Environmental Modelling & Software 23 (2008) 12621270

Contents lists available at ScienceDirect

Environmental Modelling & Software

journal homepage: www.elsevier.com/locate/envsoft

Do air pollution emissions and fuel consumption models for roadways

include the effects of congestion in the roadway trafc ow?
R. Smit*, A.L. Brown, Y.C. Chan
Grifth School of Environment, Grifth University, Brisbane, QLD 4111, Australia

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 4 October 2007
Received in revised form 2 March 2008
Accepted 3 March 2008
Available online 21 April 2008

Road transport emission and fuel consumption models are currently used extensively to predict levels of air
pollution along roadway links and networks. This paper examines how, and to what extent, models which
are currently used to predict emissions and fuel consumption from road trafc include the effects of
congestion. A classication framework is presented in which a key factor, driving pattern, connects emissions to congestion. Prediction of the effects of different driving patterns in emission models is generally
restricted to certain aspects of modelling, i.e. hot-running emissions of regulated pollutants. As a consequence, the effects of congestion are only partially incorporated in the predictions. The majority of emission
models explicitly incorporate congestion in the modelling process, but for one important family of emission
models, namely average speed models, this could not be determined directly. Re-examination of the (lightduty) driving patterns on which three average speed models (COPERT, MOBILE, EMFAC) are based, shows
that it is likely that congestion is represented in these patterns. Since (hot-running) emission factors are
based on these patterns, this implies that the emission factors used in these emission models also reect
different levels of congestion. Congestion is thus indirectly incorporated in these models. It is recommended, that, in order to get more accurate (local) emission predictions and to achieve correct application
in particular situations, it is important to improve current average speed models by including a congestion
algorithm, or alternatively, at least provide information on the level of congestion in the driving patterns on
which these models are based and recommendations on what applications the models are suitable for.
2008 Elsevier Ltd. All rights reserved.

Keywords:
Congestion
Average speed
Emission model
Driving patterns
Road trafc
Fuel consumption

1. Introduction
Major efforts have been made in recent decades to reduce air
pollution and improve urban air quality. Despite this, air quality
issues such as photochemical smog formation and visibility degradation have proven to be persistent. Estimates are that, worldwide,
nearly one billion people in urban environments are continuously
being exposed to health hazards from air pollutants (Ahrens, 2003).
Around the world road trafc is the dominant anthropogenic
source of air pollution in urban areas (Fenger, 1999). This is not only
because of the magnitude of its emissions, but also because pollutants are emitted in close proximity to people, thus enhancing
exposure levels.
Road transport has grown continuously over the last decades
and further increase in the demand for transport is projected (e.g.
BTE, 2000). Not only is there ongoing growth in the trafc ows on
road links and networks, the incidence of congested trafc conditions has become increasingly common and severe, particularly in

* Corresponding author. Present address: Pacic Air & Environment, PO Box 3306,
South Brisbane, QLD 4101, Australia. Tel.: 61 7 3004 6400; fax: 61 7 3844 6400.
E-mail address: robin.smit@yahoo.com.au (R. Smit).
1364-8152/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.envsoft.2008.03.001

major cities (e.g. Austroads, 2000). This paper addresses the issue of
whether current emission models for road transport take congestion into account, either implicitly or explicitly.
Trafc congestion has repeatedly been indicated as a major
factor in road trafc emissions and air quality degradation (e.g.
Oduyemi and Davidson, 1998; Carr et al., 2002). A sensitivity
analysis conducted for an urban network (Brisbane, Australia) indicated that, after trafc activity (expressed as vehicle kilometres
travelled (VKT)), congestion is the most important contributor to
predicted total emissions for CO and HC (Smit, 2006). Emission
tests on modern cars with advanced emission control systems
(DoTRS, 2001) have demonstrated that their emissions are particularly sensitive to the occurrence of congestion in the driving cycle.
Emission models are commonly used to provide trafc emission
information for the prediction and management of air pollution
levels near roadways (Smit et al., 2008) and this may be at different
scales ranging from road links to city-wide road networks (Affum
and Brown, 1999). How, and to what extent, do these current
emission models1 include the effects of congestion in the link or

1
For readability purposes, the term emission is used in this paper to indicate
air pollutant emissions, fuel consumption and CO2 emissions.

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

road network trafc streams? A search of the literature yielded no

study to date that had specically investigated this question, and
the current paper begins to address this issue.
2. Road trafc congestion and vehicle emission models
Road trafc congestion is the deterioration of the quality of road
trafc ow on a network element or in an entire road network that
is a departure from smooth free-owing driving conditions due to
increased travel demand and/or reduced capacity for trafc
movement. Congestion can be associated with a number of interrelated and observable effects that include increased trafc
density, increased travel times on the same route (hence lower
average speeds), increased deviation from free-ow speeds (hence
increased speed uctuation and delays) and increased queuing
(hence increased number of stops). Many different indicators can
and have been used to quantify congestion (e.g. TRB, 2000) and
some relevant ones are discussed in Section 4.
Intuitively, an examination of how congestion is included in
current vehicle emission models should address the extent to
which these different indicators are utilized as variables. However,
it will be shown below that most emission models do not explicitly
include such parameters. This means that the link between
measures of congestion and emission models has to be sought
elsewhere. This paper focuses on vehicle driving patterns2 and
driving cycles,3 which are inherent components of most emission
models.
Trafc engineering literature (e.g. ARRB, 1990), trafc ow
theory (e.g. shock wave theory, queuing theory; May, 1990) and
empirical evidence (e.g. Barth et al., 1996) shows that, when
congestion is present in a road link, driving patterns of individual
vehicles in the trafc stream will be measurably different to those
that will occur on the link when congestion is not present.

1263

consumption, with emission factors (amount of pollutant per unit

of vehicle activity) to estimate emission levels. Vehicle activity data
may be obtained from national statistics or, when a higher resolution is required, through a combination of modelled or measured
trafc volume data (with corresponding numbers of lanes and road
lengths) in the case of VKT.
Type C models are most widely used in practice. A review of 58
international studies that involved application of trafc emission
models to road networks, revealed that 81% of these studies used
Type C emission models. Type A models were applied in 16% of
these studies, whereas Type B models were applied in only 3%
(Smit, 2006).
In addition to the type of emission model, congestion may be
implicitly or explicitly included as a variable in the modelling
process. Implicit means that the effects of congestion are modelled,
but the level of congestion in the model cannot be modied and is
therefore beyond the control of the user. Explicit means that the
effects of congestion are modelled and modiable, and that level of
congestion can be controlled by the user.
Smit (2006) found that prediction of the effects of different
driving patterns is generally restricted to certain aspects of modelling. Emission models often consider the effects of congestion on
hot-running emissions alone, inherently assuming that cold-start
and running-loss emissions are independent of the driving pattern.
This is despite indications that congestion signicantly affects coldstart and running-loss emissions (Joumard and Serie, 1999; Pierson
et al., 1999). Similarly, congestion effects are often only predicted
for regulated pollutants, but not for other unregulated pollutants,
although the effects on speciated hydrocarbons may be predicted
indirectly through the use of hydrocarbon proles (Smit et al.,
2002). As a consequence, the effects of congestion are only partially
incorporated in the predictions.
3.1. Type A models

3. Classifying emission models according to their

consideration of congestion
It is useful to suggest a typology of emission models based upon
the way in which congestion is addressed. The factor driving pattern
is key to relating vehicle emissions to congestion, but as different
emission models include driving pattern data in the modelling
process in different ways, three categories of emission models are
suggested:
 those that require driving pattern data as input (Type A);
 those that generate driving pattern data as part of the emission
modelling process (Type B); and
 those that have incorporated driving pattern data in the development phase of the model (Type C).
Type A and B models are well suited for analyzing changes to
emission levels at the local level (e.g. intersections, small networks), whereas Type C models are better suited to operate at
a larger scale (e.g. urban regions, small areas) (Smit et al., 2007).
Fig. 1 shows how these three types of model use driving pattern
data to predict the effects of congestion on emissions. Basically, all
emission models combine information (by vehicle class) on vehicle
activity, i.e. vehicle kilometres travelled (VKT) and sometimes fuel

2
A driving pattern quanties a particular trafc situation and it is dened as
a speedtime prole that has been recorded (measured) on the road.
3
A driving cycle is a speedtime prole which has been synthesized from a set of
driving patterns. It is commonly assumed to be representative of (a mixture of)
certain trafc and road conditions in a particular geographic area. In this paper, the
difference between driving pattern and driving cycle is not relevant. Hence both are
referred to as driving patterns in the remainder of this paper.

Type A models require actual driving pattern data (modal

emission models) or specic congestion variables (multivariate
regression models) as input to determine emission factor values for
different vehicle classes. Modal emission models basically combine
instantaneous emission factors (g s1) for each vehicle class, which
may be modelled as a function of engine power and gear shift
behaviour (e.g. Leung and Williams, 2000; Schulz et al., 2000;
Zalinger et al., 2005), with information on time spent in each
driving mode by location. Multivariate regression models require
driving pattern data to quantify a substantial set of model input
variables, which is then used to estimate emission factors (g km1,
which is equivalent to g VKT1) for different vehicle classes.
Examples of these are MEASURE (Fomunung et al., 2000) and
VERSIT (Smit et al., 2007). Type A models explicitly take congestion into account as they require actual driving pattern data as
input. They are thus fully capable of modelling the effects of
congestion on emissions, provided that congestion is reected in
the input driving patterns.
3.2. Type B models
Type B emission models generate (simplied) driving pattern
data as a function of a number of macroscopic trafc variables relating to trafc characteristics (e.g. trafc volume, average speed,
trafc density) and road infrastructure characteristics (e.g. link
length, number of lanes, free-ow speed, type of intersection, signal settings). These driving patterns are either combined with instantaneous emission factors (g s1) or are used to compute
correction algorithms for incorporated emission factor tables (g
km1). Examples are the TEE model (Negrenti, 1999) and queuing
emission models (e.g. Matzoros, 1990). Since Type B models use

1264

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

Fig. 1. The relationship between trafc activity, congestion and emissions in Type A, B and C emission models.

congestion-related variables in the emission modelling process,

they explicitly take congestion into account.
3.3. Type C models
In Type C models measured driving pattern data (driving patterns) are used to develop distance-based emission factors (g km1)
or fuel-based emission factors (g kg1 of fuel). Emission factors
used in Type C models are either constant, as is the case for areawide and fuel-based models, or are a function of:
 one continuous quantitative driving pattern variable (average
speed model);
 a discrete quantitative or qualitative description of a trafc
situation (trafc situation model).
Area-wide models (e.g. AGO, 2003) use data on total VKT and
a single emission factor to compute total area emissions. Fuelbased models use data on total fuel consumption and fuel-based
emission factors to compute total area emissions (e.g. Pokharel
et al., 2002).
So-called average speed models, like MOBILE (USEPA, 2007),
EMFAC (CARB, 2002), COPERT (EEA, 2000) and others (e.g.
Namdeo et al., 2002), are regularly applied in practice (50% of the
58 studies referred to above). In these models, emission factors
(g km1) are stated as a function of average speed. Trafc situation

models use discrete emission factors (g km1) for certain trafc

situations, which can be dened in terms of a textual description
(e.g. INFRAS, 2007) or by a set of quantitative variables (e.g. TNO,
2001).
A relevant feature of Type C models is that driving patterns
are xed and cannot be changed by the user. Importantly, for
all Type C models, except the trafc situation model, it is not
immediately obvious what level of congestion is contained in
the driving patterns used during model development since
they are hidden in the model. For instance, it may be that
the driving patterns are smooth speed-time proles with almost constant speed (uncongested conditions) or, alternatively,
they may be representative of different levels of congestion
with different degrees of speed uctuation, but this is not
evident.
A notable exception is MOBILE 6. The driving patterns used
in MOBILE 6 are congestion and road-type specic, as collected
driving pattern data were categorized according to the measured congestion level, using the qualitative measure LOS
(level of service), and basic road type (USEPA, 1997). MOBILE 6
thus implicitly takes different levels of congestion into account.
The term implicit is still an appropriate label indicating that it
is not supercially obvious that MOBILE 6 takes congestion
into account because it uses a non-congestion variable (average speed) as model variable for congestion, as discussed
later.

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

1265

Table 1
Potential congestion indicators for driving cycle analysis including symbols, formulation, units and thresholds
Symbol
n
T*idle
n*
Pidle
Pacc dec
sa

Name
Average speed
Unit idle time
Number of full stops
Proportion idle
timef
Proportion speed
uctuationf
Acceleration noise

Formulaea

Coefcient of variation of speed

TAD

Total absolute
difference in speed
Positive kinetic
energye
Delay rate
Congestion index
Speed reduction
congestion index

PKE
DR
CI
SRCI

1

n 60d=T

Tidle
Tidle =d
n n=d
Pidle 100Tidle =T

km h
min km1
stops km1
%

0
0
No idle

Paccdec 100Tacc Tdec =T

r
R

sa

ms

0.15c/0.20d

6%

m s1 km1

m s2

0.20

Trun

at a2 dt
Trun

r
R
T

COV

Congestion thresholdb

Unit

nt n dt

0
COV n1
T
P
jnt;a vt;b j
TAD
d
P 2 2
nt;a nt;b
PKE
d

d
d

3600n1

DR
CI nnff
SRCI nffnn
ff

2

n1

ff

s km

1

(close to zero)
(close to unity)
(close to zero)

a
a, mean acceleration (m s2); at, instantaneous acceleration (m s2); d, length of driving cycle (km), d, average delay (min); n, total number of stops (), i.e. number of times
a vehicle becomes stationary (nt 0 km h1); T, total travel time of a driving cycle (min); Tidle, total stopped time in a driving cycle (min); Tacc, total time spent in acceleration
driving mode (min); Tdec, total time spent in deceleration driving mode (min); Trun, running time (s); n, mean cycle speed (km h1); nff, mean speed under free-ow conditions
(km h1); nt, instantaneous speed (m s1); nt,a, initial instantaneous speed in an acceleration manoeuvre (km h1); nt,b, nal instantaneous speed in an acceleration manoeuvre
(km h1).
b
, no particular value is available, an indication for the threshold value is given within parentheses.
c
For uninterrupted roads (e.g. freeways).
d
For interrupted roads (e.g. arterial roads).
e
Subject to at > 0 m s2.
f
Percent of time spent in a particular fundamental driving mode (i.e. idle, acceleration, cruise, deceleration) where idle is dened as nt 0 km h1 and at 0 m s2;
acceleration is dened as nt > 0 km h1 and at > 0.1 m s2; deceleration is dened as nt > 0 km h1 and at < 0.1 m s2; and cruise is dened as nt > 0 km h1 and
0.1  at  0.1 m s2.

For other commonly used Type C models, like COPERT and

EMFAC, further analysis of the driving patterns is required to investigate the extent to which congestion is incorporated. This is
discussed in the next section.
4. A procedure to assess the extent of congestion present in
driving patterns in Type C emission models
For the driving patterns used in emission models, information is
rarely available on the prevailing trafc and road characteristics at
the time the patterns were measured. Therefore an indirect approach to quantify the level of congestion in any driving pattern
had to be developed by the current authors.
The trafc engineering literature was reviewed to identify
potential congestion indicators which could be calculated directly
from the driving pattern data used in emission models (Smit,
2006). Nine such indicators were selected: average speed (n),
unit idle time (T*idle), proportion idle time (Pidle), number of full
stops4 (n*), acceleration noise (sa), coefcient of variation of
speed (COV), total absolute difference in speed (TAD), positive
kinetic energy (PKE) and proportion of speed uctuation
(Pacc dec).
Furthermore, indicators that explicitly quantify the extent of
departure of any particular driving pattern from free-ow speed
conditions are useful because (1) they directly correspond to the
denition of congestion, (2) unlike some other indicators, they
possess an unambiguous relationship with the level of congestion,
and (3) they can circumvent the need for multiple congestion
indicators as discussed later. Three such indicators were found in
the literature: namely delay rate (DR), congestion index (CI) and
speed reduction congestion index (SRCI).

4
A full stop is dened as a vehicle that becomes stationary for at least one
second.

Table 1 presents an overview of the 12 potential congestion

indicators for driving pattern analysis, including their formulation and, where available, congestion thresholds as determined by Smit (2006). A congestion threshold is a particular
value of a congestion indicator which attempts to create a dichotomous separation of uncongested and congested trafc
conditions.
DR, CI and SRCI have been used in the trafc eld (Taylor, 2000),
but have not yet been used for driving pattern analysis as they
require average free-ow speed in their computation and this is not
provided by the speed-time data. Therefore, a new methodology
was developed to estimate free-ow speed for each driving pattern
segment in the driving patterns. Appendix A presents this method
and its validation.
Study results provided by Greenwood (2003) and others (e.g.
Barth et al., 1996; Turner et al., 1996; Lyons et al., 1996; Skabardonis
and Dowling, 1997) suggest that all the indicators in Table 1, except
average speed, show a consistent correlation with the level of
congestion in a driving pattern.
Average speed is not an adequate congestion indicator in
certain speed intervals (between about 15 and 60 km h 1 ) as
the relationship between average speed and level of congestion is road-type specic. For instance, an average speed
of 60 km h 1 on an arterial road could represent uncongested free-owing conditions, whereas the same speed on
a freeway would represent much more congested conditions,
possibly involving low speed stop-and-go conditions. However, since average speed is directly linked to emission
factors in the average speed models, average speed is used
here as a gauge to assess the level of congestion in emission
models.
In the next section the procedure is applied to three average
speed models. As there is not enough space to present the 11
computed indicator values for all 60 driving patterns used in these
models, the next section reports on the main ndings and includes

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

100

1.2
1.0

Pacc + Pdec [%]

Acceleration Noise [m/s^2]

1266

0.8
0.6
0.4
0.2

80
60
40
20
0

0.0
0

20

40

60

80

100

120

20

Average Speed [km/h]

40

60

80

100

120

Average Speed [km/h]

SRCI [-]

log(DR) [s/km]

10

6
4
2

0
0

20

40

60

80

100

120

20

Average Speed [km/h]

60

80

100

120

MOBILE 6 ARTERIAL
1.5

MOBILE 6 FREEWAY
EMFAC 2000

10

COV [-]

Unit Idle Time [min/km]

15

40

Average Speed [km/h]

COPERT III
5

1.0

0.5

0.0
0

20

40

60

80

100

120

Average Speed [km/h]

20

40

60

80

100

120

Average Speed [km/h]

Fig. 2. Congestion in three average speed models as predicted with different variables (COPERT III, 33 cycles; EMFAC 2000, 14 cycles; MOBILE 6 ARTERIAL, 7 cycles; MOBILE 6
Freeway, 6 cycles).

a discussion on which of the 12 congestion indicators are most

appropriate to use.5
5. Congestion in the driving patterns of
three average speed models
To quantify the level of congestion that is actually incorporated
in three commonly used average speed emission models, congestion indicators were computed for MOBILE 6, EMFAC 2000 and
COPERT III. These emission models are now commonly used around
the world without knowledge of the extent to which congestion is
reected in the emission algorithms. To prevent potentially
signicant errors in emission predictions at the local scale,6 one

Full details can be found in Smit (2006).

For instance, it has been shown (EC, 1995) that the same average speed can
have a (substantial) difference in emissions of up to a factor of four, depending on
the actual level of speed variation (due to e.g. congestion) that is encountered
during the journey.
6

needs to be able to verify if the level of congestion in the particular

emission model corresponds to the level of congestion for the situation that is of interest. In fact, it is important to improve current
average speed models by including an appropriate congestion related variable such as DR (discussed later) in the modelling software, or alternatively, at least provide data on the level of
congestion in the driving patterns on which these models are
based. The information presented in this section can be used for
this purpose.
The driving pattern analysis was restricted to light-duty vehicles
due to lack of information and limited availability of speedtime
data on heavy-duty vehicles. It can be noted that there was considerable difculty in obtaining the original speedtime data that
were used in the original formulation of these models.
Fig. 2 shows scatter plots where six congestion indicators are
plotted against average speed. These indicators have been selected
from Table 1 as they demonstrate the range of possible shapes of
the relationship between mean speed and congestion variables
quite well. In this respect, Pidle and n* are comparable to unit idle

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

time (T*idle), PKE is comparable to acceleration noise (sa) and TAD is

comparable to coefcient of variation of speed (COV). CI correlates
with SRCI but should not be used because it underestimates the
level of congestion in highly congested situations, as is shown in
Appendix A. The trends in Fig. 4 have been quantied by tting
a second order regression model to the data. The relationships
between average speed and all 12 congestion indicators shown in
Table 1 are generally strong with average coefcients of determination (R2) values of 0.85 for MOBILE 6, 0.93 for EMFAC 2000
and 0.89 for COPERT III.
It was found that all MOBILE 6 and EMFAC 2000 driving patterns
could be considered as representing congested conditions according to the threshold values specied in Table 1. Most COPERT III
driving patterns included congestion, with the exception of four
driving patterns with average speeds larger than 105 km h1. This
does not mean that uncongested conditions are never present in
the driving patterns. It is possible that certain parts of the driving
pattern are in fact uncongested, though as emission factors are
based on the entire driving pattern, and not on specic parts of the
driving pattern, mean congestion levels are of interest.
Although differences in congestion level between emission
models can be quite substantial, a general trend of increasing
congestion indicator values with decreasing average speed was
observed for all models. In this respect, Fig. 2 also reveals that some
congestion indicators possess unwanted properties, namely ambiguity and undistinctiveness:
 The acceleration indicators sa, PKE and Pacc dec are ambiguous
congestion indicators because they suggest a decrease in congestion for average speeds below approximately 20 km h1.
This is in contrast with the other congestion indicators which
do show a consistent relationship with congestion, as is (partly)
shown in Fig. 2. The reduction in the computed values for speed
uctuation variables does not mean that the level of congestion
is reduced. It seems likely that acceleration indicators are reduced due to a limited ability of vehicles to accelerate freely in
very congested conditions and the increase in percent of time
spent idling, which leaves less time spent in the other driving
modes.
 The congestion indicators that reect idle operation (T*idle, n*,
Pidle) can be non-distinctive. In high speed trafc situations
(e.g. freeway driving) idling operation would only occur at
certain congestion levels, which makes these variables nondistinctive at mean speeds higher than roughly 70 km h1.
Congestion indicators that are ambiguous or non-distinctive
cannot be used independently to assess congestion levels in driving
patterns. Instead, various congestion indicators need to be considered simultaneously. The advantage of free-ow based congestion indicators (DR, SRCI) or speed uctuation congestion indicators
(TAD, COV) is that they circumvent the use of multiple congestion
indicators.
Fig. 2 shows that congestion levels may uctuate substantially
when patterns are compared individually, which indicates that,
for small changes in average speed the level of congestion may go
either up or down. This should not be the case for the MOBILE 6
patterns, which are known to represent increasing congestion
levels with decreasing average speed. Interestingly, only the
variables based on free-ow speed or (to a lesser extent) the ones
that reect speed uctuation show the expected consistent increase with decreasing average speed, as can be seen in Fig. 2.
This result lends support for the use of DR and SRCI, which
directly correspond to the denition of congestion but require
more effort, or alternatively COV and TAD, over other congestion
indicators in the quantication of congestion levels in driving
patterns.

1267

DR and SRCI indicate that MOBILE 6 uses more congested patterns for freeway driving in the average speed range of 3090 km
h1 compared to the (non-road-type specic) patterns that are
used in COPERT III and EMFAC 2000, whose congestion levels are in
fact quite similar. Outside this average speed range, congestion
levels appear to be quite similar for all three models.
This nding is, however, subject to some uncertainty. Since
driving patterns are also a function of other factors (driving style,
gender, power-to-mass ratio, etc.), it is possible that the observed
change in congestion levels in driving patterns is not (only) a result
of congestion, but (partly) a result of other factors. The extent to
which driving patterns have been affected by these non-congestion
related factors, and their relative importance with respect to congestion-related effects could not be determined. Nevertheless, it
appears likely that congestion was a major, if not the most important, factor inuencing the driving patterns on which the driving
patterns are based, when the following factors are considered: (1)
the investigated patterns used in the models are all based on
driving pattern data that were collected in urban areas where
congestion is naturally prevalent (e.g. Gammariello and Long,
1996); (2) current trafc engineering literature (e.g. TRB, 2000)
dictates that these non-congestion related factors are less important than congestion related factors; and (3) empirical evidence
(Ericsson, 2000) suggests that variation in the type of urban road
has a larger inuence on driving patterns than human factors
(gender, different drivers).
6. Conclusion
Congestion causes changes in driving patterns of individual
vehicles in a trafc stream and changes in emission levels. A congestion typology of emission models has been identied based on
the ways in which congestion has been incorporated in these
models. Given the increasing importance of congestion, it would be
appropriate to extend current emission models with algorithms
that enable the prediction of the effects of congestion on emissions.
The majority of emission models explicitly take congestion into
account in the modelling process through their use of actual driving
pattern data in the modelling process. For one family of emission
models (average speed models), the extent that congestion is incorporated in the models was not clear. This has been investigated
through obtaining and analysis of the original driving pattern data
that were used in the development of these models. A methodology
to assess the extent of congestion in driving patterns was presented
in this paper.
Assuming that congestion was the most inuential factor with
respect to the investigated driving patterns, it is concluded that,
with respect to light-duty vehicles, three average speed models
(COPERT III, MOBILE 6, EMFAC 2000) examined in this study
implicitly take varying levels of congestion into account, the level of
which depends on the average speed that is chosen. In order to get
more accurate (local) emission predictions and to achieve correct
application in particular projects, it is important to improve current
average speed models by including a congestion algorithm, or
alternatively, at least provide data on the level of congestion in the
driving patterns on which these models are based and recommendations on what applications the models are suitable for.
Acknowledgements
This work was funded, in part, by the Australian Research
Council and Queensland Transport. This support is gratefully acknowledged. The United States Environmental Protection Agency,
California Air Resources Board, Laboratory of Applied Thermodynamics (Aristotle University, Greece), TRL (UK) and EMPA (Switzerland, data from TuV-Rheinland and INFRAS provided by EMPA)

1268

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

Free-flow travel times (s)

140

51

35 29 25

60

58

355

27

120

Speed (km/h)

100
80
60
Minimum Speed Limit

40

20
0

100

200

300

400

500

600

700

800

900

1000

Elapsed Time (s)

Fig. 3. Speedtime prole graphically showing the methodology to estimate mean free-ow speed (A, stop-go-stop segment; B, approach stop segment; C, queuing behaviour
segment).

are also kindly acknowledged for providing the required driving

pattern data.
Appendix A
Quantication of free-ow speed congestion indicators
Fig. 3 shows how the method is applied to a particular driving
pattern (total length of 17.2 km, average speed of 66 km h1).
As a rst step, the driving pattern is divided into several segments (as shown by the vertical lines in Fig. 3) with segments
identied as stop-go-stop components of the pattern (A), e.g.
between signalised intersections or that include an approach
stop (B) without becoming stationary, such as a turning movement. An approach stop is dened as a decrease in speed from
free-ow cruise speed by more than 33% (Newman et al., 1992). In
addition, the drop in cruise speed in the driving segments only
applied if maximum segment speed was higher than the general
minimum legal speed limit (see below). Low speed stop-and-go
patterns (e.g. C) where the maximum speed does not exceed
20 km h1 are considered to be queuing behaviour and taken as
a one-pattern segment.
The pattern segments likely represent driving on a homogeneous section of road, where the speed limit would not change. This
assumption seems reasonable as minimum speed zone lengths are
used in practice, which may be up to 10 km long for high speed
limits.
As a second step, the general minimum legal speed limit is determined (shown as the broken line in Fig. 3). Since Fig. 3 represents European driving pattern data, this limit was set to 50 km h1
(EC, 2004). The posted speed limit on a road is assumed to approximate the free-ow speed. Drivers tend to comply, on average,
within a margin of 5 km h1 below to 13 km h1 above, the speed
limit in free-owing trafc conditions (Harkey et al., 1990; Wesseling et al., 2003). In practice, speed limits may be raised above or
lowered below the general speed limit by local authorities on the
basis of trafc engineering studies and trafc surveys. Thus, it must
be assumed that the general speed limits applied in most cases
during driving pattern data collection.
The third step assigns free-ow speeds to the pattern segments.
It is assumed that in each speed-time segment free-ow speed
will be achieved at least once as, even in quite congested conditions, there will normally be some small period of time in which
free-ow conditions are achieved. Segment free-ow speed is

approximated by the maximum instantaneous speed, provided that

its value is higher than the minimum national statutory speed limit.
This will not always hold, for instance in very congested conditions where only low-speed driving occurs (e.g. C in Fig. 3).
Free-ow speeds could range from high speed freeway driving to
low speed residential driving, though this cannot be determined
from the speedtime data. To address this issue, it is considered
that free-ow speed can never be lower than the minimum national
statutory speed limit that is used in the country where the driving
pattern data were collected.7
The difference between the estimated segment free-ow speed
and the actual instantaneous speed is indicated in grey in Fig. 3. The
upper boundary of the grey area does not represent a free-ow
speedtime prole, as it takes less travel time to traverse the segment distance in free-ow conditions. The free-ow travel times
(i.e. segment length divided by segment free-ow speed) are
therefore presented at the top of Fig. 3.
As a last step, mean free-ow speed is computed by dividing
total pattern length by the sum of free-ow travel times for each
pattern segment, which in this case results in a mean free-ow
speed of 97 km h1, as follows from 17.2 km/(640 s/3600 s h1).

Validation of free-ow speed methodology

To validate this methodology, it has been applied to a set of 9
motorway driving patterns (TNO, 2001) for which the speed limits
that applied at the time of driving pattern data collection were
known (100 or 120 km h1, depending on the pattern). For each
driving pattern, the three congestion indicators that use free-ow
speed as input (DR, CI and SRCI) were calculated using estimated
segment free-ow speeds (the predicted values) and compared to
the computations based on the known speed limits as the segment
free-ow speeds (the observed values). The results of this validation are presented in Fig. 4.
The scatter plot shows that estimated indicator values correlate
reasonably with the observations, but that some indicators perform

7
This assumption is only valid as long as driving cycles do not consist of articially constant low speed driving. Hence, a visual inspection is required for driving
cycles with maximum speeds lower than the general speed limit to verify the
validity of this assumption. It is noted that Type C emission models are commonly
based on real-world data where speed limits would apply and not on test data
related to constant steady-speed driving (Smit, 2006).

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

1.0

DR
CI
SRCI

Predictions

0.8

0.6

0.4

0.2

0.0
0.0

0.2

0.4

0.6

0.8

1.0

Observations
Fig. 4. Validation of free-ow speed estimation method based on 9 motorway driving
cycles (range-standardised x- and y-axes).

better than others. Errors increase in more congested conditions

(i.e. higher indicator values). However, direction (increase or decrease) is predicted correctly by all indicators in these conditions.
DR has the lowest root-mean-square-error and is able to predict
true values well in all conditions. This is in contrast with CI, which
provides underestimates in more congested conditions.
Further validation was not possible as no other datasets with
known speed limits were available. It is expected, however, that the
estimation method would perform better in situations where speed
limits are lower than 100 km h1 (e.g. on urban roads) because the
difference between free-ow speed and average speed is substantially lower. As a result, errors in the estimation of free-ow
speed would have a smaller impact on congestion indicator values.
The validation is therefore considered to be a worst-case assessment of method performance. This limited validation indicates that
the free-ow speed estimation procedure gives acceptable results,
but that the congestion indicators DR and SRCI are preferred.

References
Affum, J.K., Brown, A.L., 1999. Estimating urban air pollution levels from road trafc
in TRAEMS. Journal of Eastern Asia Society for Transportation Studies 3 (1), 139
150.
AGO, 2003. Australian Methodology for the Estimation of Greenhouse Gas Emissions and Sinks 2003 Energy (Transport). National Greenhouse Gas Inventory
Committee. Australian Greenhouse Ofce. 1 928040 16 8.
Ahrens, C.D., 2003. Meteorology Today An Introduction to Weather, Climate and
the Environment, seventh ed. Thomson Brooks/Cole, ISBN 0 534 39771 9.
ARRB, 1990. Calibrating SIDRA. In: Akelik, R. (Ed.), ARR Research Report No. 180.
Australian Road Research Board.
Austroads, 2000. Roadfacts 2000 An Overview of the Australian and the New
Zealand Road Systems. Austroads Incorporated, Sydney, Australia, ISBN 0 85588
484 3.
Barth, M.J., Johnston, E., Tadi, R.R., 1996. Using GPS technology to relate macroscopic
and microscopic trafc parameters. Transportation Research Record 1520, 89
96.
BTE, 2000. Urban congestion the implications for greenhouse gas emissions. In:
Cosgrove, D. (Ed.), Information Sheet 16. Bureau of Transport Economics.
Canberra, ACT, Australia.
CARB, 2002. EMFAC 2002, California Air Resources Boards Emission Inventory
Series, September. http://www.arb.ca.gov//msei/on-road/latest_version.htm.
Carr, D., Von Ehrenstein, O., Weiland, S., Wagner, C., Wellie, O., Nicolai, T., Von
Mutius, E., 2002. Modelling annual benzene, toluene, NO2, and soot concentrations on the basis of road trafc characteristics. Environmental Research
Section A 90, 111118.
DoTRS, 2001. Comparative Vehicle Emissions Study. Commonwealth Department of
Transport and Regional Services, Canberra, Australia, ISBN 0 642 45684 4.

1269

EC, 1995. ESTEEM European Scenarios on Transport-Energy-Environment for

Metropolitan Areas, Final Report. Contract No. JOE3-CT95-0016. The European
Commission, Brussels.
EC, 2004. Comparative Study of Road Trafc Rules and Corresponding Enforcement
Actions in the Member States of the European Union. Directorate General
Energy and Transport. European Commission, Brussels.
EEA, 2000. COPERT III Computer Programme to Calculate Emissions From Road
Transport Methodology and Emissions Factors (Version 2.1). In:
Ntziachristos, L., Samaras, Z., Eggleston, S., Gorissen, N., Hassel, D., Hickman, A.-J.,
Joumard, R., Rijkeboer, R., White, L., Zierock, K.-H. (Eds.), November 2000,
Technical Report No. 49. European Environment Agency, Copenhagen.
Ericsson, E., 2000. Urban driving patterns characterisation, variability and
environmental implications. PhD thesis (unpublished), Bulletin 186, Lund
University, Sweden.
Fenger, J., 1999. Urban air quality. Atmospheric Environment 33, 48774900.
Fomunung, I., Washington, S., Guensler, R., Bachman, W., 2000. Validation of the
MEASURE automobile emissions model: a statistical analysis. Journal of
Transportation and Statistics September, 6584.
Gammariello, R.T., Long, J.R., 1996. Development of Unied Correction Cycles,
Proceedings of the 6th CRC On-Road Vehicle Emissions Workshop, 18 March,
San Diego, CA, pp. 4.154.29.
Greenwood, I.D., 2003. A new approach to estimate congestion impacts for highway
evaluation effects on fuel consumption and vehicle emissions. PhD dissertation (unpublished), University of Auckland, New Zealand.
Harkey, D.L., Robertson, H.D., Davis, S.E., 1990. Assessment of current speed zoning
criteria. Transportation Research Record 1281, 4051.
INFRAS, 2007. Handbook Emission Factors for Road Transport HBEFA. http://www.
hbefa.net/.
Joumard, R., Serie, E., 1999. Modelling of Cold Start Emissions for Passenger Cars,
MEET Deliverable No. 8, COST319 Action. Report No. LTE 9931. Institut National
de Recherche sur les Transports et leur Securite (INRETS), France.
Leung, D.Y.C., Williams, D.J., 2000. Modelling of motor vehicle fuel consumption and
emissions using a power-based model. Environmental Monitoring and
Assessment 65, 2129.
Lyons, G.D., McDonald, M., Hounsell, N.B., Williams, B., Cheese, J., Radia, B., 1996.
Urban trafc management: the viability of short term congestion forecasting
using articial neural networks. Proceedings of the 24th European Transport
Forum, PTRC, 112.
Matzoros, A., 1990. Results from a model of road trafc air pollution, featuring
junction effects and vehicle operating modes. Trafc Engineering and Control
31 (1), 2435.
May, A.D., 1990. Trafc Flow Fundamentals. Prentice-Hall Inc., Englewood Cliffs, NJ,
USA, ISBN 0 13 926072 2.
Namdeo, A., Mitchell, G., Dixon, R., 2002. TEMMS: an integrated package for
modelling and mapping urban trafc emissions and air quality. Environmental
Modelling & Software 17 (2), 179190.
Negrenti, E., 1999. The corrected average speed approach in ENEAs TEE model: an
innovative solution for the evaluation of the energetic and environmental
impacts of urban transport policies. Science of the Total Environment 235,
411413.
Newman, P.W.G., Kenworthy, J.R., Lyons, T.J., 1992. The ecology of urban driving II,
driving cycles across a city: their validation and implications. Transportation
Research A 26A (3), 273290.
Oduyemi, K.O.K., Davidson, B., 1998. The impacts of road management on urban air
quality. Science of the Total Environment 218 (1), 5966.
Pierson, W.R., Schorran, D.E., Fujita, E.M., Sagebiel, J.C., Lawson, D.R., Tanner, R.L.,
1999. Assessment of nontailpipe hydrocarbon emissions from motor vehicles.
Journal of Air and Waste Management Association 49, 498519.
Pokharel, S.S., Bishop, G.A., Stedman, D.H., 2002. An on-road motor vehicle emissions inventory for Denver: an efcient alternative to modelling. Atmospheric
Environment 36, 51775184.
Schulz, D., Younglove, T., Barth, M., 2000. Statistical analysis and model validation
of automobile emissions. Journal of Transportation and Statistics September,
2938.
Skabardonis, A., Dowling, R.G., 1997. Improved speed-ow relationships for planning applications. Transportation Research Record 1572, 1823.
Smit, R., 2006. An examination of congestion in road trafc emission models and
their application to urban road networks, PhD dissertation, Grifth University,
Brisbane, Australia.
Smit, R., Bawden, K., Ormerod, R., 2002. Speciated hydrocarbon emissions from road
trafc. Proceedings of the 16th Clean Air and Environment Conference, 1922
August. CASANZ, Christchurch, New Zealand, pp. 746757.
Smit, R., Smokers, R., Rabe, E., 2007. A new modelling approach for road trafc
emissions: VERSIT. Transporation Research Part D 12, 414422.
Smit, R., Poelman, M., Schrijver, J., 2008. Improved road trafc emission inventories
by adding mean speed distributions. Atmospheric Environment 42 (5),
916926.
Taylor, M.A.P., 2000. Network modelling of the trafc, environmental and energy
effects of lower urban speed limits. Road and Transport Research 9 (4),
4857.
TNO, 2001. Emissions and Congestion Estimation of Emissions on Road Sections
and the Dutch Motorway Network. In: Gense, N.L.J., Wilmink, I.R., Van de
Burgwal, H.C. (Eds.), Report No. 01.OR.VM.0441/NG. TNO Automotive, The
Netherlands.
TRB, 2000. Highway Capacity Manual. Transportation Research Board. National
Research Council, Washington, DC, ISBN 0 309 06681 6.

1270

R. Smit et al. / Environmental Modelling & Software 23 (2008) 12621270

Turner, S.M., Lomax, T.J., Levinson, H.S., 1996. Measuring and estimating congestion
using travel time-based procedures. Transportation Research Record 1564, 11
19.
USEPA, 1997. Development of speed correction cycles. In: Carlson, T.R., Austin, T.C.
(Eds.), Sierra Research, Report No. M6.SPD.001. US Environmental Protection
Agency, Washington, DC.
USEPA, 2007. MOBILE 6 Vehicle Emission Modelling Software and Documentation.
US Environmental Protection Agency, Washington, DC. http://www.epa.gov/
otaq/m6.htm.

Wesseling, J.P., Hollander, K., Teeuwisse, S., Keuken, M.P., Gense, R., Van de
Burgwal, E., Hermans, L. Th.M., Voerman, J.W.T., Kummu, P.J., Van den Elshout, J.
H.H., 2003. Onderzoek naar Effecten van 80 km/u Maatregel voor A13 op de
Luchtkwaliteit in Overschie. TNO Report 2003/258. TNO Automotive, The
Netherlands.
Zalinger, M., Ahn, T.L., Hausberger, S., 2005. Improving an instantaneous
emission model for passenger cars, Proceedings of the 14th Symposium
Transport and Air Pollution 1, 13 June. Graz, Austria (ISBN 3902465166
167-176).