Professional Documents
Culture Documents
* Corresponding author:
Tel: 1-919-5151155
Fax: 1-919-5157908
Email: frey@ncsu.edu
Number of Pages: 42
Number of Figures: 16
Number of Tables: 10
This supporting information (SI) provides supplemented texts, tables and figures to further
describe the VSP-based model development and validation, and speed- and facility- specific
average emission rates estimates and their evaluation. The information provided includes:
(2) Accuracy of measurements from the Portable Emission Measurement System (PEMS);
(3) Bus link speed profiles to get insights on link-based bus activity patterns and their
(4) Data division to investigate the representative of model calibration and validation
datasets;
(5) Exploring the relationship between VSP and emissions before model development;
(10) Time distributions of VSP modes for multiple links with the same attributes;
(11) ANVOA results regarding the effect of facility type and link mean speed on transit bus
emission rates;
(12) Time distribution of VSP modes for various speed ranges; and
S1
S2. Portable Emission Measurement System (PEMS) Accuracy
The mass-based emission rates are calculated by fuel flow rate, mole fraction of pollutant and
molecular weight ratio of pollutant versus fuel (2). In an independent evaluation, the cycle
average measurements from the SEMTECH-D were compared to Federal Reference Method
(FRM) measurements (3). Based on 6 sets of measurements, the average difference in cycle
emission rate was 3% for CO2, -5% for hydrocarbons, 9% for NOx, and 35% for CO. Thus, the
measurements agreed well for CO2, HC, and NOx. The larger error for CO is attributed to low
exhaust gas concentrations, which are subject to larger errors because of detection limits.
Various vehicle activities such as deceleration and acceleration produce different levels of
vehicle emissions. Link speed profiles of transit buses were analyzed in order to get insights on
link-based activities patterns for transit buses and their association with link mean speeds and
emissions. For transit buses, a link was defined as the road segment between two bus stops. The
defined links were characterized by the segment attribution such as roadway type. Link classes
included principal arterials, minor arterials, local and collector streets. There were two types of
bus speed profiles as shown in Figure S-1: ( ) the one that had no traffic control stops; and ( )
the other that had traffic control stops. As shown in Table S-1, the speed profile ( ) had lower
mean speed than speed profile ( ). However, CO2, CO and NOx emission rates based upon the
distance (g/mile) from speed profile ( ) were larger than from speed profile ( ) whereas its
average emission for HC was lower than from speed profile , which implied that traffic control
S2
S4. PEMS Database Division
The PEMS database was divided into two parts: a calibration dataset and a validation dataset for
model development and validation. Because driving modes affect vehicle emission rates, the
similarity of bus activities between both datasets was investigated prior to using them in order to
evaluate whether both datasets were representative or not. The cumulative distribution function
(CDF) plots of vehicle speeds and accelerations were generated for the calibration and validation
datasets to ensure that activities in both datasets were comparable. As shown in Figure S-2, the
values of CDF for a given speed or acceleration were very close to each other for both datasets,
which indicates that there are similar patterns of vehicle activities between the calibration and
validation datasets and that both datasets were representative of each other.
The relationship between VSP and emissions was explored through the use of scatter plots. As
shown in Figure S-3, emission rates were very low and were almost the constant for VSP less
than -10 m2/s3. However, emission rates gradually increased with the VSP values when VSP
values were in the range from -10 to 0 m2/s3. There was generally a monotonic increase in
emissions with positive VSP for all pollutants, with the exception of CO emission rates which
tended to decrease when VSP values were above about 10 m2/s3 (see paper for an explanation as
to why this may be the case) In general, monotonic relationships motivated the development of
Examination of the data showed that there were positive VSP values in previous seconds
for the current data point whose VSP value was in the range from -10 to 0 m2/s3. Therefore, it is
S3
necessary to further investigate the randomness in the dataset. The randomness in the dataset is
ascertained by computing autocorrelations for data values at varying time lags. The sample
autocorrelation plot in Figure S-4 showed that the time series of emissions had an autocorrelation
between current and adjacent observations in previous several seconds. The autocorrelation
explained the reason that emission rates corresponding to VSP values from -10 to 0 m2/s3
As shown in Figure S-5a, CO emissions are sensitive to bus acceleration. During heavy
acceleration, more fuel is injected and more incomplete combustion may occur, which result in
high CO emissions. There are fewer samples with heavy acceleration for Mode 8 than Modes 6
As shown in Figure S-4a for Bus 1, for the time lag up to 2 seconds for CO, NOx, and HC and 6
seconds for CO2, the autocorrelation is larger than 0.50. There are similar situations for other
buses. Thus, the effect of autocorrelation on the VSP modal emissions model needs to be
evaluated.
Before evaluating the effect of autocorrelation on emission rates, the averaging duration
per VSP mode event for typical speed profiles is investigated, as shown in Figure S-6. These
two speed profiles are typical. During the vehicle driving, one speed profile has traffic control
stops, and the other has no traffic stops. These two speed profiles were on different links. The
duration per VSP event ranged from 1.0 second to more than 30.0 second for two speed profiles.
Each VSP mode event is affected by speed, acceleration and road grade. As shown in Figure S-6
(a) and (b), when the vehicle started to accelerate, there were some 1-second VSP mode events
S4
distributed among various VSP modes. When the vehicle lightly accelerated or cruised for a
while, and then decelerated on a downhill, VSP may change from positive to negative values.
Thus, VSP mode event changed, and some 1-second mode events may appear. The average
duration per VSP mode event for each mode for typical speed profiles are summarized in Table
S-2. The average duration per VSP mode event for each mode ranged from 1.2 seconds to 5.0
seconds for two speed profiles. The overall average duration per event across modes for the
original datasets are 1.7 seconds and 2.0 seconds for two speed profiles.
In order to reduce autocorrelation, the data were grouped into segments of various
consecutive times. Given average durations from 1.2 seconds to 5.0 seconds per VSP modal
event for real-world link speed profiles, the second-by-second data are averaged using
VSP modal modeling. Modal models involve averaging of the data. Therefore, it is useful to
As an example, the autocorrelation coefficients of emission rates for the 2-, 3- and 5-
second consecutive moving average datasets for Bus 1 are shown in Figure S-4. Compared to the
original second-by-second dataset, the autocorrelation coefficients for the consecutive average
t
VSP =
1
(VSPi ) (S-1)
t i =1
Where:
i = time index;
S5
VSPi = vehicle specific power at a given second i (kW/ton);
The modal emission rates were estimated based on each version of the 2-, 3- and 5-second
consecutive average calibration data. The differences for modal emission rates based on
consecutive average dataset and the original calibration dataset are given in Table S-3. The
differences of modal emission rates are classified into three ranges ( ±5%, ±10%, >±10%), and
the number of VSP modes for each range is summarized for each pollutant in Figure S-7.
The overall average differences in modal emission rates for eight modes are not more
than 3% for CO2, NOx and HC, and not more than 5% for CO, when comparing original second-
by-second and each of 2, 3 and 5 second consecutive averages. For CO2, NOx and HC, the
absolute differences in modal emission rates for individual modes are not more than 10% for the
2- and 3-second consecutive average datasets; however, for the 5-second consecutive average
dataset, there is one mode for each of these pollutants for which the absolute differences in
modal emission rates are more than 10% and less than 16%. For CO, there are two modes for
the 2-second consecutive average dataset, three modes for the 3-second consecutive average
dataset and four modes for the 5-second consecutive average dataset for which such absolute
differences are more than 10% for CO. These results indicate that the consecutive averages have
relatively larger effects on individual modal emission rates for CO than those for other pollutants
because CO has lower autocorrelation for time lags of more than 2 seconds, compared to other
Using the fleet validation dataset, modal models based on 2-, 3- and 5- second
consecutive average datasets are evaluated. Total emissions are estimated using these new modal
emission rates and compared to total measured emissions. The estimation errors for all trips
S6
emissions are given in Table S-4 and compared for different versions of consecutive average-
based models. The average difference of estimation errors in total emissions estimated for 12 trip
speed profiles is 0.1 to 1.3 percent for CO2, HC and NOx, and 1.4 to 4.1 percent for CO when
comparing each of 2, 3 and 5 second consecutive average-based models to the original modal
significant effects on trips emissions predicted by the VSP modal emissions model.
Vehicle activities affect emissions. The CDFs of vehicle speed were plotted for individual buses.
As shown in Figure S-8, buses had different activities patterns. For example, Buses 7, 9 and 10
were related to low speed activities. However, Buses 11, 14 and 15 are related to high speed
activities. For Buses 7, 9 and 10, there are 65 to 70 percent of speeds less than 20 mph, and 60
percent of speed less than 20 mph for Bus 6, whereas there are 40 percent of speeds less than 20
mph for Bus 11, and 50 to 54 percent of speeds less than 20 mph for Buses 14 and 15. Trip
average speed was 12 mph and 14 mph for Buses 9 and 10, which are much lower than average
Inter-vehicle variability in VSP modal average emission rates was investigated before
combining data from different vehicles to estimate fleet average emission rates. As shown in
Figure S-9, the modal average emission rates were estimated based upon individual vehicle data,
which in general, are significantly different. As summarized in Table S-5, the number of
samples for each of eight VSP modes for individual vehicle is more than 100, except for VSP
Mode 8 for Buses 1, 11 and 15, which indicates that there are sufficient samples to estimate
modal average emission rates. As shown in Figure S-10, in general, VSP modal average
S7
emission rates of CO2, NOx and HC monotonously increased with the VSP mode for all buses
except for modal average emission rates of HC from Buses 14 and 15.
There were similar varying trends for CO emissions for all buses. For CO, modal
average emission rates increased with VSP mode up to Modes 5 or 6 or 7 and then decreased.
Comparisons of multiple buses emissions indicated that there was a large inter-vehicle
variability in emissions for HC. Compared with modal average HC emission rates of other
vehicles, each of the eight modal average emission rates for Bus 10 were almost all the lowest.
For Buses 14 and 15, modal average emission rates for the VSP Modes 1, 2 and 3 are larger than
those for high VSP modes, which are also much larger than modal average emission rates of the
VSP Modes 1 and 2 of other buses. Different from other buses, the correlation coefficients
between HC emission rates and VSP for Buses 14 and 15 were 0.04 and -0.18, which indicated
ratio is defined as actual fuel versus air ratio normalized with respect to the stoichiometric fuel
mf
ma
= actual
(S-2)
mf
ma
stoichiometric
Where:
= Equivalence ratio;
f = index of fuel;
a = index of air;
m = the mass;
S8
mf
= fuel versus air mass ratio.
ma
As shown in Figure S-11(a), high HC emission rates in gram per second occurred across fuel
lean equivalence ratios. As an alternative, HC emission rates were estimated based on brake
horse power. As shown in Figure S-11(b), there is a more clear relationship between equivalence
ratio and brake-specific HC emissions. Similar to results by Frey et al, 2002, high HC emission
rates occurred at low values of equivalence ratio less than 0.40. As shown in Figure S-11(c),
equivalence ratio and VSP are weakly related; the R2 value between them is 0.38. In addition, the
R2 value between brake horse power and VSP is 0.26. Thus, the VSP modal approach does not
explain a large portion of the variability in HC emission rates. On the other hand, HC emission
rates and total emissions from on-road diesel engine are small relative to those from gasoline
engines.
There was a smaller inter-vehicle variability in modal average emission rates on a relative
VSP modes helps illustrate the trends that CO modal emission rate for VSP Mode 8 is smaller
than for VSP Mode 6 or 7 for some vehicles shown in Figure S-9b. Percentages of data for
heavy acceleration events in individual high VSP modes for each vehicle are summarized in
Table S-6. As shown in Figure S-11, the ratio of modal emission rates for VSP Mode 8 versus 6
or 7 increases with the ratio of percentage of heavy acceleration more than 3 mph/s. For vehicles
with the modal emission rates ratios less than 1.0 for VSP Mode 8 versus 6 or 7, there are higher
percentages of samples with heavy acceleration for VSP Modes 6 and 7 than for VSP Mode 8,
S9
which account for lower modal emission rate for VSP Mode 8; for vehicles with the modal
emission rates ratios more than 1.0, there are higher percentages of heavy acceleration events
Once VSP mode average emission rates were estimated based upon the model calibration dataset,
validation was carried out. The validation database consisted of 12 subsets, each of which was
selected from individual vehicles and was ten minutes in length as described in the main paper.
Using modal average emission rates with 95% confidence intervals shown in Figure 2 of the
main paper, total emissions of individual subsets were estimated for each of buses and then
compared with corresponding measured total emissions. Subsequently, estimation errors were
Because VSP mode average emission rates had large confidence intervals, especially for
CO and HC, uncertainty estimates of total emissions were included in the model validation. Total
emissions for each bus are estimated with uncertainty ranges as follows:
8
TE = Ti × ERi (S-3)
i =1
8
t (11,0.025) × SDi
TE Lower = Ti × ( ERi ) (S-4)
i =1 12
8
t (11,0.025) × SDi
TE Upper = Ti × ( ERi + ) (S-5)
i =1 12
Where:
TE Lower = Lower boundary of total estimated emissions for a bus trip (g);
S10
TE Upper = Upper boundary of total estimated emissions for a bus trip (g);
SDi = Standard deviation of modal average emission rates of 12 buses for Mode i (g/s)
intervals.
Total emissions were estimated based on individual data subset. Uncertainty ranges of
estimation errors for 12 buses are given in Figure S-13. The average values of mean estimation
errors of 12 buses for selected pollutants are displayed in Figure S-14. The results show that the
VSP modeling approach generated different estimation errors for different pollutants. The mean
estimation error for CO2 for one bus was less than +/- 20%. For CO, NO and HC, estimation
errors substantially decreased when emissions increased. However, for HC emissions, the
approach exhibited significantly different performance for different vehicles. Zero estimation
error was enclosed by the uncertainty ranges of estimation errors for many buses. The average
values of the 12 mean estimation errors with 95% CIs shown in Figure S-14 indicated that VSP
modeling has good ability to predict CO2, NOx and CO emissions from diesel transit buses.
However, it had a larger estimation error for HC emissions due to relatively larger inter-vehicle
S10. Time Distributions of VSP Modes for Multiple Links with the Same Attributes
Emission estimates based on the VSP-modeling approach and measured speed profiles were
determined by the fraction of travel time spent in each VSP mode. It was necessary to
S11
investigate whether patterns of VSP mode time distributions were similar among multiple links
with the same speed ranges and facility type before data from multiple links (in the same speed
range and roadway type) were combined to estimate link-based speed- and facility-specific
average emission rates. For example, time distributions of VSP modes for selected link speed
profiles on principal arterials were calculated for mean link speed from 25 mph to 30 mph. As
shown in Figure S-15a, about 37 % of the time on average was spent in VSP Mode 1, and less
than 12 % was spent in any other mode. As shown in Figure S-15b, the CV value for the
percentage of time distribution is 0.25 for VSP Mode 1 and ranges from 0.41 to 0.72 for other
modes. Time distribution was similar among multiple links for this speed range and data from
S11. ANOVA Test on the Effects of Facility Type and Mean Speed on Buses Emissions
An ANOVA test was carried out to evaluate the effect of facility type and mean speed on transit
buses emissions for each of selected pollutants. Three speed ranges (10-15 mph, 15-20 mph and
20-25mph) and three facility types (major arterial, minor arterial, local and collector) were
included in the ANOVA test. As shown in Table S-7, all p-values corresponding to facility type
were more than .05 and p-values corresponding to mean speed were less than .001, which
implied that facility type did not appear to be a major explanatory variable when compared with
speed.
Facility type did not appear to be a major explanatory variable. Thus, data across facility types
could be combined to increase the sample size and reduce the statistical sampling error. Based on
S12
the combined database, the time distribution of VSP Modes for various speed ranges is shown in
Figure S-16. VSP Mode 1 dominated the link travel time for all speed bins. With an increase in
link mean speed, the percent of time spent in Mode 1 decreased whereas the time spent in higher
Tables S-8, S-9 and S-10 are given to complement the results for Figures 1(b), 2 and 3 in the
main paper.
References
http://www.dieselnet.com/standards/us/hd.php.
(2) Carl Ensfield Sensors. On-Road Emissions Testing of 18 Tier 1 Passenger Cars and 17
Ensfield Sensors, Inc. for U.S. Environmental Protection Agency, Ann Arbor, MI,
October 2002.
(3) Durbin, T.D.; Johnson, K.; Cocker III, D.R.; Miller, J.W.; Maldonado, H.; Shah, A.;
Ensfield, C.; Weaver, C.; Akard., M.; Harvey, N.; Symon, J.; Lanni, T.; Bachalo, W.D.;
measurement systems and federal reference methods for emissions from a back-up
generator and a diesel truck operated on a chassis dynamometer. Environ. Sci. Technol.
S13
List of Figures and Tables in the Supporting Information
Table S-1 Link mean speed and average emissions for selected speed profiles
Table S-2 Averaging length of time per VSP mode event
Table S-3 Differences of modal emission rates based on the original and consecutive average
calibration datasets
Table S-4 Estimation errors for the fleet validation dataset based on the original and consecutive
average calibration datasets
Table S-5 Sample size for VSP modes for individual vehicles
Table S-6 Percentage of samples for heavy acceleration events in individual high VSP modes for
each vehicle (%)
Table S-7 ANOVA tests for effects of facility type and mean speed on link average emission
rates
Table S-8 Definition of VSP modes for transit buses
Table S-9 Fleet-based modal average emission rates for diesel buses
Table S-10 Link-based average emission rates for diesel buses
S14
50
I
40 II
Speed (mph)
30 Bus Stop Traffic Control Stop Bus Stop
20
10
0
0 50 100 150 200 250
Time Trace (sec)
S15
100
Calibration Dataset
Validation Dataset
Percent (%) 75
50
25
0
0 10 20 30 40 50 60
Speed (mph)
(a) Speed
100
Calibration Dataset
Validation Dataset
75
Percent (%)
50
25
0
-10 -5 0 5 10
Acceleration (mph/s)
(b) Acceleration
Figure S-2 Empirical cumulative distribution functions for calibration and validation datasets.
S16
0.24
0.12
0.06
0.00
-30 -10 10 30
VSP (kW/ton)
(a) CO
0.6
NO Emission Rate (g/s)
0.45
0.3
0.15
0
-30 -10 10 30
VSP (kW/ton)
(b) NOx
HC Emission Rate (mg/s)
3.6
2.7
1.8
0.9
0.0
-30 -10 10 30
VSP (kW/ton)
(c) HC
Figure S-3 Emission rates vs. vehicle specific power based on Bus 1 data
S17
1.00
CO2
CO
0.75
NOx
Autocorrelation
HC
0.50
0.25
0.00
-0.25
0 1 2 3 4 5 6 7 8 9 10
Time Lag (sec)
HC
0.50
0.25
0.00
-0.25
0 1 2 3 4 5 6 7 8 9 10
Interval Lag
S18
1.00
CO2
CO
0.75
NOx
Autocorrelation
HC
0.50
0.25
0.00
-0.25
0 1 2 3 4 5 6 7 8 9 10
Interval Lag
1.00
CO2
CO
0.75
NOx
Autocorrelation
HC
0.50
0.25
0.00
-0.25
0 1 2 3 4 5 6 7 8 9 10
Interval Lag
S19
1
0.1
CO (g/s)
0.01
0.001
0.0001
-6 -4 -2 0 2 4 6
Acceleration (mph/s)
0.1
CO (g/s)
VSP Mode 6
0.01
VSP Mode 7
VSP Mode 8
0.001
0 1 2 3 4 5
Acceleration (mph/s)
100%
Cumulative Frequency
75%
50%
VSP Mode 6
25% VSP Mode 7
VSP Mode 8
0%
0 1 2 3 4 5
Acceleration (mph/s)
S20
50 8
7
40 6
Speed (mph)
VSP Mode
30 5
4
20 3
10 2
1
0 0
0 25 50 75 100 125
Time(sec)
Speed
(a) Speed Profile I VSP Mode
50 8
7
40 6
Speed (mph)
VSP Mode
30 5
4
20 3
10 2
1
0 0
0 25 50 75 100 125 150 175 200 225
Time (sec)
Speed
(b) Speed Profile II VSP Mode
100%
Cumulative Frequency
75%
50%
Speed Profile I
25% Speed Profile II
0%
0 5 10 15 20 25 30 35 40 45 50
Duration per VSP Mode Event (sec)
(c) Cumulative time-weighted frequency of duration per VSP mode event for two speed profiles
Figure S-6 VSP mode events for typical speed profiles
S21
25
Difference
0
CO2 CO NOx HC Total
20 ±5%
±10%
15 > ±10%
10
0
CO2 CO NOx HC Total
a) Original versus 3-second consecutive average-based models
25
Differences
Number of VSP Modes
20 ±5%
±10%
15
> ±10%
10
0
CO2 CO NOx HC Total
S22
100% Bus1
Bus 10
Bus2
Bus 9
Bus4
75% Bus 7 Bus 5
Bus5
Bus6
Bus7
Percent
50% Bus8
Bus 11 Bus9
Bus 14 Bus10
Bus11
25% Bus 15
Bus14
Bus15
0%
0 10 20 30 40 50 60 70
Speed (mph)
Figure S-8 Empirical cumulative distribution function of speeds for individual buses
S23
40
30
CO2 (g/s)
20
10
0
1 2 3 4 5 6 7 8
VSP Mode
(a) CO2
0.5
0.4
NOx (g/s)
0.3
0.2
0.1
0
1 2 3 4 5 6 7 8
VSP Mode
(b)
Figure S-9 VSP modal average emission rates based on one bus data
Continued on next page
S24
0.10
0.08
CO (g/s) 0.06
0.04
0.02
0.00
1 2 3 4 5 6 7 8
VSP Mode
(c)
4.0
3.0
HC (mg/s)
2.0
1.0
0.0
1 2 3 4 5 6 7 8
VSP Mode
(d)
Figure S-9 Continued
S25
40 Bus1
Bus2
Bus4
30 Bus5
Bus6
CO2 (g/s) Bus7
20 Bus8
Bus9
Bus10
10 Bus11
Bus14
Bus15
0
0 1 2 3 4 5 6 7 8
VSP Mode
(a) CO2
Bus1
0.25
Bus2
Bus4
0.20 Bus5
Bus6
0.15 Bus7
CO (g/s)
Bus8
Bus9
0.10 Bus10
Bus11
0.05 Bus14
Bus15
0.00
0 1 2 3 4 5 6 7 8
VSP Mode
(b) CO
Figure S-10 Inter-vehicle variability in modal average emission rates.
Continued on next page
S26
0.50 Bus1
Bus2
Bus4
0.40 Bus5
Bus6
NOx (g/s) 0.30 Bus7
Bus8
Bus9
0.20 Bus10
Bus11
0.10 Bus14
Bus15
0.00
0 1 2 3 4 5 6 7 8
VSP Mode
(c) NOx
Bus1
4.0
Bus2
Bus4
Bus5
3.0
Bus6
HC (mg/s)
Bus7
Bus8
2.0
Bus9
Bus10
Bus11
1.0
Bus14
Bus15
0.0
0 1 2 3 4 5 6 7 8
VSP Mode
(d) HC
Figure S-10 Continued
S27
0.008
0.006
HC (g/s) 0.004
0.002
0
0.0 0.2 0.4 0.6 0.8 1.0
Equivalence Ratio
0.6
0.4
0.2
0
0.0 0.2 0.4 0.6 0.8 1.0
Equivalence Ratio
50
R2 = 0.38
25
VSP (kW/ton)
-25
-50
0.0 0.2 0.4 0.6 0.8 1.0
Equivalence Ratio
S28
Ratio of Modal Emission Rate for
1.5
y = 0.18Ln(x) + 0.79
0.5
R2 = 0.88
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
Ratio of Percentage of Heavy Acceleration for VSP
Mode 8 versus 6
1.5
VSP Mode 8 versus 7
1.0
y = 0.21Ln(x) + 0.80
0.5
R2 = 0.77
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
Ratio of Percentage of Heavy Acceleration for VSP
Mode 8 versus 7
S29
75
25
-25
-50
0 2000 4000 6000 8000 10000
CO2 Measured Emissions per Bus (g)
(a) CO2
150
Error of Estimate (%)
100
50
-50
0 30 60 90 120 150
NOx Measured Emissions per Bus (g)
(b) NOx
Figure S-13 Errors of estimates of total emissions based upon individual validation data subsets a
Continued on next page
S30
250
200
-50
-100
0 5 10 15 20 25 30
CO Measured Emissions per Bus (g)
(c) CO
1000
Error of Estimate (%)
750
500
250
-250
0.0 0.5 1.0 1.5 2.0 2.5
HC Measured Emissions per Bus (g)
(d) HC
Figure S-13 Continued a
a
represents uncertainty ranges of errors of emission estimate
S31
Individual Average Estimation Error (%)
200
150
100
50
-50
-100
CO2 CO NOx HC
Figure S-14 Individual bus average estimation errors for trip emissions a
a
The boxplot has lines at the smallest observation, lower quartile, median, upper quartile and
largest observation. Any data observation which lies more than 1.5*IQR (interquartile range,
subtracting the lower quartile from the upper quartile) lower than the first quartile or 1.5*IQR
higher than the third quartile is considered an outlier.
S32
50%
40%
Percentage of Time
30%
20%
10%
0%
1 2 3 4 5 6 7 8
VSP Mode
(a) Average percentage of time for a given VSP mode and 95% confidence interval on mean
0.8
Coefficient of Variation
0.6
0.4
0.2
0.0
1 2 3 4 5 6 7 8
VSP Mode
S33
80
Link Mean Speed
5-10 mph
Percentage of Time (%) 60 10-15 mph
15-20 mph
20-25 mph
40
25-30 mph
20
0
1 2 3 4 5 6 7 8
VSP Mode
Figure S-16. Average time distribution of VSP modes for all facility types based upon data
for 7 buses for which GPS data were available
S34
Table S-1 Link mean speed and average emissions for selected speed profiles
Link Mean Speed Average Emission (g/mile)
Speed Profile
(mph) CO2 CO NOx HC
32.9 1322.5 3.68 20.44 0.25
24.3 1360.5 5.65 21.07 0.17
S35
Table S-3 Differences of modal emission rates based on the original and consecutive average calibration datasets (%)
VSP Mode CO2 CO NOx HC
2-Sec 3-Sec 5-Sec 2-Sec 3-Sec 5-Sec 2-Sec 3-Sec 5-Sec 2-Sec 3-Sec 5-Sec
1 -6 6 -1 -2 3 7 -3 5 0 2 -1 4
2 -5 8 -12 -7 9 -12 -7 8 -16 -4 5 -11
3 -2 3 -4 9 -7 13 1 3 -6 2 3 -1
4 -4 5 -3 -11 8 5 -2 2 -4 -4 2 -3
5 0 -2 2 0 -4 -3 2 -3 3 -1 2 -4
6 3 -5 6 8 -11 -5 2 -4 5 -1 -1 1
7 6 -7 8 2 14 -19 4 -5 8 5 -2 1
8 5 -6 7 -11 13 -29 2 -1 9 1 1 -10
Average across modes 0 0 0 -2 3 -5 0 1 1 0 1 -3
Table S-4 Estimation errors for total emissions based on the fleet validation dataset using modal models developed from the original
and consecutive average calibration datasets (%)a
Calibration Dataset
Pollutant
Original Dataset 2-Sec Moving Average Dataset 3-Sec Moving Average Dataset 5-Sec Moving Average Dataset
CO2 2.6 2.5 2.4 3.3
CO 5.9 4.5 2.7 1.8
NOx 4.3 3.7 3.0 3.4
HC 3.0 3.6 2.7 1.8
a
Based on the fleet validation dataset, total emissions estimates using modal models are compared to total measured emissions to
calculate estimation errors.
S36
Table S-5 Sample size for VSP modes for individual vehicles
VSP Bus Number
Sum
Mode 1 2 4 5 6 7 8 9 10 11 14 15
1 991 1577 3874 4286 3801 4743 4037 4503 4162 2097 2570 2704 39345
2 213 415 960 678 643 364 786 538 1142 627 455 685 7506
3 221 306 760 514 491 254 465 324 646 772 379 472 5604
4 164 266 549 426 384 250 484 550 555 704 298 386 5016
5 154 206 491 355 335 269 509 606 412 425 225 336 4323
6 170 166 462 369 281 417 471 356 253 177 168 240 3530
7 166 161 496 403 343 606 400 254 150 115 217 165 3476
8 98 124 252 279 409 419 118 206 146 55 371 97 2574
Sum 2177 3221 7844 7310 6687 7322 7270 7337 7466 4972 4683 5085 71374
S37
Table S-6 Percentage of samples for heavy acceleration events in individual high VSP modes for
each vehicle (%)
Percent of Samples for Acceleration > 3 mph/s in Given VSP Modes
Vehicle Number
Mode 6 Mode 7 Mode 8
2 9 7 5
4 15 21 12
5 10 8 3
6 11 12 12
7 13 8 2
8 4 2 2
9 9 14 55
10 11 23 54
11 7 11 35
14 6 7 11
15 10 7 42
S38
Table S-7 ANOVA tests for effects of facility type and mean speed on link average emission
rates
Pollutant Source DF Sum of Squares F Ratio Prob. > F
Mean Speed 2 2.94E+02 4.69E+01 <.001
CO2
Facility Type 2 1.84E+00 2.94E-01 0.75
Mean Speed 2 3.45E-03 5.61E+01 <.001
CO
Facility Type 2 7.21E-05 1.17E+00 0.31
Mean Speed 2 3.46E-02 6.07E+01 <.001
NOx
Facility Type 2 2.25E-04 3.94E-01 0.67
Mean Speed 2 5.15E-07 6.23E+01 <.001
HC
Facility Type 2 1.14E-08 1.38E+00 0.25
S39
Table S-8 Definition of VSP modes for transit buses
VSP Mode VSP Range (kW/ton)
1 VSP 0
2 0<VSP<2
3 2 VSP<4
4 4 VSP<6
5 6 VSP<8
6 8 VSP<10
7 10 VSP<13
8 VSP 13
S40
Table S-9 Fleet-based modal average emission rates for diesel buses
Modal Average Emission Rates
VSP Mode
CO2 (g/s) CO (g/s) NOx (g/s) HC (mg/s)
1 2.4 0.009 0.04 1.23
2 7.8 0.036 0.13 1.70
3 12.5 0.045 0.18 1.75
4 17.1 0.072 0.22 1.84
5 21.2 0.085 0.24 1.94
6 24.8 0.091 0.26 2.05
7 27.6 0.084 0.28 2.08
8 29.5 0.062 0.31 2.15
S41
Table S-10 Link-based average emission rates for diesel buses
Link Mean Link-based Average Emission Rates
Speed (mph) CO2 (g/s) CO (g/s) NOx (g/s) HC (mg/s)
5-10 6.3 0.024 0.09 1.42
10-15 8.8 0.031 0.11 1.49
15-20 10.6 0.038 0.13 1.57
20-25 11.8 0.042 0.145 1.61
25-30 12.7 0.044 0.155 1.66
S42