0.241
0.138
0.118
0.0980
0.0786
0.0599
0.0407
0.0180
0.0177
0.0176
0.0171
Start
(g)
1.34
1.12
1.04
0.956
0.726
0.553
0.376
0.206
0.175
0.174
0.132
0.983
0.672
0.621
0.569
0.465
0.366
0.262
0.127
0.125
0.124
0.119
Running
(g/mi)
0.184
0.144
0.126
0.109
0.0780
0.0554
0.0327
0.0140
0.0120
0.0120
0.00922
0.149
0.080
0.0662
0.052
0.0415
0.0312
0.0208
0.00899
0.00886
0.00884
0.00863
57
-------
Figure 1 - 29. Weighted Ratios for Composite, Start and Running CO Emissions, for LOT and
LDV.
Model Year
Model Year
58
-------
Figure 1 - 30. Weighted Ratios for Composite, Start and Running THC Emissions, for LDT and LDV.
CM CM CM
Model Year
1.0
Mode I Year
59
-------
Figure 1 - 31. Weighted Ratios for Composite, Start and Running NOx Emissions, for LOT and
LDV
Mode I Year
Mo del Year
1.6.2.4
Estimating Emissions by Operating Mode
1.6.2.4.1
Running Emissions
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/tonne), and the ranges covered by the SFTP standards (primarily the
60
-------
US06 cycle). For convenience, we refer to these to regions as "the FTP region" and "SFTP
region," respectively (See Figure 1 - 32).
Figure 1 - 32. Operating Modes for Running Exhaust Pollutant Processes, divided into "FTP " and
"SFTP" Regions.
30 +
27-30
f 24-27
21-24
1, 18-21
I 15-18
i 12'15
'§. 9-12
6 6-9
j§ 3-6
0-3
<0
Speed Class
1-25 ^5-50
16
30
29
28
~^^^^^a
27
15 25
14 24
13 23
12 22
11 21
(mph)
50 +
40
Z3.S
&"'
38
^^^^^^^^^^^^^^^^
37
^^B
35
^
33
^^
S
The "SFTP Region"
High Power
*
The "FTP Region"
Low to Moderate
Power
(also includes braking (0)
and idle (1)
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, to represent emissions for
that year. For the FTP operating modes, we applied the ratios shown in Figure 1 - 29 to Figure 1
- 31 above.
For the SFTP operating modes, we followed a different approach. 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. However, it was not obvious that the degree of
control would increase as dramatically for the SFTP standards. Thus, in preparation of the draft
rates, we adopted a conservative assumption that emissions in the SFTP region would not drop as
dramatically as those in the FTP region.
It was therefore necessary to estimate a different set of ratios. However, it was not feasible to
calculate ratios for the SFTP modes as described above for the FTP modes because SFTP
standards do not apply to Tier 1 engines, leaving no point of reference for the ratio calculation.
So, as an initial effort for the draft, we adopted an alternative approach. Returning to the Phoenix
I/M data, 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 SFTP operating modes (as
a group), and calculated ratios between the two groups.
61
-------
77
_ ^poll.SFTP.01-03
-^FTP ~ TT 1 - 4U
-^poll,SFTP,98-00
The resulting ratios are 0.55, 0.35 and 0.30 for CO, THC and NOx, respectively. These values
exceed those for the FTP modes for MY -2004 and later. However, the calculation and
application of these values is considered preliminary and will be reevaluated for development of
the final MOVES rates.
Figure 1-33 and Figure 1-34 show application of the ratios to the FTP and SFTP operating
modes in model years 2000 (the reference year), 2005, and 2010, both calculated with respect to
2000. The ratios shown in Figure 1 - 29 to Figure 1-31 are used for the FTP modes, and the
uniform values .RSFTP are used for the SFTP modes. Note that the values for the SFTP modes are
equal in 2005 and 2010, because the SFTP ratios are constant by model year. 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.
NOTE: the assumptions described in this section as slated for review and potential revision
before releases of the final model.
62
-------
Figure 1 - 33. Projected Emission Rates by Operating Mode for Age group 0-3 years, in three
Model Years (LINEAR SCALE).
a
500 00 -
nnn
4 2000
2005
D2010
(a) CO n
» n A
n n n n n r*i n n ri n 1*1 n [*i n n n n n n "-1
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
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
120
110
100
90
80
70
60
50
40
30
20
10
0
*2000
2005
2010
(c) NOx
fTrrTrnTTTTH*Trl*lT-|\rnTnTl\rnTCH!^
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
63
-------
Figure 1 - 34. Projected Emission Rates by Operating Mode for Age group 0-3 years, in three
Model Years (LOGARITHMIC SCALE).
64
-------
Brisson Rate (g/hr)
100.00 :
10.00 ;
1.00 :
0.10 ;
n m
* 2000 S
2005 "-
D2010 * D 5
.*** ***gfl »-5
1 *flfl I *fl
(a) CO
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
1 n
0 1
n n 1 -
n nnm -
*2000 ^ ^
ZUUb ^ 4
D2010 * ^ »
**' ** 1 **
4
HHC C
flfl" B
(b) THC
s
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
1,000.00 r
100.00
10.00
1.00
0.10
0.01
»
n _.nn
n
(c) NOx
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
1.6.2.4.2 Start emissions
65
-------
As mentioned, the values for "start" shown above in Table 1 - 18 are assumed to represent cold-
start emissions, denoted by opModelD 108, and defined as start emission following a soak period
of 12 hours or longer (720 min). Additional operating modes for starts are defined in terms of
shorter soak periods, as shown in Table 1-19.
Table 1 - 19. Operating-Mode Definitions for Start Exhaust Emissions
opModelD
101
102
103
104
105
106
107
108
Description
"hot start"
"cold start"
Soak Period (min)1
<6
6-30
30-60
60-90
90-120
120-360
360 - 720
>720
1 Defined in terms of lower-bound < soak period < upper-bound.
To estimate start emissions for the other operating modes, we applied "soak fractions" to the
"cold-start" emissions. The soak fractions were adapted from the approach applied in the
MOBILE model13. Specifically, the part-wise regression equations used in MOBILE 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 33 shows the
soak fractions for HC, CO and NOx, with each value plotted at the midpoint of the respective
soak period.
Figure 1 - 35. Soak Fractions Applied to Cold-Start Emissions (opModelD = 108) to Estimate Emissions
for shorter Soak Periods (operating modes 101-107).
120 240 360 480
Soak Time (minutes)
600
720
1.6.2.5
Apply Deterioration
66
-------
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.
Note that we applied deterioration only to the running emissions, not to the start emissions. We
carry this process out in four steps.
1.6.2.5.1 Recalculate the logarithmic mean
Starting with the values of the arithmetic mean (xa~) calculated above, we calculate a logarithmic
mean (xj), as shown in Equation 1-41. Note that this equation is simply a rearrangement of
Equation 1 - 27.
x, =lnx - 1-41
' a 2
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 applied were 1.30, 0.95 and 1.6 for
CO, THC and NOx, respectively.
1.6.2.5.2 Apply a logarith mic Age slope
After estimating logarithmic means for the 0-3 age class (fyo-s), we estimate additional
logarithmic means for successive age classes (x/;age), by applying a linear slope in In-space (mi).
x
/,age
= ^,0-3+^ (age-1.5) I-42
The values of the logarithmic slope are adapted from values developed for the 1996-98 model -
year group. The values applied were 0.18, 0.15 and 0.17 for CO, THC and NOx, respectively.
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.
Figure 1-36 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
67
-------
deterioration. In this view, deterioration appears as the magnitude of the gaps between a family
of similar trends against power.
Figure 1 - 36. Example of Logarithmic Deterioration Model for THC (LDV, MYG 96-98): (a) InTHC vs.
Age, by VSP level (kW/tonne), (b) InTHC vs. VSP, by Age (yr).
12
21
-30
-10.0
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0
Age (Years)
-9.0
-10.0
0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0
VSP (kW/tome)
1.6.2.5.3
Apply the reverse transformation
68
-------
After the previous step, the values of x/;age were reverse-transformed, as in Equation 1 - 27. 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 1.6.2.5.1 and may vary with age.
Values of logarithmic variances for all three pollutants are shown in Table 17.
Table 1 - 20. 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
10-14
Pollutant
CO
2.5
2.7
2.7
2.7
2.7
THC
1.50
1.80
2.00
2.10
2.10
NOx
1.95
2.10
2.00
2.00
2.00
No values are presented in Table 1-20 for the 15-19 and 20+ year age Groups. This omission is
intentional, in that we did not want to extrapolate the deterioration trend beyond the 10-14 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.
Figure 1-37 shows the same results as Figure 1-36, 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 with increasing age.
69
-------
Figure 1 - 37. Example of Reverse Transformation for THC (LDV, MYG 96-98): (a) THC vs. Age,
by VSP level (kW/tonne), (b) THC vs. VSP, by Age (yr).
110.0
100.0
3 kW/tonne
6 kW/tonne
9 kW/tonne
12 kW/tonne
21 kW/tonne
30 kW/tonne
2.0
4.0
6.0 8.0
Age (Years)
10.0
12.0
14.0
200.0
180.0
5.0
10.0
15.0 20.0 25.0
VSP (kW/tonne)
30.0
35.0 40.0
1.6.2.6
Estimate non-I/M References
Completion of steps 1.6.2.1 - 1.6.2.6 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.
70
-------
1.7 Replication and Data-Source Identification
The rates developed as described in Sections 2 and 3 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?) In addition, we
replicated all the gasoline rates for the advanced engine technologies. The fuel types and engine
technologies represented in the table segment are listed in Error! Reference source not found..
Table 1-21 Fuel Types and Engine Technologies Represented for Criteria-pollutant
Emissions from Light-Duty Vehicles.
Attribute
Fuel type
Engine Technology
sourceBin attribute
fuelTypelD
engTechID
Value
01
02
05
01
02
11
12
20
21
22
Description
Gasoline
Diesel
Ethanol
Conventional internal combustion (CIC)
Advanced internal combustion (AIC)
Moderate hybrid - CIC
Full hybrid -CIC
Hybrid -AIC
Moderate hybrid - AIC
Full hybrid - AIC
Throughout the process, we assigned dataSourcelDs to subgroups of rates, which identify the
data and methods used for particular rates. The dataSourcelDs developed for these analyses are
listed and described in Table 1 - 22.
Table 1 - 22. Descriptions of Data Sources and Methods used in Development of Criteria-
Pollutant Emission Rates for Light-Duty Vehicles.
DataSourcelD
4400
4427
4437
4500
4527
4537
4601
Description
Data driven rates: averaged from second-by-second IM240/IM147 data from Phoenix random
evaluation sample, CY1995-99 and CY2002-05, on temperature range of 68-86 °F.
replaces a 4400 value for opModes 28-30, calculated by ratio relative to opMode 27
replaces a 4400 value for opModes 38-40, calculated by ratio relative to opMode 37
imputed using statistical hole-filling models.
replaces a 4500 value for opModes 28-30, calculated by ratio relative to opMode 27
replaces a 4500 value for opModes 38-40, calculated by ratio relative to opMode 37
calculated by ratio relative to ageGroup 10-14 (modelyeargroups 2000 and previous only,
ageGroupID 15-19 and 20+ only)
71
-------
4602
4800
4801
4802
4803
4805
4806
4807
4900
4901
4910
calculated by ratio relative to ageGroup 15-19, modelyeargroups 2000 and previous only,
ageGroupID 20+ only, meanBaseRate only (corresponding meanBaseRatelM is 4601).
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 10-14)
calculated by ratio relative to ageGroup 10-14 (modelyeargroups 2001 and later / ageGroupID
15- 19 and 20+ only).
calculated by ratio relative to ageGroup 15-19, modelyeargroups 2001 and later / ageGroupID
20+ only, meanBaseRate only (corresponding meanBaseRatelM is 4802).
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 or ethanol rates with conventional internal combustion to represent rates
for advanced engine technologies.
replicated from gasoline rates for all engine technologies to represent rates for tier-2 light-duty
diesel engines (MY 2010 and later only).
Finally, Table 1-23 shows the accounting for all rates developed for light-duty criteria-pollutant
emissions and included in the draft emissionRateByAge table. The leftmost four columns
delineate subsets of rates by the pollutant processes included (Running, Start), and the respective
fueltypes, engtechs 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, i.e., two
regClasses always refers to LDV and LDT.
The rates for datasourcelD = 4400 - 4602 were summed as a single category, as these groups
represent the outcome of a set of interrelated processes, as described in section 1.5. The count of
15 modelyeargroups includes groups through MY 2000 (see Table 1-11). The dataSourcelDs
4800 - 4803 represent running emissions for MY 2001+, as described in Section 1.6. The total
of 21 modelyeargroups represent groups 2001 - 2021-2050. For these rows, a count of one
agegroup refers to the 0-3 year ageGroup, whereas a count of four refers to the 4-5, 6-7, 8-9 and
10-14 year age Groups.
DataSourcelDs 4805 - 4807 represent start emissions for MY2001-2021. For this group, counts
of 26 or 36 modelyeargroups denote MYG 1996-2021 and 1980 and earlier - 20212050,
respectively. Counts of one or six ageGroups refer to the 0-3 ageGroup and the remaining six
72
-------
ageGroups, respectively. Counts of one or seven opModes refer to the cold-start emissions
(opmode 108) or the remaining seven start modes, respectively.
The dataSourcelDs 4900 and 4901 refer to the replication of the gasoline/conventional rates for
ethanol and the advanced engine technologies, respectively. 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. The counts for dataSourcelD 4910 is similar,
except that the 12 modelyeargroups include only 2010-20212050, as mentioned.
73
-------
Table 1 - 23. Accounting for the Segment of the Draft emissionRateByAge Table contributed by Rates for Criteria-Pollutant
Emissions, Light-Duty Vehicles.
Process(es)
Start
Running
Running
Running
Running
Running
Running
Running
Running
Running
Running
Running
Running
Start
Start
Start
fuelTypelD
01
01
01
01
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
01
01
01
01
dataSourcelD
101
4400
4427
4437
4500
4527
4537
4601
4602
4800
4801
4802
4803
4805
4806
4807
Accounting (No. classes or groups)
fueltypes
1
engTechs
1
regClasses
2
MYG
10
ageGroups
1
opModes
1
polProcesses
3
No. records
60
11 2 15 7 23 3 14,490
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
21
21
21
21
26
36
36
1
4
1
1
1
6
7
23
23
23
23
1
1
7
3
3
3
3
3
3
3
2,898
11,952
2,898
2,898
156
1,296
10,584
SUBTOTAL 46,872
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
Running & start
05
01
01
01
01
01
01
05
05
05
05
05
05
02
02
02
02
02
02
02
01
02
11
12
20
21
22
02
11
12
20
21
22
01
02
11
12
20
21
22
4900
4901
4901
4901
4901
4901
4901
4901
4901
4901
4901
4901
4901
4910
4910
4910
4910
4910
4910
4910
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
36
36
36
36
36
36
36
36
36
36
36
36
36
12
12
12
12
12
12
12
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
46,872
46,872
46,872
46,872
46,872
46,872
46,872
46,872
46,872
46,872
46,872
46,872
46,872
15,624
15,624
15,624
15,624
15,624
15,624
15,624
TOTAL 765,576
-------
75
-------
2. Particulate-Matter Emissions from Light-Duty Vehicles
2.1 Introduction and Background
There is currently a large body of research on the formation and measurement of Particulate
Matter (PM) emissions from combustion engines. This chapter describes the process by which
emissions measured in a subset of the past PM research programs from light-duty gasoline
vehicles was employed to generate emission rates for MOVES. The emission rates determined
by this approach embody strictly "bottom-up" methodology whereby emission rates are
developed from actual vehicle measurements following intensive data analysis, and are then
input into an 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 2004-200514. 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 phase, 261
vehicles were tested, while 278 vehicles were tested in the winter. The testing was conducted on
a portable dynamometer using the LA92 driving cycle in ambient temperature conditions.
Much of the data from this program was analyzed in the report: "Analysis of Parti culate Matter
Emissions from Light-Duty Gasoline Vehicles in Kansas City"15 This "analysis report" (which
is the partner to this "modeling" chapter) presented preliminary emission rates for PM, elemental
carbon fraction (EC), organic carbon fraction (OC), as well as temperature adjustment factors for
start as well as hot running emissions processes for MOVES. These emission rates form the
basis for the emission rates developed in this chapter. The rates from the analysis report were
based on the aggregate or "bag" emissions measured on the PM filters in the program, thus they
are in units of grams per start for starts and grams/mile for hot running operations. The rates
were inclusive of gasoline powered light-duty vehicles of all ages, but only modeled calendar
year 2005, when the emission rates were actually measured. The measurement program could
say very little of what the emission rates of vehicles were in, say 1990, or will be in 2020
because the vehicle fleet looks very different in these other calendar year scenarios. This chapter
describes the development of a deterioration model based on a comparison of past PM studies
with the 2005 Kansas City study. The rates from this deterioration model include calendar years
in the past, present and future as required by MOVES.
The analysis in the 2008 paper also did not describe how the emissions would change if the
driving differed from the LA92 (unified) drive cycle used in the program as it would in the real-
world. MOVES has the capability to capture hot running "modal" emission rates so that
emissions vary by the Vehicle Specific Power (VSP) of the vehicle. This chapter describes how
the real-time PM measurements collected in the study were used to populate the VSP modal rates
76
-------
for MOVES. Because of the reliance on real-time 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 2 - 1). The LA92 cycle consists of three phases or "bags".
Phase 1 ("bag 1") is a "cold start" that lasts the first 310 seconds (1.18 miles). "Cold start" is
technically defined as an engine start after the vehicle has been "soaking" in a temperature
controlled facility (typically ~72°F) with the engine off. In the Kansas City study, the vehicles
were soaked over night in 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 2-1. Phases 1 and 2 of the LA92 Cycle, "cold-start" and "hot-running," respectively.
70.0
60.0
200 400 600 800 1000 1200 1400
0.0
In addition to the regulated gas pollutants measured via the constant-volume sampler (CVS),
continuous measurements of PM mass were taken using an EPA-supplied Booker Systems
Model RPM-101 QCM manufactured by Sensors, Inc. and a Thermo-MIE Inc. DataRam 4000
Nephelometer. An estimate of black carbon was measured continuously with a DRI
photoacoustic instrument and 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
77
-------
maintained at 47°C ± 2°C. The samples were extracted from the dilution tunnel through a low
particulate loss 2.5 |j,m outpoint pre-classifier. Further details and a schematic of the sampling
instrumentation are shown in Figure 2-2 and Figure 2-3.
Figure 2-2. Schematic of the constant-volume sampling system used in the Kansas-City
Study.
y*
Dessicator
J
V
Air Conditioner
Diluted exhaust to aldehyde sample aldehyde
"f46C flow comtroller i cartridge
1U
i cartrit
to particle sample Jlf heated f vent
flow controller i sample '
T line I I
particle f l
M^-""^ ^^ ^ background
. \v sample line
V \
dilution air \r
heater (46 C) \ 1
^
Backgrd HC analyzer
vent
water p-
trap -^*" H
pump -»«~l
filter »" |
flow ^-*' *
measurement
and control
n
, 1
1
1 '
. 1
f"
Positive
Displacement
Pump (POP)
( heated
sample Heated
HC analyzer
LJLJLIJ
measurement
and control
4 * * *
high CO r NOx analyzer
analyzer I
low CO
analyzer
Dilute exhaust
collection bags
CO2 analyzer
Figure 2-3. Continuous PM analyzers and their locations in the sample line.
<-Dyno
" -PM 2.SIM FACTORS
PHOTO ACOUSTIC
(black carbon)
DutiTruk
(light
scattering)
78
-------
It is worth briefly describing the real-time PM measurement apparatus used in the study. A more
thorough description maybe found in the contractor's report14. As of the date of this
measurement program, there existed no perfect means of measuring real-time PM. Each of the
devices has specific advantages and disadvantages. For this study, it has been assumed that the
cumulative mass as measured (weighed) on the Teflon filters is the benchmark. Thus all real-
time measurements have been normalized to the filters in order to minimize systematic
instrument errors.
The Quartz Crystal Microbalance measures the cumulative mass of the PM deposited on a crystal
face by measuring the change in 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
diluting the temporal effects of transients. The QCM can accurately capture cumulative PM that
collects over time, however the measurement uncertainties increase at any given moment in time
because they are dependent on a calculation 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, QCM measurements were not
viewed as practical for use with MOVES at this time, except as a check for 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 real-time mass
concentrations though it may miss a substantial portion of the smaller (nano) particles. To
provide a qualitative check on supposition, the time-series for the QCM and optical instruments
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 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 real-time 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. It has been validated in a number of
studies [references]. 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 report14. 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
79
-------
means of determining modal emission rates, and the use of such data do not qualify them as EPA
recommended or approved devices or processes.
2.1.2 Causes of Gasoline PM Emissions
Particulate matter is formed from gasoline-fueled engines originating from 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 also
compare 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 form particulate. This PM is generally ccomposed 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 as compared to diesels but does form in larger
quantities under relatively rich 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 number is about 20% of PM mass
compared to about 70% for a diesel engine. There are also other compounds in the fuel or
engine oil such as trace levels of sulfur and phosphorus which, in combustion, form sulfates and
phosphates, both of which form particulate. The sulfur level in gasoline is now very low, almost
eliminating sulfate formation from gasoline sulfur content but motor oil contains significant
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.
80
-------
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 supplanted it; 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 controls, however it should be noted that the correlation between these
emissions is far from perfect. Many examples of this are shown in the 2008 analysis report.
Amounts of of PM emitted are very sensitive to the amount of fuel in combustion as well as the
airfuel ratio. 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). When high acceleration rates or
loads are encountered (such as in 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 also get into the combustion chamber via several
pathways. Some manufacturers 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, HC from oil is also small. However,
organic PM from oil consumption can be quite significant because oil is a high molecular weight
81
-------
hydrocarbon, and more likely to remain as uncombusted 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 recently, the first two are less of a factor in the Kansas City program than aromatics
would be. In the calendar years that MOVES models, lead is not a significant portion of the
inventory, thus is largely ignored. Sulfur (as a fuel rather than a tailpipe phenomenon) is
modeled separately and is described in another MOVES document. Impurities may be captured
in a future version of MOVES.
Some of these PM forming mechanisms clearly affect HC emissions. So 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 a modeling approach to extend the PM emission trends from Kansas
City presented in the analysis report to average emissions across the fleet. The section also
compares the new vehicle results from many different studies in order to determine the zero mile
level (ZML) emission rates for all model years in MOVES. Before modeling deterioration, it is
first necessary to capture ZML emission rates.
In constructing a model of emissions from the Kansas City data (Error! Reference source not
found.), the most significant challenge is distinguishing between model year and age effects.
This problem arises because program was conducted over a two-year period, thus ensuring a 1 to
1 correspondence between model year and age. As a result, one cannot know for sure whether
emissions are decreasing with model year, or increasing with age, or both. 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.
82
-------
Figure 2-4. Average PM Emissions from the Kansas City by Model Year.
E
^3>
E
n
n
n
n
n
n
<
I
*
1975 1980
KC measured
T T
I'nhhl
f S
1985 1990 1995
Mode I Year
« ₯ S .,
2000
In concept, the most accurate means of quantifying emissions from vehicles over time is to
conduct a longitudinal study, where emissions are measured for the same vehicles over several
(or many) years. However, implementing such a study would be costly. Moreover, it is
impossible to obtain recent model year vehicles that have been significantly aged. In the
following sections, we will describe some limited longitudinal studies conducted in the past.
Then we will present our modeling methodology to isolate model year (technology) in this
chapter from age (deterioration) in the next.
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 from 0 to 55,000 miles (odometer) on 3
different cycles for the EPA16. 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.
83
-------
Figure 2-5. Hydrocarbon emission as a function of mileage (Gibbs et al., 1979)
0 C
0
9 R
E 9
2
7T" -I K
O 1 -3
1
n *i
*
^ »
******
0 10 20 30 40 50 60
mileage (*1000)
Hammerle et al. (1992) measured PM from two Ford model vehicles over 100,000 miles.17
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 only 2 vehicle models.
Figure 2-6. Particulate emissions as a function of odometer for two Ford vehicles (Hammerle et
al., 1992)
y=2E-05x +0.8218
R7 = 0.555
0.5
20000
40000
60000
odometer
80000
100000
120000
84
-------
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) is not
capable of accounting for all causes of deterioration.
Whitney (2000) re-recruited 5 vehicles that had been tested from a previous large study 2 years
prior (CRC-E24)18. 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
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 PM studies. This entails capturing the near ZML or Zero Mile Level
emission rates. New vehicle emission rates tend to have a much smaller variability than older
vehicles (in absolute terms) since they have lower emissions that comply with more stringent
standards. These standards, which decrease over time, are for hydrocarbons, but tend to have an
effect on PM emissions as well since many of the same mechanisms for HC formation also form
PM.
Several independent studies have been conducted , which 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. The following table lists the 15 studies employed for
this analysis.
Historical gasoline PM studies including new vehicles
Table 2-1. Historical gasoline PM studies, including new vehicles.
Program
Gibbsetal.19
Cadle et al.20
Uiban&Garbe21,22
Langetal.23
Volkswagen24
Yr of Study
1979
1979
1979, 1980
1981
1991
#new
vehicles
27
3
8
8
7
Drive
cycle
FTP
FTP
FTP
FTP
FTP
85
-------
CARB25
Ford26
CRCE24-1 (Denver)27
CRC E24-3 (San
Antonio)28
CRC E24-2 (Riverside)29
Ford,Chrysler,GM
Chase etal.30
Whitney (SwRI)31
KC (summer)32
EPA (MSAT)33
1986
1992
1996
1996
1997
2000
1999
2004
2006
5
2
11
12
20
19
2
13
4
FTP
FTP
FTP
FTP
FTP
FTP
LA92
LA92
FTP
Before, we plot 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,
Riverside34. 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
this program. However the Figure 2-7 shows the HC emissions compared for the two cycles.
The trends indicate that there is little cycle effect for HC.
86
-------
Figure 2-7. Hydrocarbon emissions on the LA92 versus corresponding results on the FTP.
18
16
14
CM
O>
6
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 2-8 shows the new-vehicle emission rates from the 11 studies listed in Table 2-1 The
data points represent each individual test, and the points with error bars represent the average for
each test program. 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 2
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). 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 parti culate
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 there is insufficient documentation to even
attempt it. Therefore no attempts at adjustment or normalization were made except for size
fraction, lead and sulfur, as described above.
87
-------
Figure 2-8. Particulate emission rates for new vehicles compiled from 11 independent studies.
30
25
E
?
o
i
Q.
15
10
A
| '
Vw
^ .
_
IQ
D 5
. i
A
^>^
x x >
10
I
I PM = 15.6exp(-.0814 l\
i^ _
"""T .
15
all
Gibbsetal., 1979
o Cadleetal., 1979
n Urban&Garbe, 1980
A Lang etal., 1981
X VW, 1991
CARB, 1986
Ford, 1992
n E24-1 Denver
A E24-2 UC Riverside
Chrysler/Ford/GM
0 SwRi/NREL
x KC-Summer LA92
+ MSAT-Tier2
mean for each program
Fit
A
/IY) n
0 A
0 g X
RR II*"
20 25 30
Model Year (+1975)
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 bag. 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 LDV (^HR,^) is:
F - F
^
^-0.814^
2-1
Where
y = model year - 1975, and
= hot running zml rate for MY 2005.
To estimate equivalent rates for trucks, we multiplied this expression by 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 g/mi).
These values were estimated by running a general linear model of bag 2 and bag 1-3 with respect
to composite PM respectively in the SPSS statistical software tool. The averages of these ratios
by model year are shown in Figure 2 - 9, in which no clear trend is discernable. The parameters
of the model are summarized in Table 2-2.
88
-------
Figure 2-9. Ratios of hot-running/composite and cold-start/composite, Bag2 and Bagl-Bag3,
respectively, averaged by model year.
7
R
*
(/) .-
Q.
1 4
o *f
o
*? 3
O)
° 9
-1
n
coldstart/comp
bag2/comp
1960
1
1970
***
^. A* .a. * *
*v» *: *
» ** * *
: * *t* ,» * .
B t"|"Ml »! .!
* * **
1980 1990 2000 2010
model year
Table 2-2. 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 2-10 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.
89
-------
Figure 2 -10. Particulate ZML emission rates (g/mi) for cold-start and hot-running emissions, for
LDV and LOT.
0.050
E
0)
0.000
0.045 -
0.040 -
0.035 -
0.030 -
0.025 -
«
0.020 -
0.015 -
0.010 -
0.005 -
truck new bag1_3
car new bag1_3
-truck new bag2
-car new bag2
1975
1980
1985
1990
model year
1995
2000
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 must be done 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 (even if there are a family of 2 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 bins 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.
90
-------
The bag measurements from all of the vehicles measured in Kansas City were first adjusted for
temperature using the equation derived in the analysis paper. The equation used is:
F _p -0.03344(72-7-)
^PM,72 ^PM,rc z " z
where £pM,72, is the adjusted rate at 72°F for cold-start or hot-running emissions, EfM,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.
There are two simple methods for determing the deterioration rates. The first method is to
subtract the aged rates from the new rates, the difference by age would define the deterioration
rate. This model will be referred to as the "additive deterioration model". The second method is
to ratio the aged rates with the new rates so that the deterioration rates are all proportional. This
will be referred to as the "multiplicative deterioration model".
In the additive mode, the Kansas City data were grouped or averaged within the age Groupss,
with the prerequisite that no newer age Group can have greater rates than an older age Group.
This constraint is maintained by averaging 2 (or more) age bins together as necessary. Figure 2 -
11 shows the additive age deterioration rates in units of g/mi for running and g/start for starts.
Figure 2 -12.shows the aggregate deterioration rates compared to the data. The cold start and hot
running were combined with the proper weighting to calculate an LA92-equivalent rate. Finally,
Error! Reference source not found, shows the LA92 equivalent emission rates, combining
ZMLs and deterioration, compared to the data.
91
-------
Figure 2 -11. The incremental effect of age on Particulate emissions based on the Kansas-City
Results.
0.16
0.14
s 0-1
ce
c
o
I 0.08
LU
c
.0
2 0.06
0)
Q
0.04
0.02
Dear hot running g/mi
Dtruck hot running g/mi
car start g/mi
D truck start g/mi
0-3
20+
Figure 2 - 12. Cold-start and hot-running Particulate emissions by age, with estimated rates
compared to data (as LA92 composite, mg/mi).
"Si
E
1 50
I
D
< 40
i^S
5
j
T
7\_ ''
10
' ? '
/
'
1
'
5
I
Age
[
-
1 ,
r
//
/
/
:
2
0
2
I
5
(
[
30
0
cars
trucks
car age model
truck age model
35
92
-------
Figure 2 -13. Emission rates by age, as LA92 composites (mg/mi), compared to data.
100
10
10
15
20
25
30
Age
The model appears to fit the data well. One can also observe from Error! Reference source not
found, the relative effect of ZML vs deterioration. The slight curvature after each large step
represents the ZML exponential curves, whereas the larger steps represent the deterioration.
This additive deterioration model suffers from some significant limitations. The first is that it
assumes that all model years will deteriorate the same way independent of the emission control
technologies employed. As an example, the model assumes that a 1975 carbureted engine will
deteriorate in the same fashion as a 2005 fuel injected engine car; i.e. they will have different
ZML rates to start, but both will add about 50 g/mi after 20 years. So, while new 1975 and 2005
cars have emission rates of approximately 15 and 2 g/mi, respectively, (a factor of 7.5), after 20
years the rates would be 65 and 52 g/mi, respectively, thus the cars will have nearly the same
emission rates. While it is possible that a small number of fuel-injected vehicles would
experience complete failures of their aftertreatment and fuel control systems, it is unlikely that
they would (on average) have similar emission rates as a carbureted vehicle. The second
problem with the model is that it is not consistent with the way hydrocarbon deterioration rates
are represented in MOVES.
It is likely that some of the same mechanisms that cause HC 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. Error! Reference source
not found, 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. This pattern means that the
deterioration rate is exponential over this time interval. This observation, combined with the
93
-------
approximate parallelism of the trends for successive model years, imply that emissions follow a
multiplicative pattern across model-year or technology groups, thus calling for a multiplicative
deterioration model. In such a model, the aged rates and the new rates are converted to a
logarithmic scale, then the slopes are estimated by fitting a general linear model. The average
slope is determined, and the ZMLs determined earlier define the y-axis offsets. This results in a
series of ladder-like linear lines in on a log scale as show in Error! Reference source not
found.. The lines fan out exponentially on a linear scale as shown in Error! Reference source
not found.. 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.
Figure 2 -14. The natural log of THC emissions vs. Age for LDV in the Phoenix (AZ) Inspection
and Maintenance program over a ten-year period (1995-2005).
LDV WEIGHTED
In (THC) vs. Age (years), LDV
Vehicle age (yeans)
Model Year
OOO 1981
B-B-B 1337
000 1962
ODD 1338
&-&-& 1994
2000
1333
B-B-B 1339
&-&-& 1995
I I I 2001
OOO 1564
B-B-B 1990
A A A 1936
I I I 2002
OOO 1335
B-B-B 1991
r 1997
I I I 2003
94
-------
Figure 2 -15. 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.
fe
=
Age
Figure 2 - 16. 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 represente the
Kansas-City results (with 95% confidence intervals).
100 --
--1975
^»^1980
--1985
-1990
--1995
-2000
^2005
D KG measured in 2004
- - KG model in 2004
1985
[ 11980
40
20
1975[T]
i]2005
10
15
Age
20
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 fill the remainder of the rates. Because no data is
95
-------
available describing how particulate start emissions vary by soak time, we have used the HC
soak curves shown previously (see Figure 1 - 35).
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 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
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 and higher emitters. We conclude by describing the
procedure by which results from the Dustrak were used to develop emission rates by operating
mode.
2.2.1 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 release 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 2n (hot
running) bag at around 850 seconds. This peak is captured in Figure 2-17, which includes 2
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.
96
-------
Figure 2-17. A typical time-series plot of continous particulate emissions as measured by several
instruments.
x ,(,-' Test [84714] Model [STATION WAGON] MY [1994] Bag1 PM [82.09 mg/m] Bag2 PM [6.69 mg/m] Temp [51.5 f]
E 3 -
1 -
,1
in
1 1 f 1
J U\M fc Mrt^Jh Hi f LJ - * 1 . /
200 400 600 BOO
1 1 i
JL*. «L . 1 i .1. J
1000 1200 1400
QCM (raw)
Time, seconds
' DustTrak (shifted, normalized) DataRAM (shifted, normalized)"
PA (g, shifted)
x10
400
PA (g, shifted)
1000
600 800
Time, seconds
Speed (mph) HC (g, normalized to bag 2 PA)
1200
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 2-18 shows an older 1976
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.
97
-------
Figure 2 - 18. Continous particulate emissions from a 1976 Nova measured at 54°F.
1()-' Test [B4712] Model [NOVA] MY [1976] Bag1 PM [354.43 mg/m] Bag2 PM [13.71 mg/m] Temp [54.3 F]
D_
0.5
yn.
200
QCM (raw)
400
1000
600 800
Time, seconds
' DustTrak (shifted, normalized) DataRAM (shifted, normalized)"
1200 1400
PA (g, shifted)
-
200
400
- PA (g. shifted)
1000
600 BOO
Time, seconds
~~ Speed (mph) HC (g, normalized to bag 2 PA)
1200
1400
Figure 2-19 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
Figure 2 - 19. Particulate time-series for a 2002 Trailblazer at 50°F.
these vehicles need to go into enrichment during this relatively mild acceleration.
98
-------
x 10'* Test [B4628] Model [TRAIL BLAZER] MY [2002] Bag1 PM [64.21 mg/m] Bag2 PM [12.17 mg/m] Temp [50 F]
=
2.3
200
QCM (raw)
400 600 800 1000
Time, seconds
' DustTrak (shifted, normalized) DataRAM (shifted, normalized)"
1200 1400
PA (g, shifted)
x10
100
E
50 -c
LU
400
1000
600 800
Time, seconds
' PA (g, shifted) ~ Speed (mph) ~ HC (g, normalized to bag 2 PA)
1200
1400
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 2-20 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 VSP methodology. 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 2 - 20. Particulate time-series for a 1988 Dynasty.
99
-------
Test [84118] Model [DYNASTY] MY [1988] Bag1 PM [52 mg/m] Bag2 PM [47.61 mg/m] Temp [87.5 F]
x10
0.01
0.008
0.006
0.004
0.002
0
I
-
it§
200 40
A
0
tTrak (si
I I
W^ji/Wuu^ll li .
I I I
A. J A. . ».. i
600 800 1000 1200 1400
Time, seconds
lifted, normalized) DataRAM (shifted, normalized) PA (g, shifted)
400
- PA (g, shifted)
1000
600 800
Time, seconds
Speed (mph) HC (g, normalized to bag 2 PA)
1200
1400
Figure 2-21 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 dustrack, nepholometer, and the QCM all register the hump, while the PA shows very
little EC, 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 emitter or vice-versa in back-to-back tests. These
observations have ramifications for future PM test programs, in that sample sizes should be large
and fleets properly representative.
100
-------
Figure 2 - 21. Continuous particulate time series for a 1995 Lincoln Continental.
x 10 * Test [84380] Model [CONTINENTAL] MY [1995] Bag1 PM [12.24 mg/m] Bag2 PM [41.35 mg/m] Temp [62.1 F]
200
QCM (raw)
600 800 1000
Time, seconds
"DustTrak (shifted, normalized) DataRAM (shifted, normalized)"
1200 1400
PA (g, shifted)
x10
lA
i It j lit in 111 in1* inn
_,-U..-: W4alA.v^!:i
200
400
1000
600 800
Time, seconds
PA (g, shifted) Speed (mph) ~~ HC (g, normalized to bag 2 PA)
1200
1400
The next figure (Figure 2 - 22) shows a more typical high PM emitter, where the bag 2 emission
rate is 266g/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.
101
-------
Figure 2 - 22. Continuous particulate (and HC) time series for a 1978 MG.
Test [84277] Model [MG] MY [1978] Bag1 PM [251.11 mg/m] Bag2 PU [266.15 mg/m] Temp [69.7 F]
200
QCM (raw)
400
1000
600 800
Time, seconds
' DustTrak (shifted, normalized) DataRAM (shifted, normalized)"
1200 1400
PA (g, shifted)
xlO
200
400
1000
600 800
Time, seconds
- PA (g, shifted) ~ Speed (mph) HC (g, normalized to bag 2 PA)
1200
1400
We are now ready to bin the emission rates into operating modes based on vehicle-specific
power (VSP) The following two figures show Dustrak PM emissions binned by VSP and
classified by model year Groups. Error! Reference source not found, shows this relationship
on a linear scale and Error! Reference source not found, 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 linear on a log scale. [This is consistent with some of the other
criteria pollutant trends - is this true Jim?}. Thus we assume smooth log-linear relations when
calibrating our VSP based emission rates.
102
-------
Figure 2 - 23. Particulate emissions, as measured by the Dustrak, averaged by VSP and model-year
Group (LINEAR SCALE).
Cars
x10
10 15 20 25 30 35 40 45 50
VSP, kw/tonne
Figure 2 - 24. Particulate emissions, as measured by the Dustrak, averaged by VSP and model-year
Group (LOGARITHMIC SCALE).
Cars
-51 1 1 1 1 1 1
-6
-7
ca
E .«
(0 -8
n -9
"3
-11
-12
-13
1983
-1989
1996
-2000
2001
2004
0 5 10 15 20 25 30 35 40 45 50
VSP. kw/tonne
In order to determine the actual MOVES VSP based rates, followed seven steps:
103
-------
1. The LA92 equivalent hot running emission rate in 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 activity 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
VSP bin.
4. The VSP bin rates are then combined with the VSP activity rates by operating mode and
then summed to give a total bag 2 emission factor that must match the aggregate LA92
emission rates in step 2 (as calculated from the filter measurements).
5. The emission rates are constrained to match the filter values through a normalization
factor that is applied to every model year age group (test by test??).
6. The rates from step 5 are then multiplied by the corresponding EC and OC factors from
the analysis report(cite) to give all of the hot running rates.
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
2-25. For operating-mode definitions, see Table 1-2.
Figure 2 - 25. Operating-Mode distribution for cars and light trucks representing the hot-running
phase (Bag 2) of the LA92 cycle.
160
140
120
.£ 100 1
T3
C
O
ou 80
o
1 60
40
20
Dear
D light truck
m
Jli
11
n
1 11 12 13 14 15 16 21 22 23 24 25 27 28 29 30 33 35 37 38 39 40
VSP Bin
104
-------
The output of step 5 for each model year ZML (0-3 year age Group) is shown in Error!
Reference source not found..
Figure 2 - 26. Particulate emissions for passenger cars (LDV) from Kansas City results, by model
year Group, normalized to filter mass measurements.
1fi
14 -
19
3T
en m
0)
-t-J
CO
o: o _
c b
O
'(/)
- 6 -
E b
LU
,1
o
n -
A A
.
x
* - x
* * -
-jjj_-_: -_BJLU llU S ! sSil
1960-1980
1981-1982
1983-1984
X1985
1986-1987
+ 1988-1989
-1990
1991-1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
I
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 that has exponential factors risks
having excessively high emission rates under extreme conditions. For example any rate over 100
g/s 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.
2.4 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
105
-------
effect of age is determined by comparing the new vehicle rates to the Kansas City data. It is
determined that emissions deteriorate exponentially with the age of the vehicle, but remain
constant after about 20 years. It is also found that PM emission increase exponentially with VSP
(or road or engine load).
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 emission and also quantifying the PM emissions due to oil consumption. This 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 the PM. It is important to continue to collect PM
emissions data in the field, since it validates the deterioration model.
106
-------
3. Criteria Pollutant Emissions from Light-Duty Diesel Vehicles
(THC, CO, NOx)
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 the emission
rates by VSP. Therefore, we used results (in grams per mile) from the Federal Test Procedure
(FTP ) to estimate corresponding modal rates (in grams per hour).
3.1. 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 tested either to certify or to generate fuel
economy estimates (labels or CAFE). These test vehicles were all new (age = zero years), each
vehicle having accumulated about 4,000 miles. These individual tests 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 USA.) The sample sizes (by model year group)
and the mean composite FTP emissions are given in Table 3-1 for LDV and Table 3 - 2 for
LDT:
Table 3-1. Mean Composite FTP Emissions (g/mile) for diesel-fueled LDV.
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
2011+2
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
0.0196
CO
1.3425
1.0861
0.9809
1.1386
1.3581
1.6854
1.2233
0.4140
0.5367
0.5367
NOx
1.4126
1.1859
1.0517
0.8436
0.5952
0.5685
0.8567
0.8180
0.3925
0.0500
PM1
*
0.2999
0.2881
0.2751
0.5668
0.4990
0.1747
0.0848
0.0312
0.0100
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.
2 For 2011 and newer model years, we used the smaller of the Tier-2 Bin-5
standard or actual test results from the 2006-2008 LDV.
107
-------
Table 3 - 2. Mean Composite FTP Emissions (g/mile) for diesel-fueled light-duty trucks
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
2011+**
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
0.0181
CO
1.7923
1.3277
1.0302
1.1200
0.9985
1.7000
1.6256
1.6179
0.2767
0.2767
NOx
1.6577
1.3748
1.3052
0.9473
1.4435
1.2600
1.3814
1.4629
0.4583
0.0500
PM
*
0.3296
0.2700
0.2673
0.2790
0.1280
0.1114
0.0960
*
0.0100
Because measurements of PM emissions were not performed for the Pre-
1981 model year cars (ortrucks), 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).
. For 201 1 and newer model years, we used the smaller of the Tier-2 Bin-5
standard or actual test results from the 2006-2008 LOT.
To achieve the substantially lower FTP PM emissions, manufacturers are now equipping
their diesel-fueled vehicles (cars and trucks) with particulate traps.
All of the Tier-2 diesel test vehicles in this sample were certified to the Bin-10 standards,
implying that all 2006-2008 light-duty Tier-2 diesels sold in the USA were certified to the same
standards. Since manufacturers are likely to target the Bin-5 standards in future model years,
those average FTP emissions are probably not appropriate to represent model years beyond 2010.
Therefore, we used the mean Bin 10 emissions of those Tier-2 vehicles to estimate the typical
fleet emissions for only model years 2006 through 2010. For model years 2011 and later, we
used the Bin-5 standards for NOx and PM. However, since the actual HC and CO (average)
emissions (of those 2006 to 2008 test vehicles) were lower than the Bin-5 standards, EPA
proposes using those lower test results for HC and CO (assuming that moving from Bin-10 to
Bin-5 will not lead to an increase in HC or CO).
3.1.2 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.
108
-------
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 (£Hc,Bagi, g) to the
corresponding composite FTP HC emission rate (£Hc,composite,g/niile):
^HC.Bagl
= -0.6433 + 4.702885 E,
HC, composite
3-1
Graphing this equation along with the 146 FTP test results, as shown in Figure 3-1 below,
illustrates the relationship between the individual bag HC emission and the composite HC
emission.
Figure 3-1. Example: Bag-1 HC (g) versus Composite FTP HC (g/mile)
16
12
0.0
0.5
1.0 1.5 2.0
FTP HC Rate (grams / mile)
2.5
3.0
We then applied those 14 equations (derived from the regression analyses) to the
corresponding composite FTP emissions from Tables 1 and 2. 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
109
-------
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 3 - 3 to
Table 3 - 6), one table for each of the four pollutants.
Table 3-3. Estimated Aggregate HC Emission Rates.
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
2011+
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
0.1477
Cold-Start
(g)rt
1.0961
0.5505
0.4325
0.4729
0.4486
0.4600
0.4953
0.2816
0.0071
0.0071
Hot-Start
(g)
0.1688
0.1626
0.1349
0.1444
0.1387
0.1414
0.1496
0.0995
0.0351
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
0.1226
Cold-Start
(g)rt
1.6077
0.7784
0.5668
0.5009
0.5165
0.6707
0.4811
0.3196
0.0036
0.0036
Hot-Start
(g)
0.3280
0.2161
0.1664
0.1510
0.1546
0.1908
0.1463
0.1084
0.0342
0.0342
Table 3-4. Estimated Aggregate CO Emission Rates.
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
2011+
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
8.0052
Cold-Start
(g)rt
3.0900
2.5146
2.2787
2.6326
3.1250
3.8594
2.8226
1.0066
1.2818
1.2818
Hot-Start
(g)
1.0957
0.8647
0.7700
0.9121
1.1098
1.4046
0.9884
0.2592
0.3698
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
3.6954
Cold-Start
(g)rt
4.0993
3.0567
2.3892
2.5908
2.3181
3.8922
3.7252
3.7079
0.6984
0.6984
Hot-Start
(g)
1.5010
1.0824
0.8144
0.8953
0.7858
1.4178
1.3508
1.3438
0.1355
0.1355
110
-------
Table 3-5. Estimated Aggregate NOx Emission Rates.
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
2011+
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
0.9813
Cold-Start
(g)rt
1.6481
1.4573
1.3444
1.1692
0.9602
0.9377
1.1802
1.1477
0.8433
0.3793
Hot-Start
(g)
1.5561
1.3466
1.2227
1.0304
0.8009
0.7762
1.0425
1.0067
0.6673
0.1751
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
0.9813
Cold-Start
(g)rt
1.8543
1.6162
1.5576
1.2565
1.6740
1.5196
1.6218
1.6903
0.9325
0.3793
Hot-Start
(g)
1.7824
1.5211
1.4568
1.1262
1.5846
1.4151
1.5272
1.6025
0.7619
0.1751
Table 3-6. Estimated Aggregate PM Emission Rates.
Model Year
Group
Pre-1981
1981-82
1983-84
1985
1986-90
1991-93
1994
1995-2005
2006-2008
2011+
Diesel-Fueled Passenger Cars
Running
(g/hr)
7.0131
3.3778
3.2356
3.0774
6.6108
5.7897
2.4073
1.1338
0.3738
0.0739
Cold-Start
(g)rt
2.4362
1.2427
1.1960
1.1441
2.3041
2.0346
0.6682
0.3368
0.1390
0.0609
Hot-Start
(g)
1.2789
0.6436
0.6188
0.5911
1.2086
1.0651
0.3020
0.1378
0.0397
0.0010
Diesel-Fueled Light-Trucks
Running
(g/hr)
7.0131
3.7378
3.0160
2.9830
3.1250
1.7460
1.5101
1.2931
0.3738
0.0739
Cold-Start
(g)rt
2.4362
1.3609
1.1239
1.1131
1.1597
0.4961
0.4347
0.3782
0.1390
0.0609
Hot-Start
(g)
1.2789
0.7065
0.5804
0.5746
0.5995
0.2167
0.1863
0.1583
0.0397
0.0010
The PM rates in the preceding table represent the PM10 rates for all particulate 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 another
report.
NOTE: 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 1-23 (dataSourcelD 4910)
3.1.2 Assigning Operating Modes for Starts (Adjustment for Soak Time)
MOVES has start emission rates for eight different 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).
Ill
-------
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 EopModero) by exponential functions of the form:
3-2
where £108 = 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
(£101) from the preceding four tables, we solved algebraically for both the a and ft coefficients,
specifically:
a - e720/?+ln0.001
F \
In i_±M. -inO.OOl 3-3
n_
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 plot 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.
112
-------
Figure 3 - 2. 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
emissions of gasoline-fueled vehicles and that EPA used in MOBILE6
35,36
3.2 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 1 - 2). However, we lacked
the requisite second-by-second data for the diesel-fueled passenger 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 3-7.
113
-------
Table 3 - 7. Operating-Mode Distribution for the LA-4 Drive Cycle
ooModelD
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
o
J
0
0
LOT
164
255
96
139
103
66
33
20
70
164
110
33
19
15
7
2
0
29
33
11
3
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 LDD (Table 3-3 through Table 3 - 6), 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 an
adjustment 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 plot of the estimated zero-mile HC emission
rates by VSP bins for 1995-1998 model year diesel-fueled passenger cars.
114
-------
Figure 3 - 3. Modal Emission Rates for HC (g/hour) for 1995-98 diesel-fueled LDV.
HC Emissions (g/ hour) by VSP Bins
1995-1998 Diesel-Fueled Passenger Cars
10 20 30
VSP Bin Operating Modes
40
115
-------
4. 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." This blow-by is potentially a
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 gasoline vehicles since model year 1969.
Turbocharged diesel engines have only required PCV valves since model year 2008. MOVES
emission rates 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 percentage of tailpipe emissions. These emissions are
calculated selected pollutant processes, including THC, CO, and NOX; and the paniculate
fractions Organic Carbon PM 2.5, Elemental Carbon PM 2.5, Sulfate PM 2.5, Sulfate PM 10,.
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.
With a working PCV valve, emissions are considered zero. Based on EPA tampering
surveys, MOVES assumes a 4% PCV valve failure rate.37 Consequently, the emission rates in
fuel type/model year combinations that have PCV valves are estimated as 4% of the emission
rates of those years with uncontrolled emissions.
Very little information is available on crankcase emissions, especially those for gasoline
vehicles. A literature review was conducted in order to determine the best data sources for
emission factors (Table 4 - 1).
116
-------
Table 4-1. Selected Results for Crankcase Emissions Data
Authors
Hare and Baines38
Heinen and Bennett39
Bowditch40
Montalvo and Hare41
Williamson42
Kittelson43
Hill44
Ireson45
Zielinka46
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
117
-------
Based on this literature review, emission factors were estimated for years without
mandated PCV valves (Table 4 - 2). In absence of better information, gasoline emission factors
are largely a reflection of diesel research. As noted previously, model years with PCV valves
were assigned emission factors which were 4% of the emission factors belows
Table 4-2. Emission Rates for Vehicles without PCV systems (percent of exhaust
emissions)
Emission Type
HC
NOX
CO
PM (all speciations)
Gasoline
33%
0.03%
0.005%
20%
Diesel
2%
0.03%
0.005%
20%
The crankcase emission factors for HC, CO and NOxmay underestimate emissions.
These percentages of exhaust emissions are generally based on 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 significantly
higher than current tailpipe exhaust.
A 1995 study by Williamson 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 factors in MOVES
maintain consistency with those emission factors used in the NONROAD model.
Emission factors 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.
-------
References
1 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.
2 Heirigs, Philip, Robert Dulla, Robert W. Crawford. Processing 0/IM240 Data for Use in MOVES. SR007-05-02.
Sierra Research, Inc., Sacramento, CA. May, 2007.
3 Singer, B.C., R.A. Harley, D. Littlejohn, Jerry Ho and Thu Vo. 1998. Scaling of infrared remote sensor
hydrocarbon measurements for motor vehicle emission inventory calculations. Environ. Sci. Technol. 32:21 3241-
3248.
4 Bishop, G.A. and D.H. Stedman. 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.
Kish, L. 1965. Survey Sampling. John Wiley & Sons, New York.
Neter, J.; Kutner, M.H.; Nachtsheim, C.J.; Wasserman, W. Applied Linear Statistical Models. Irwin, Chicago.
Fourth Edition. 1996.
McClement, D., Dulla, R.G. Identification ofNon-I/M Vehicles in I/MProgram Vehicle Emission Datasets. Draft
Report, Sierra Research, Sacramento, CA., October, 2007.
8 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.
9 DeHart-Davis, L., Corley, Elizabeth and Micheal O. Rodgers. 2002. Evaluating vehicle inspection/maintenance
programs using on-road emissions data. Evaluation Review. 26(2) 111-146.
10 Bishop, G.A., and D. H. Stedman. 2008. A decade of on-road emissions measurements. Environ. Sci. Technol.
Xxs, sss-sss.
11 M6. EXH.??? Report for NOx
12 M6.EXH.012. Report for HC, CO
13 M6.STE.003.
14 Kishan, S.; Burnett, A.; Fincher, S.; Sabisch, M.; Crews, B.; Snow, R.; Zmud, M.; Santos, R.; Bricka, S.; Fujita,
E.; Campbell, D.; Arnott, P. (2007), Kansas City PM Characterization Study - Final Report. EPA Contract Report
(ERG No. 0133.18.007.001), October, 2006.
15 Nam et al., 2008.
16 Gibbs, R.E., Wotzak, G.P., Byer, S.M., Kolak, N.P., Sulfates and? articulate Emissions from In-Use Catalyst
Vehicles. Regulated/Unregulated Emissions and Fuel Economy. Report: EPA-600/9-79-047, 1979.
17 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 Mehthylcyclopentadienyl
Manganese Tricarbonyl. SAE920731, 1992.
119
-------
18 Whitney, K.A., Collection ofln-Use Mobile Source Emission Samples for Toxicity Testing, Southwest Research
Institute. 08.02602, October, 2000..
19 Gibbsetal., 1979
20 Cadleetal., 1979
21 Urban, C.M., Garbe, R.J., Regulated and Unregulated Exhaust Emissions from
Malfunctioning Automobiles. SAE790696, 1979.
22 Urban, C.M., Garbe, R.J., Exhaust Emissions from Malfunctioning Three-Way Catalyst
Equipped Automobiles. SAE800511, 1980.
23 Lang etal., 1981
24 Volkswagen, 1991
25 California Air Resources Board. Paniculate Exhaust Emissions: Gasoline Powered
Vehicles. Emissions Studies Section, El Monte, CA. Project No. 2R8618, 1986.
26 Ford, 1992
27 CRCE-24-1 (Denver)
28 CRC E-24-3 (San Antonio)
29 CRC E-24-2 (Riverside)
30 Ford, Chrysler, and Chase, 2000.
31 Whitney, K., Measurement of primary exhaust paniculate matter emissions from light-duty
motor vehicles. Southwest Research Institute Report E-24-3, 1998.
32 Kansas City study (summer)
33 EPA (MSAT)
34 Li, W., Collins, J.F., Norbeck, J.M., Cocker, D.R., Sawant, A., Assessment of Paniculate
Matter Emissions from a Sample ofln-Use ULEVandSULEV Vehicles. SAE 2006-01-1076,
2006.
35 Sabate, S. Methodology for Calculating and Redefining Cold and Hot Start Emissions.
March 1996. (Available at: www. epa.gov/otaq/models/mobile6/m6tech.htm)
36 Glover, E. and P. Carey, Determination of Start Emissions as a Function of Mileage and
Soak Time for 1981-1993 Model Year Light-Duty Vehicles. USEPA Office of
Transportation and Air Quality. EPA420 R 01 058 (M6.STE.003), November 2001.
(http://www.arb.ca.gov/msei/onroad/downloads/pubs/starts.pdf)
37 US Environmental Protection Agency. EPA Motor Vehicle Tampering Survey. EPA # 420-A-90-001.
1990.
38 Hare, Charles T.; Baines, Thomas M.; Characterization of Diesel Crankcase Emissions. Society of
Automotive Engineers Off-Highway Vehicle Meeting and Exhibition. MECCA, Milwaukee. 1977.
39 Bennet, PA et al. Reduction of Air Pollution by Control of Emissions from Automotive Crankscase,
SAE Paper 142A, January 1960.
120
-------
40 Bowditch, Fred W. The Automobile and Air Pollution, SAW Paper No. 680242 presented at SAE Mid-
Year Meeting Milwaukee, WI, May 1968.
41 Montalvo and Hare, 1985 EPA 460/3-84-011 Crankcase Emissions with disabledPCVsystems
42 Williamson, James. 1995. Exhaust emissions and fuel economy evaluation of SV technologies power
valve crankcase ventilation and oil recovery system. SV Technologies report
43 Kittelson, David B. Matthew Spears, Todd Taubert, and Bob Waytulonis, "Measurement and
Characterization of Crankcase Blow-by Emissions Phase I Report to Caterpillar, Inc.", University of
Minnesota, Center of Diesel Research, February 25, 1998
44 Hill, Zimmerman, Gooch A Multi-City Investigation of the Effectiveness of Retrofit Emissions
Controls in Reducing Exposures to Particulate Matter in School Buses. January 2005. Clean Air Task
Force.
45 Ireson, Robert. Intentional Tracer Measurements of Tailpipe and Crankcase Particulate Matter
Concentrations Inside School Buses: Results, Refinements, and Effects of Bus Operating Conditions .
18thCRC.
46 Detailed Characterization and Profiles of Crankcase and Diesel Particulate Matter Exhaust Emissions
Using Speciated Organics .Zielinska, B.; Campbell, D.; Lawson, D. R.; Ireson, R. G.; Weaver, C. S.;
Hesterberg, T. W.; Larson, T.; Davey, M.; Liu, L.-J. S. Environ. Sci. Technol.; (Article); 2008; 42(15);
5661-5666.
121
------- |