Effects of traffic signal coordination on noise and air pollutant emissions

Abstract

Traffic management solutions are increasingly called for to address problems of transport and mobility. In particular, coordinated traffic lights that create green waves along major arterials are an increasingly used strategy to reduce travel times. Although it is usually assumed that an improved traffic flow will result in lower vehicle emissions, little scientific research has been spent on the effects of synchronized traffic lights on emissions. Moreover, because changes in traffic flow do not necessarily influence travel times, noise and air quality in the same way, there is a clear need for a combined approach. This paper reports on a computational study in which a microscopic traffic simulation model (Paramics) is combined with submodels for the emission of noise (Imagine) and air pollutants (VERSIT+). Through the simulation of a range of scenarios, the model is used to investigate the influence of traffic intensity, signal coordination schemes and signal parameters on the noise, carbon dioxide, nitrogen oxides and particulate matter emissions along an arterial road equiped with a series of traffic lights. It was found that the introduction of a green wave could potentially lower the emissions of the considered air pollutants by 10%–40% in the most favorable conditions, depending on traffic flow and signal timing settings. Sound pressure levels were found to decrease by up to 1 dB(A) near the traffic signals, but to increase by up to 1.5 dB(A) in between intersections. Traffic intensity and green split were found to have the largest influence on emissions, while the cycle time did not have a significant influence on emissions.

Full text

Effects of traffic signal coordination on noise and air pollutant emissions
B. De Coensel
a,∗
, A. Can
a
, B. Degraeuwe
b
, I. De Vlieger
b
, D. Botteldooren
a
a
Ghent University, Department of Information Technology, Acoustics research group, St.-Pietersnieuwstraat 41, B-9000 Ghent, Belgium
b
Flemish Institute for Technological Research (VITO), Boeretang, B-2400 Mol, Belgium
Abstract
Traffic management solutions are increasingly called for to address problems of transport and mobility. In particular, coordinated
traffic lights that create green waves along major arterials are an increasingly used strategy to reduce travel times. Although it is
usually assumed that an improved traffic flow will result in lower vehicle emissions, little scientific research has been spent on the
effects of synchronized traffic lights on emissions. Moreover, because changes in traffic flow do not necessarily influence travel
times, noise and air quality in the same way, there is a clear need for a combined approach. This paper reports on a computational
study in which a microscopic traffic simulation model (Paramics) is combined with submodels for the emission of noise (Imagine)
and air pollutants (VERSIT+). Through the simulation of a range of scenarios, the model is used to investigate the influence of
traffic intensity, signal coordination schemes and signal parameters on the noise, carbon dioxide, nitrogen oxides and particulate
matter emissions along an arterial road equiped with a series of traffic lights. It was found that the introduction of a green wave
could potentially lower the emissions of the considered air pollutants by 10 % to 40 % in the most favorable conditions, depending
on traffic flow and signal timing settings. Sound pressure levels were found to decrease by up to 1 dB(A) near the traffic signals, but
to increase by up to 1.5 dB(A) in between intersections. Traffic intensity and green split were found to have the largest influence on
emissions, while the cycle time did not have a significant influence on emissions.
Keywords: Urban traffic, Microscopic simulation, Synchronization, Green wave, Sound level, Air pollution
1. Introduction
Next to being a major cause of stress for drivers, traffic con-
gestion causes travel delays, and thus imposes a substantional
cost on society. It is estimated that every year nearly 1 % of
the EU’s GDP is lost as a result of this phenomenon (Euro-
pean Commission, 2007). One way to moderate congestion is
to expand the road network, but in an urban area this is often
not feasible because of the presence of buildings. On the other
hand, traffic management solutions—such as introducing and
enforcing variable speed limits, installing local-express lanes
or reversible lanes, imposing differentiated road pricing or opti-
mizing traffic signal timing—try to improve the performance of
the existing infrastructure. Increasingly, new information and
communication technologies are used in the implementation of
measures, and the deployment of intelligent transport systems
is actively promoted by the European Commission (2008).
Optimization of traffic signal parameters has a long history,
starting in the late 1950’s with the work by Webster (1958) on
the timing of isolated intersections based on statistical meth-
ods. Since then, the state-of-the-art has evolved over actuated
signals, which lengthen the green period to some extent if a
queue is observed, to adaptive and cooperative methods, which
are realized using actual flow information supplied by traffic
detectors, and which involve series of intersections (Bretherton
∗
Corresponding author. Tel.: +32 9 264 9994; fax: +32 9 264 9969.
Email address: bert.decoensel@intec.ugent.be (B. De Coensel)
et al., 2004; Warberg et al., 2008; Osorio and Bierlaire, 2008).
Usually, systems are designed to create green waves along ar-
terial roads facing high demands, and a number of optimiza-
tion techniques exist in order to accomplish this strategy (e.g.,
Gartner and Stamatiadis, 2002; Cheng et al., 2006). Because
vehicular traffic flow through a network of signalized intersec-
tions represents a complex system far from equilibrium, it has
been studied extensively from a statistical mechanics point of
view, using microscopic traffic simulation models (Chowdhury
and Schadschneider, 1999; Brockfeld et al., 2001; Huang and
Huang, 2003; Nagatani, 2007, 2009; Varas et al., 2009). Most
of the optimization techniques use the average vehicle delay or
the number of stops as a measure of effectiveness.
However, there are some conflicts of interest in the selec-
tion of objectives for signal timing optimization (Li et al., 2004;
Warberg et al., 2008). For example, minimizing the delay for
vehicles along an arterial road may cause longer waiting times
for reverse-flow traffic and for crossing pedestrians; prioritiz-
ing public transport (e.g. by skipping a phase) may lower the
performance for private transport; optimizing flows may have
an impact on the safety of drivers and pedestrians (Tindale and
Hsu, 2005; Shinar et al., 2004), etc. Therefore, signal timing
optimization is considered to be a multi-objective problem. The
potential positive effects of green waves on emissions (noise
and air pollutants) are often called upon as an additional support
for their introduction. The rationale behind the claim of lower-
ing (air pollutant) emissions is that congestion causes vehicles
to function at sub-optimal speeds and accelerations, leading to
Preprint submitted to Environmental Modelling and Software February 9, 2012

incomplete combustion and additional emissions of NO
x
, CO,
etc. Although the potential of green waves to reduce travel de-
lays are widely accepted, the side-effects on vehicle emissions
(both noise and air pollutants) are however much less clear.
In this paper, the influence of traffic signal coordination on
vehicle emissions will be studied in detail. In particular, a
microscopic traffic simulation model is coupled with emission
models for noise and air pollutants (CO
2
, NO
x
and PM
10
). With
this methodology, two options are possible. The first option
is to model a limited number of scenarios in great detail (e.g.
based on an existing case study setting), thereby including the
effect of a large number of contextual factors. The second op-
tion is to extract information from the simulation of a large
number of (more simplified) scenarios. This option is chosen
in the present paper, as it hopefully leads to more general in-
sights, not tied to a particular existing context. The downside is
that the effect of only a limited number of factors can be con-
sidered, because of computational complexity.
A simplified setting consisting of an urban arterial road with
several consecutive signalized intersections will be considered,
and through the simulation of a range of scenarios, the influ-
ence of traffic demands and signal timing parameters on emis-
sions will be investigated (no air pollutant dispersion modelling
is considered). The work discussed in this paper differs from
earlier research in several aspects. Noise and air pollutants are
considered jointly, and state-of-the-art instantaneous emission
models are used for both types of emissions. These emission
models were specifically designed for use with microscopic
traffic simulation models, and have been validated extensively
on a European scale. An important feature of both models is
that results are representative for a complete vehicle fleet (in
this case the Dutch fleet), instead of representing only a lim-
ited sample of vehicle types. Additionally, the ranges of traffic
intensities and signal timing parameters are larger than those
considered in earlier studies. In Section 2, an overview is given
of previous work that considered the influence of traffic sig-
nal coordination on noise and air pollutant emissions. In Sec-
tion 3, the general methodology will be described, including
the assumptions and simplifications made. Section 4 will then
present the results for a series of scenario simulations, followed
by a discussion in Section 5.
2. Literature overview
Research on the influence of traffic light control on noise and
air pollutant emission is by no means complete; most studies
have considered the emission at a single intersection at most—
see De Coensel et al. (2007) and Pandian et al. (2009) for an
overview of literature on noise and air pollutant emissions near
traffic intersections. When the effect of traffic signal coordina-
tion is considered, usually only the emission of a single vehi-
cle is measured (using on-board equipment), or the immission
caused by all vehicles is measured at a few locations. A wide
range of methods are used, and often no details are given about
the fleet composition, which makes it difficult to compare re-
sults across studies.
Based on a review of measurements performed in the UK and
Switzerland, Desarnaulds et al. (2004) found that coordination
of traffic lights may lower the sound pressure level near inter-
sections by up to 2 dB(A). Unal et al. (2003) performed on-
board air pollutant emission measurements along a signalized
arterial road in North Carolina, US, using four different drivers
and eight gasoline fueled light-duty vehicles, before and after
the coordination of traffic signals. They found that, depending
on the type of vehicle and the level of congestion, the imple-
mentation of traffic signal coordination yielded reductions in
HC, NO and CO emissions per unit of distance between 10
and 20 %. Zhang et al. (2009) used a portable emission mea-
surement system to compare the NO
x
, HC and CO emissions
of a single vehicle, when driven along two different roads in
Bejing, China, one with and one without coordinated traffic
signals (both carried similar traffic flow and composition). It
was found that the emission of HC and CO per unit of distance
was lower along the road with coordinated signals, by resp.
50 % and 30 %, but the emission of NO
x
per unit of distance
was higher by 10 %. A detailed analysis of the driving cycles
showed that NO
x
emission increased slightly with increasing
average vehicle speed, while HC and CO emissions decreased
with increasing average vehicle speed. Subsequently, simula-
tions using emission laws extracted from measurements were
used to estimate the change in emission that would result from
removing the signal coordination along the particular road with
coordinated traffic lights. It was found that this action would re-
sult in an increase in air pollutant emissions per unit of distance
between 9 % and 14 %.
A main reason for the relative lack of scientific data on emis-
sions at intersections is that well-controlled field experiments
during which emissions are measured are quite complex and
expensive to carry out, and therefore not always feasible. On
the other hand, computational models for estimating emissions
that return realistic results for the stop-and-go behavior of ve-
hicles near intersections have become available recently, for
noise (De Coensel et al., 2005, 2007; Can et al., 2008) as well as
for air pollutants (Ahn et al., 2002; Int Panis et al., 2006; Silva
et al., 2006; Chen and Yu, 2007; Smit et al., 2008; Mensink and
Cosemans, 2008; Smit and McBroom, 2009), and these models
will become increasingly important for evaluating environmen-
tal policies and infrastructural developments.
Considering the case of noise emissions, as part of the SI-
LENCE project, simulations were carried out for a road with
three signalized intersections with 200 m and 500 m in be-
tween (B
´
erengier and Picaut, 2008). Two situations with co-
ordinated traffic lights were compared: a green wave (cars only
have to stop at the first traffic light) and a red wave (all cars
have to stop at all traffic lights). Only a single set of traffic
light parameters and a single traffic intensity (1440 vehicles/h)
were considered. Results indicated that the green wave could
lower L
Aeq
levels up to 4 dB(A) at the intersections, but could
increase levels by as much as 3 dB(A) between intersections,
due to higher average speeds.
Considering the case of air pollutant emissions, early work
was performed by Rakha et al. (2000), who used a microscopic
traffic simulation model coupled with an instantaneous emis-
2

sion model (HC, CO and NO
x
) to assess the influence of the
implementation of a green wave along an arterial road with
three equally spaced (350 m apart) traffic signals. They con-
sidered the extreme cases of all vehicles having to stop at all
signals, and the perfect green wave, and found reductions in
emissions in the range of 50 % for the latter. The emission
model (instantaneous emissions as a function of speed and ac-
celeration) was obtained through non-linear regression based
on measurement data collected for 8 light-duty vehicles under
hot stabilized conditions, and did not account for cold starts
or high emitters. Zito (2009) used a microscopic traffic sim-
ulation model (DRACULA), combined with an artificial neu-
ral network (trained on measured data at a monitoring station
along the road) to estimate roadside CO and C
6
H
6
concentra-
tions near a particular arterial road in Palermo, Italy, which con-
tains a sequence of coordinated traffic lights, spaced 80 m to
150 m apart. Using this unconventional model, quite extreme
variations in CO concentration between 0.1 and 1.5 mg/m
3
, and
in C
6
H
6
concentrations between 0.1 and 4.0 µg/m
3
were found,
depending on the settings for the (common) traffic light cycle
time and synchronization offset. Neunh
¨
auserer and Diegmann
(2010) used a microscopic traffic simulation model (VISSIM)
to extract mean speeds on a 1-minute basis, along an arterial
road in Cologne-M
¨
ulheim, Germany, containing several sig-
nalized intersections over a length of 1 km. Two scenarios,
without and with coordinated traffic signals, were considered.
Average NO
x
emissions were subsequently estimated for each
street section, using a linear combination of driving patterns fit-
ted to the simulated mean speed distributions. Depending on
the section considered, they found changes in NO
x
emissions
ranging from a decrease by 45 % to an increase by 18 %. Fi-
nally, Zallinger et al. (2010) also used the VISSIM microscopic
traffic simulation model, coupled with an instantaneous emis-
sion model (PHEM), to study the effect of signal coordination
along an existing arterial road with 12 signalized intersections
in Graz, Austria. Simulations showed that optimized signal set-
tings could reduce fuel consumption, NO
x
and PM emissions
by resp. 14 %, 19 % and 17 %. Results were compared with on-
board measurements along the actual arterial using 2 cars, and
in general a good agreement was found.
Note that the concept of calming green waves (Ellenberg
and Bedeaux, 1999) has been proposed with safety purposes
in mind, rather than minimizing travel delay. In this case, sig-
nal coordination is tuned to encourage drivers to adopt a slower
and safer, but more consistent speed by avoiding that drivers ac-
celerate excessively in order to catch up one signalization cycle
between two intersections. Based on a reduction of the aver-
age speed by 10 to 15 km/h, a noise reduction of about 3 dB(A)
could be expected (Ellebjerg, 2007; B
´
erengier, 2009). How-
ever, one concern about this type of signals is that they not
only stop traffic that exceeds the speed limit, but also traffic that
is not; experimental data suggests that the fraction of unfairly
stopped vehicles may be as high as 30 %, leading to increases of
air pollutants (CO, NO, HC) between 10 % and 40 % (Coelho
et al., 2005b). On the other hand, if the speed control traffic
signals modify the behavior of the drivers by inducing a speed
reduction, they will also result in a decrease in relative pollutant
emissions (Coelho et al., 2005a; Barkenbus, 2010). Neverthe-
less, the effects of calming green waves will not be studied in
this work.
3. Methodology
Basically, the study of the potential benefits or drawbacks of
traffic signal coordination on emissions can be decomposed into
an emission related and a traffic related subproblem. The for-
mer problem handles about what can be expected in terms of
reduction or increase in emissions, depending on factors such
as the traffic intensity or traffic signal parameters, but given that
the traffic signal coordination works as expected. The latter
problem handles about finding those conditions in which coor-
dinated signals are effective in creating a green wave. In this
paper, we will mainly focus on the first subproblem. The sec-
ond subproblem is purely traffic related and has been studied
extensively, as already mentioned in the introduction; we will
only briefly review some important aspects in the discussion in
Section 5.
3.1. Microscopic traffic simulation
The aim of the virtual experiment described in this paper
is to investigate the influence of signal parameters and traffic
intensity on vehicle emissions along a typical urban arterial
road with coordinated traffic signals. Traffic simulation mod-
els that aim to be accurate in the vicinity of interrupted traf-
fic flows have to model the temporal and spatial behavior of
vehicle speeds and accelerations. Microscopic traffic models
incorporate these dynamic effects by modelling vehicles indi-
vidually, and during recent years they have been used success-
fully as part of traffic noise and air quality prediction models,
mostly for scholarly purposes (De Coensel et al., 2005, 2009;
Botteldooren et al., 2006; Can et al., 2009; Della Ragione et al.,
2009; Zallinger et al., 2010; Madireddy et al., 2011). A traf-
fic network, consisting of an arterial road with signalized inter-
sections, was constructed using Quadstone Paramics (Fritzsche,
1994; Cameron and Duncan, 1996); a schematic view of the
setting is shown in Figure 1(left). Note that, because the set-
ting described in this paper is relatively simple, it is expected
that the use of a different microscopic traffic simulation model
would lead to the same results.
Because it is not feasible to simulate all conceivable config-
urations of intersections, a number of assumptions and simpli-
fications had to be made. A one-way arterial road with a single
lane is considered. As a consequence, the influence of cross-
flow and reverse-flow traffic, lane changing and overtaking on
emissions is neglected. This may lead to an overestimation of
the effects; we will discuss the influence of reverse-flow traffic
more in detail in Section 5. Not accounting for reverse-flow
traffic can be justified given the typical application of signal
coordination on urban roads during morning (into the city) or
evening (out of the city) rush hour, i.e. periods during which
the traffic flow is assumed to be dominant in a single direction.
For the same reason, not accounting for cross-flow traffic can be
somewhat justified, at least when arterial roads not too close to
3

Time
5
4
3
2
1
d
1
d
2
d
3
d
5
td
4
green
red
x-2L -L 0 L 2L
12345
Section of interest
Traffic demand D
Figure 1: Schematic view of the simulated arterial road (left), together with an example of unsynchronized traffic light timing (right).
the urban core are considered. The modelled arterial road has a
traffic demand D [vehicles/h] and a speed limit v
max
= 50 km/h,
which is typical for urban roads in Europe. Five traffic sig-
nals are located at regular distances L = 200 m from each other.
This distance was chosen to be realistic for urban situations,
and results in only minimal platoon dispersion, occurring when
vehicles leave a signalized intersection. Additionally, given the
urban rush hour context, only a single light duty vehicle type is
used for the traffic simulation. Nevertheless, emission calcula-
tions are representative for a complete vehicle fleet of light duty
vehicles (see Sections 3.3 and 3.4).
Vehicle behavior parameters, such as the aggression, aware-
ness and reaction time distributions of the drivers, the queue
gap distance, the mean target headway between a vehicle and
the car in front, the distance at which signposting is perceived
etc., were not varied in this study. In order to adhere to
a generic methodology, default Paramics values were used,
which are already carefully selected based on extensive mea-
surements (Fritzsche, 1994). Although these parameters have
an influence on traffic dynamics (Helbing, 2001; Knospe et al.,
2002), they are highly dependent on specific configurations of
infrastructure and even on social and cultural aspects. These
parameters could be adjusted for specific cases to which the
methodology described in this paper is applied.
The simulation time considered, noted T
sim
, was 1 hour, with
a simulation timestep ∆t = 0.2 s. However, the actual simula-
tions included two additional 5-minute periods: one at the be-
ginning, for traffic build up, and one at the end, to make sure all
vehicles can complete their trip. Actual simulated traffic flows
Q [vehicles/h] and travel times are calculated on the basis of the
trips that are started during the considered 1-hour period. Ve-
hicles are loaded onto the network at a distance of 500 m from
the first traffic signal (western side in Figure 1), randomly dis-
tributed in time according to a negative exponential distribution.
Because of the statistical nature of microsimulation, results dif-
fer between runs of the model: the actual simulated traffic flow
Q will, in each particular simulation run, be near the demanded
value D, but will not necessarily be exactly the same. In partic-
ular, when the demand is higher than the capacity of the road,
the actual flow will reach a saturation value.
3.2. Traffic signal timing
The main parameters controlling the operation of a single,
isolated signalized intersection are the cycle time τ [s], the
green split α
i
for the different approaches i of the intersection,
and the offset δ [s]. The cycle time is defined as the sum of the
durations of all distinct phases of the signalized intersection,
where a phase corresponds to a particular state of the traffic
lights. In other words, it is the time it takes for a traffic signal
to get from the start of the green light through amber and red to
the start of the next green light. For isolated signalized intersec-
tions, a short cycle time will generally lead to lower average de-
lays (vehicles never have to wait long for a green light), but will
lower the overall capacity of the intersection (Webster, 1958).
Upper and lower bounds for the cycle time are also set by safety
concerns: if the cycle time is too long, drivers may start to ig-
nore the red phase; if it is too short, there is an increased risk of
collisions (Warberg et al., 2008). The green split for the i’th ap-
proach of a single signalized intersection is defined as the ratio
between the amount of green time γ
i
[s] and the cycle time, i.e.
α
i
= γ
i
/τ. The green split is usually divided among the different
approaches of an intersection according to the traffic intensity
expected for each approach; arterial roads are given the larger
split. The offset of a signalized intersection is defined as the dif-
ference in time between the start of a cycle of this intersection
and the start of a cycle of some reference intersection. The off-
set is used to provide signal coordination between consecutive
intersections for vehicles travelling along an arterial road. Next
to these main parameters, the operation of a signalized inter-
section is also characterized by the order of the phases during
a cycle (called the phase sequence). The influence of the phase
sequence was not taken into account in this work, because the
intersections that are considered only have a single approach.
When an isolated intersection is considered, the cycle time
and green split are usually optimized in order to minimize the
average vehicle delay; the method developed by Webster (1958)
is widely used for this. However, the cycle time and green split
of the intersections considered in this work were not optimized,
because the optimal settings depend on the amount of cross-
flow traffic, which is not taken into account in this work. Rather,
a range of settings is considered, in order to make the results
applicable to a range of cross-flow traffic intensities.
When a series of intersections is considered, as in the present
study, signal coordination is usually accomplished through the
use of a common cycle time (which can be time-dependent).
Both pretimed and adaptive signals can be used to install green
waves (Warberg et al., 2008). The former use static plans for
the signal parameters and offsets, according to the time of day.
The latter are set to work together and periodically adjust their
settings in order to minimize travel delays, based on detection
and short-term prediction of traffic intensities. It turns out that,
in order to avoid transient side-effects such as malfunctioning
green waves, changes to the signal settings should be made in
small steps only (Bretherton et al., 2004). It is assumed that
4

Table 1: Parameter ranges for the simulation scenarios.
Parameter Range Levels
τ 30, 40. . . 90 s 7
α 0.5, 0.6, 0.7, 0.8 4
D 50, 100. . . 2000 vehicles/h 40
Signal scheme green, red, desynchronized 3
these changes are small enough to not have an influence on
emissions, so it is sufficient to consider the stationary state in
this work. The traffic signals at all intersections are considered
to operate with a common cycle time τ and green split α for
the arterial road (west to east), and with offsets δ
i
, 1 ≤ i ≤ 5.
To further simplify the discussion, no amber time is considered,
i.e. the green time for each intersection is ατ, and the red time is
(1−α)τ. An example of the signal settings of the five signalized
intersections is given in Figure 1(right).
A series of scenarios was created by varying the parameters
τ, α and D; an overview of the parameter ranges can be found in
Table 1. Next to this, three different traffic signal coordination
schemes were considered, labeled green, red and desynchro-
nized. In the first scheme, the offsets are set to create a green
wave, i.e. δ
i
= 3.6L/v
max
·i. Vehicles will only have to stop at the
first traffic light, with probability 1 − α. Note that this scheme
represents the theoretical extreme; in practice a perfect green
wave will be hard to achieve (see Section 5 for a discussion).
In the second scheme, the offsets are set to create a red wave,
i.e. δ
i
= (3.6L/v
max
− ατ) · i. Vehicles will have to stop at every
traffic light. In the third scheme, the offsets are set randomly,
and in order to desynchronize the signals, a small but random
number of seconds (< 2 s) is added to or subtracted from the
cycle times. This way, a wide range of waiting times and queue
lengths at each intersection is encountered over the course of a
simulation run; the chance that a vehicle will have to stop at a
single traffic light will again roughly be 1 − α. The results for
the desynchronized scheme thus represent the average over the
results for all possible schemes in which there is no signal co-
ordination. Finally, the total number of unique traffic scenarios
is equal to 7 × 4 × 40 × 3 = 3360.
3.3. Noise emission modelling
The output of a microscopic traffic simulation run consists of
the instantaneous position, speed and acceleration of each ve-
hicle at each timestep. Subsequently, the instantaneous noise
emission of each vehicle in the simulation is calculated using
the Imagine road traffic noise emission model (Peeters and van
Blokland, 2007). This model was specifically developed with
microscopic traffic simulation in mind, and has been validated
widely on a European scale, using measurements on a large
number of vehicles, driven on a wide range of road surface
types. The model forms the basis for a potential future Eu-
ropean standard for road traffic noise prediction, and was cali-
brated to generate the average noise emission of the European
vehicle fleet. More in particular, while there may be differences
between different types of vehicles in terms of noise emission,
the model will provide a good estimation of traffic sound levels,
when results are aggregated over a sufficiently large number of
vehicles sampled from the European fleet. Regional differences
can be accounted for through corrections on the reference road
surface (e.g. the use of porous asphalt) and on standard vehi-
cle fleet characteristics (e.g. higher fraction of diesel vehicles).
Reference values were used in this work, which makes the re-
sults valid for the light duty vehicles of the Dutch vehicle fleet.
The light duty vehicle type considered in this work corresponds
to the Imagine emission category 1.
The Imagine model produces instantaneous point source
sound power levels, in which two sources of noise are consid-
ered separately: rolling noise (generated by tire-road interaction
and aerodynamic drag) and propulsion noise (generated by the
powertrain and the exhaust). Both contributions are resp. given
by the following formulae (Peeters and van Blokland, 2007):
L
W,R
(v, a) = A
R
+ B
R
· log
10
v
v
re f
!
, (1)
L
W,P
(v, a) = A
P
+ B
P
·
v − v
re f
v
re f
!
+ C
P
· a, (2)
where v is the vehicle speed (in km/h) with v
re f
= 70 km/h,
and a is the vehicle acceleration (in m/s
2
). For values of the
coefficients A
R
, B
R
, A
P
, B
P
and C
P
, we refer to Peeters and
van Blokland (2007). Both contributions are then aggregated
to obtain the total sound power level, in dB(A), produced by a
vehicle:
L
W
(v, a) = 10 log
10

10
L
W,R
(v,a)/10
+ 10
L
W,P
(v,a)/10

. (3)
Note that instantaneous emissions are calculated on a 1/3-
octave band basis; however, in this work we will only consider
the A-weighted value.
When the noise emission of a vehicle trip through the net-
work is considered, we define the total sound power level L
tot
W
of the particular vehicle over the course of its trip:
L
tot
W
= 10 log
10







∆t
1 s
X
t
10
L
W
(t)/10







, (4)
in which the summation is calculated over all simulation
timesteps that the vehicle needs to complete its trip (i.e. T/∆t
values, with T the duration of the trip, in seconds). Trip results
averaged over all simulated vehicles are noted hL
tot
W
i. This quan-
tity relates directly to the sound power level used for noise map-
ping purposes, and as such relates to time-averaged immission
levels. In particular, the hourly averaged A-weighted sound
power level emitted by the simulated road segment equals
hL
tot
W
i + 10 log
10
(Q ).
Because of the local character of sound, it may also be use-
ful to consider the (hourly) equivalent continuous (A-weighted)
sound pressure level L
Aeq
at a number of locations along the
simulated road segment. In particular, we will consider two
receiver locations, one in the middle between traffic lights 3
and 4 (x = L/2) and one near traffic light 4 (x = L), both
at a standardized height (1.5 m) and distance from the road
(7.5 m). The hourly L
Aeq
is derived from the time series of
5

instantaneous sound pressure levels (which in turn is caused
by the instanteneous sound power level of all vehicles on the
road segment), and is calculated assuming free field propaga-
tion conditions and only considering geometric divergence. The
L
Aeq
at both locations will strongly depend on the traffic inten-
sity; for further analysis, we therefore also define the average
contribution to the sound pressure level by a single vehicle as
hL
Aeq
i = L
Aeq
−10 log
10
(Q ). Note that differences in hL
Aeq
i will
also reflect differences in L
Aeq
at the same traffic flow.
Finally, it has to be noted that, given the urban rush hour con-
text, only light duty vehicles are considered in this study, as al-
ready mentioned in Section 3.1. For heavy duty vehicles, accel-
eration has a more pronounced effect on noise emission, which
is not the case for light duty vehicles, due to the engine noise
being more controlled. Consequently, if the vehicle fleet com-
position consists of a significant amount of heavy duty vehicles,
higher noise levels could be expected near the intersections for
the red wave scheme, which would increase the beneficial effect
of a green wave on noise levels near the intersections.
3.4. Air pollutant emission modelling
The instantaneous CO
2
, NO
x
and PM
10
emission of each ve-
hicle in the simulation is calculated using the VERSIT+
micro
vehicle exhaust emission model. The VERSIT+ model, devel-
oped by TNO (Smit et al., 2005, 2007), is based on more than
12,500 measurements on vehicles of a wide range of makes and
models, fuel types, Euro class, fuel injection technology, types
of transmission etc. The model uses multivariate regression
techniques to determine emission factor values for different ve-
hicle classes. As the model requires actual driving pattern data
as input, it is fully capable of modelling the effects of conges-
tion on emission. Furthermore, it takes into account detailed ef-
fects such as cold started engines; the emission factors for PM
10
also account for non-exhaust emissions, the wearing of tires and
brakes etc. A derived model, VERSIT+
micro
, was recently de-
veloped by TNO (Ligterink and De Lange, 2009), specifically
targeted at a coupling with microscopic traffic simulation mod-
els. For this, emission parameters of different vehicles (with
varying fuel type) were aggregated into a prototypical vehi-
cle emission model representing the average emission of the
Dutch vehicle fleet. This procedure is similar to the one used
in the construction of the Imagine noise emission model, and as
such, the model will correctly predict measurement results ag-
gregated over a sufficiently large number of vehicles sampled
from the Dutch fleet. In this work, the VERSIT+
micro
light duty
vehicle class representing the light duty vehicles within the fleet
in Dutch urban environments during the year 2009 was used.
The VERSIT+
micro
model produces instantaneous vehicle
emissions in g/s, on the basis of instantaneous speed v and ac-
celeration a, extracted from the microscopic traffic simulation
runs. First, a dynamic variable w is defined as (Ligterink and
De Lange, 2009):
w = a + 0.014 · v. (5)
For constant w, emissions were found to vary only slowly with
speed. Hence, the remaining dependence on v was set in three
categories, roughly corresponding to urban, rural and freeway
driving (idling is considered apart). Subsequently, three dy-
namic domains (stationary, dynamic and aggressive) are set for
each speed category. Finally, the emissions e in g/s are given by
the following set of piecewise linear equations (Ligterink and
De Lange, 2009):
e =



















u
0
(v < 5, a < 0.5),
u
1
+ u
2
|w|
+
+ u
3
|w − 1|
+
(v < 50),
u
4
+ u
5
|w|
+
+ u
6
|w − 1|
+
(50 < v < 80),
u
7
+ u
8
|w −
1
2
|
+
+ u
9
|w −
3
2
|
+
(v > 80),
(6)
where the function |x|
+
yields 0 for x < 0, and x otherwise.
The first line in Eq. 6 models the air pollutant emissions dur-
ing idling. The 10 coefficients u
i
in each of the regions of the
speed-acceleration space were, for each air pollutant type, de-
termined through a maximum likelihood method. Similar to
the case of noise emission, we will note the total CO
2
, NO
x
and
PM
10
emission (in g) of a vehicle trip through the network as
CO
tot
2
, NO
tot
x
and PM
tot
10
, and the trip results averaged over all
simulated vehicles as hCO
tot
2
i, hNO
tot
x
i and hPM
tot
10
i. To get the
hourly emitted amount of air pollutants on the simulated road
segment, one has to multiply the latter quantities by Q.
4. Simulation results
4.1. Road capacity and travel times
One of the principal questions to ask when determining if
signal coordination would be advisable, is to see what trade-offs
exist between road capacity, travel times and vehicle emissions,
as this would enable traffic engineers to devise suitable com-
promise solutions. Therefore, we first present the influence of
signal coordination on road capacity and travel times. Figure 2
shows the traffic flow Q as a function of traffic demand D. The
capacity of the road (including all signalized intersections) can
be defined as the limit value of Q for high D. It can be seen
that, for most part, capacity depends on the green split, and
that installing the green wave slightly increases the road capac-
ity, by about 6 %. This is somewhat lower than what is found
empirically (e.g., Lum et al. (1998) found an increase of about
8.4 %), which is probably due to the simple layout and limited
size of the simulated network. Because of the small influence
of the cycle time, the results shown in Figure 2 are averaged
over the different cycle times for the purpose of clarity. It was
found that there is no significant difference between the desyn-
chronized scheme and the red wave scheme (p > 0.1). This
means that implementing the red wave scheme will not lower
the capacity significantly as compared to the case without co-
ordination, even if this scheme results in more stops. The fact
that having a red wave does not lower capacity as compared to
the case without signal coordination can be easily understood
when one takes into account that road capacity is determined
by the main bottleneck: when there is no signal coordination,
one can assume that at least one of the traffic lights will gener-
ate a bottleneck, where most vehicles have to stop because of
bad timing between two consecutive lights. More in particu-
lar, this result suggests that signal coordination in one direction
6

0
200
400
600
800
1000
1200
1400
1600
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Traffic demand [vehicles/h]D
Traffic flow [vehicles/h]Q
green wave
no synchronization
red wave
0.5
0.6
0.7
0.8
a =
Figure 2: Traffic flow, as a function of traffic demand, for various signal coor-
dination schemes and green split. Results are averaged over the different cycle
times.
will not have an adverse effect on road capacity in the oppo-
site direction, at least for the theoretical situations considered
in this work. It has to be noted that the scenarios considered do
not contain any downstream obstructions; as such, the capacity
of the simulated road is mainly determined by the traffic signal
timings (which are the same for all traffic signals, except for the
offsets). Therefore, effects of (over)saturation are mainly visi-
ble in front of the first traffic signal; further down the simulated
road, traffic is always in undersaturated conditions.
When travel times for vehicles crossing the whole network
are considered, results will also be partly determined by the be-
havior of vehicles in front of the first traffic signal, for which the
signal coordination scheme does not make a difference. In order
to assess the influence of the different coordination schemes,
we will therefore focus, for the remainder of this work, on a
particular section of interest, from stopline to stopline between
the third and fourth traffic signal (see Figure 1). This section
has a length of 200 m and contains an acceleration, a cruising
and a queueing zone. Results for this section of interest will
reflect the influence of the green wave normalized to a single
traffic light. At the given speed limit of 50 km/h, it takes at least
14.4 s for a vehicle to cross this section. In Figure 3, average
travel times for the section of interest are given as a function
of α and τ, aggregated over the different actual traffic flows Q.
For the green wave scheme, travel times for the section of in-
terest were found to be independent of α, τ and Q, as long as
the road is not saturated, as can be expected (therefore, results
for the green wave scheme are averaged over α in Figure 3).
For the other schemes, travel times increase with increasing cy-
cle time and with decreasing green split. The values for the red
wave scheme can be easily understood; e.g. in case τ = 30 s and
α = 0.5, vehicles have to stop for 15 s on average, resulting in
an average travel time of 29.4 s for the section of interest.
A stepwise multiple linear regression analysis was performed
with travel time as dependent variable, and α, τ and Q as inde-
pendent variables, for the desynchronized scheme. It was found
0
10
20
30
40
50
60
30 40 50 60 70 80 90
Cycle time [s]t
Travel time for section of interest [s]
green wave
no synchronization
red wave
0.8
0.7
0.6
0.5
a
0.8
0.7
0.6
0.5
=
Figure 3: Average travel time for the section of interest, for various signal co-
ordination schemes, as a function of cycle time and green split. Results are
averaged over the different traffic flows, and for the green wave scheme also
over the green split.
that the green split explains about 37.2 % of variance in travel
time, the traffic flow an additional 22.1 %, and the cycle time an
additional 10.8 % (r
2
= 0.70). Given the average travel time of
14.6 s over the section of interest in the green wave scheme, one
gets the following regression formula for the effect of installing
a green wave on the average travel time hT
s
i, per traffic signal:
∆hT
s
i = −24.1 + 41.3 · α − 10.3 ·
τ
100
− 16.6 ·
Q
2000
. (7)
Note that this formula is only valid for the parameter ranges as
given in Table 1, and for a limit speed of 50 km/h and a dis-
tance of 200 m between traffic lights. However, in first approx-
imation, Eq. 7 can also be used for larger distances between
traffic lights, if one assumes that vehicles drive at the free flow
speed over the additional distance, and that this free flow speed
is independent of the signal coordination scheme.
4.2. Noise emission
Figure 4 shows the total emitted sound power level, aver-
aged over all simulated vehicles, and only considering the sec-
tion of interest, as a function of traffic flow and signalization
scheme. It can be seen that introducing traffic signal coordina-
tion will increase the total noise emission in all cases except for
very low traffic flows, and this increase will be larger for high
traffic flows, up to a value of 0.6 dB(A). The implementation
of a green wave will reduce the number of vehicles decelerat-
ing/accelerating near the traffic signals, but will also increase
the average vehicle speed; from Figure 4 one can conclude
that the decrease in noise emission caused by the former ef-
fect is more than compensated by the increase in noise emission
caused by the latter effect. Note that for heavy duty vehicles, the
engine, drive train, and exhaust noise are more important com-
ponents of the total noise emission. Thus, when traffic consists
of a relatively large fraction of heavy vehicles (not considered
in this work), it can be expected that introducing traffic sig-
nal coordination will lead to a lower increase of the total noise
emission. Finally, the difference between the two extreme cases
7

106.5
107.0
107.5
108.0
108.5
109.0
16001400120010008006004002000
Traffic flow [vehicles/h]Q
áñL
W
tot
for section of interest [dB(A)]
green wave
no synchronization
red wave
a = 0.5
0.6
0.7
0.8
Figure 4: Average total sound power level emitted per vehicle for the section
of interest, as a function of traffic flow, for various signal coordination schemes
and green split α. Results are averaged over the different cycle times, and for
the green and red wave schemes also over the green split.
of a green wave and a red wave can be up to 1.2 dB(A) for high
traffic flows.
A stepwise multiple linear regression analysis was performed
with hL
tot
W
i as dependent variable, and α, τ and Q as indepen-
dent variables, for the green wave scheme. It was found that the
traffic flow Q explains about 83.2 % of variance, and the green
split α an additional 4.4 % (r
2
= 0.88). The cycle time τ did
not have a significant influence on hL
tot
W
i in this scheme. The
same analysis was performed for the scheme without synchro-
nization: Q explains about 72.5 % of variance in this scheme,
and α an additional 6.1 % (r
2
= 0.79). Again, τ did not have
a significant influence on hL
tot
W
i. By subtracting both regression
equations, one gets the following formula for the effect of in-
stalling a green wave on the average total sound power level
emitted per vehicle, in dB(A):
∆hL
tot
W
i = 0.25 − 0.59 · α + 1.19 ·
Q
2000
. (8)
The same restrictions as for Eq. 7 have to be taken into account
when using this and subsequent regression equations. Addition-
ally, these results are only calculated for the considered vehicle
fleet (Sections 3.3 and 3.4). Note that Eq. 8 also equals the in-
crease in hourly averaged A-weighted sound power level emit-
ted on the road segment associated to the section of interest,
because the additional term 10 log
10
(Q ) vanishes.
The analysis in the previous paragraphs considered the total
noise emission over the section of interest. However, the effect
of a green wave on the sound pressure level may vary depending
on the measurement location (B
´
erengier and Picaut, 2008). Fig-
ures 5 and 6 show the average contribution to the L
Aeq
by a sin-
gle vehicle, resp. in between traffic lights and near a traffic light,
at a distance of 7.5 m from the road. It is found that the imple-
mentation of a green wave will result in a decrease of hL
Aeq
i
(and consequently of L
Aeq
) by up to 1 dB(A) near the signalized
intersections (1.5 dB(A) if compared to the red wave scheme),
but will result in an increase of L
Aeq
by up to 1.5 dB(A) between
intersections (2 dB(A) if compared to the red wave scheme),
because of the higher average vehicle speeds. When one takes
31.0
31.5
32.0
32.5
33.0
33.5
34.0
16001400120010008006004002000
Traffic flow [vehicles/h]Q
áñLxL
Aeq
at = /2 [dB(A)]
green wave
no synchronization
red wave
a = 0.5
0.6
0.7
0.8
Figure 5: Average contribution of a single vehicle to the L
Aeq
, at location x =
L/2 (in between traffic lights 3 and 4), as a function of traffic flow, for various
signal coordination schemes and green split α. Results are averaged over the
different cycle times, and for the green and red wave schemes also over the
green split.
into account the fact that the absolute value of the effect will
be larger when the microphone is placed closer to the road,
these extremes are roughly in accordance with empirical results
found in literature (Desarnaulds et al., 2004). From Figures 5
and 6 it also follows that a green wave will have the least dete-
riorating effect on noise levels when traffic intensities are low.
For higher intensities, the decrease in level near the signalized
intersections will be somewhat less, while in between intersec-
tions, the increase in level will clearly be higher.
Although the results in Figures 5 and 6 are qualitatively com-
parable to the simulation results reported by B
´
erengier and Pi-
caut (2008) (4 dB(A) decrease near intersections, 3 dB(A) in-
crease in between intersections as compared to the red wave
scheme), the absolute values are somewhat smaller. In order
to make a direct comparison possible, two additional scenario
simulations were carried out for a network that reproduces the
setting as used by B
´
erengier and Picaut (2008). In particular,
the same distances between intersections, the same green wave
and red wave traffic light timings, and the same traffic flow were
used (see Section 2 for more details). A decrease by 1.1 dB(A)
was found near the intersections, while a maximal increase by
1.7 dB(A) was found in between intersections. A closer look
at the results showed that for a large part, these differences can
be linked to the LCPC vehicle noise emission model (David,
2000) used by B
´
erengier and Picaut (2008). Only a single ve-
hicle type was considered (Renault Clio 1.4l), as compared to a
whole vehicle fleet in the present paper. More importantly, the
LCPC model pronounces much more the effect of acceleration
on emission, as compared to the Imagine model, which leads
to higher noise levels near the intersections in the red wave
scheme.
4.3. Air pollutant emission
Figures 7, 8 and 9 resp. show the average amount of CO
2
,
NO
x
and PM
10
that vehicles emit while travelling over the sec-
tion of interest, as a function of traffic flow, for various sig-
nal coordination schemes and green split α. It was chosen to
8

33.0
33.5
34.0
34.5
35.0
35.5
36.0
16001400120010008006004002000
Traffic flow [vehicles/h]Q
áñLxL
Aeq
at = [dB(A)]
green wave
no synchronization
red wave
a = 0.5
0.6
0.7
0.8
Figure 6: Average contribution of a single vehicle to the L
Aeq
, at location x = L
(near traffic light 4), as a function of traffic flow, for various signal coordination
schemes and green split α. Results are averaged over the different cycle times,
and for the green and red wave schemes also over the green split.
present the results in g rather than in g/km or g/s, such that the
figures represent the absolute effect per intersection. In order to
get the average emissions in g/km, or to get the total air pollu-
tant emission per km and per hour, one has to multiply the val-
ues in these figures by resp. a factor 5 or 5Q. One can see that,
in contrast to the case of noise emissions, all types of air pollu-
tant emissions decrease when a green wave is installed. This is
in accordance with most results reported in Section 2, except for
the case of NO
x
emissions, for which reported measurement re-
sults were less clear and a potential increase was found in some
situations. This may be due to the fact that NO
x
emissions are
more related to cilinder head temperature than to the occurrence
of incomplete combustion, and thus are less correlated with av-
erage vehicle speed than the other types of emissions. In the
present simulations, because instantaneous acceleration has a
large influence on air pollutant emission, a potential increase
of emissions caused by the increase in average vehicle speed
is more than compensated by the smoother traffic flow result-
ing from the coordinated traffic signals. Irrespective of the type
of air pollutant, the difference in average emission between the
desynchronized scheme and the red wave scheme reduces to
zero for traffic flows close to capacity. This is caused by the in-
fluence of idling vehicles in the queue in front of a traffic light:
while idling vehicles still emit a considerable amount of noise,
the fraction of total air pollutant emission caused by idling ve-
hicles is relatively small.
A stepwise multiple linear regression analysis was per-
formed, resp. with hCO
tot
2
i, hNO
tot
x
i and hPM
tot
10
i as dependent
variables, and α, τ and Q as independent variables, for the
desynchronized scheme. It was found that the green split α ex-
plains resp. about 41.0 %, 54.1 % and 25.4 % of variance, and
that the traffic flow Q resp. explains 25.2 %, 11.5 % and 40.8 %
of variance additionally (r
2
= 0.66 for all pollutant types). Irre-
spective of the type of air pollutant, the cycle time τ did not have
a significant influence on emissions. The following regression
equations are found for the average air pollutant emission per
vehicle for the section of interest, in g, for the desynchronized
0
10
20
30
40
50
60
70
80
90
100
16001400120010008006004002000
Traffic flow [vehicles/h]Q
áñCO
2
tot
for section of interest [g]
green wave
no synchronization
red wave
a = 0.5
0.6
0.7
0.8
Figure 7: Average CO
2
emission per vehicle for the section of interest, as a
function of traffic flow, for various signal coordination schemes and green split
α. Results are averaged over the different cycle times, and for the green and red
wave schemes also over the green split.
0.00
0.05
0.10
0.15
0.20
0.25
16001400120010008006004002000
Traffic flow [vehicles/h]Q
áñNO
X
tot
for section of interest [g]
green wave
no synchronization
red wave
a = 0.5
0.6
0.7
0.8
Figure 8: Average NO
x
emission per vehicle for the section of interest, as a
function of traffic flow, for various signal coordination schemes and green split
α. Results are averaged over the different cycle times, and for the green and red
wave schemes also over the green split.
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
16001400120010008006004002000
Traffic flow [vehicles/h]Q
áñPM
10
tot
for section of interest [g]
green wave
no synchronization
red wave
a = 0.5
0.6
0.7
0.8
Figure 9: Average PM
10
emission per vehicle for the section of interest, as a
function of traffic flow, for various signal coordination schemes and green split
α. Results are averaged over the different cycle times, and for the green and red
wave schemes also over the green split.
9

(subscript d) scheme:
hCO
tot
2
i
d
= 102.7 − 70.1 · α + 28.4 ·
Q
2000
, (9)
hNO
tot
x
i
d
= 0.241 − 0.156 · α + 0.040 ·
Q
2000
, (10)
hPM
tot
10
i
d
= 0.0173 − 0.0126 · α + 0.0074 ·
Q
2000
. (11)
A similar regression analysis was performed for the average
emission per vehicle in the green wave scheme; it was found
that neither of the variables Q, α or τ had a significant influ-
ence on emissions, as can also be seen in Figures 7–9. The
average values for hCO
tot
2
i
g
, hNO
tot
x
i
g
and hPM
tot
10
i
g
are resp.
40.2 g, 0.095 g and 0.0073 g for the section of interest of 200 m
length. A similar regression analysis was also performed for
the average emission per vehicle in the red wave scheme; it was
found that only the traffic flow Q had a significant influence on
emissions, explaining 72.5 %, 90.6 % and 89.3 % of variance
for resp. CO
2
, NO
x
and PM
10
emissions. The following regres-
sion equations are found for the average air pollutant emission
per vehicle for the section of interest, in g, for the red wave
(subscript r) scheme:
hCO
tot
2
i
r
= 95.5 − 25.9 ·
Q
2000
, (12)
hNO
tot
x
i
r
= 0.225 − 0.089 ·
Q
2000
, (13)
hPM
tot
10
i
r
= 0.0137 + 0.0134 ·
Q
2000
− 0.0199 ·

Q
2000

2
. (14)
Finally, one may calculate the effect of installing a green wave
on the average air pollutant emission per vehicle and per traffic
signal, in g, as ∆hCO
tot
2
i = hCO
tot
2
i
g
− hCO
tot
2
i
d
, and similar for
the other pollutants.
From Figures 7–9, one can estimate the reduction in percent-
ages caused by the implementation of a green wave, although
these estimates are strictly speaking only valid for a distance of
200 m between traffic signals. It is found that reductions vary
between 10 % for low traffic flows and high green split, and
40 % for traffic flows near capacity and low green split, which
is in accordance with the ranges reported in literature (see Sec-
tion 2).
5. Discussion
The simulation results presented in Section 4 consider the ex-
treme cases of the perfect green wave, in which all vehicles are
able to traverse the simulated road segment without having to
stop, and the perfect red wave, in which all vehicles have to stop
at all traffic lights, together with the situation without signal co-
ordination. As such, this study focused on the limits of what
can be expected by introducing signal coordination. However,
in practice, introducing signal coordination will almost never
result in a perfect green wave, and a wide range of literature ex-
ists that investigates the conditions for which coordinated sig-
nals are effective in creating a green wave (see the references
in Section 1). Examples of important factors to consider, and
which could lead to the green wave (partly) breaking down, are
the amount of congestion, the amount of traffic entering from
sidestreets, the distance between signalized intersections, the
presence of slow or heavy vehicles (i.e. a distribution of target
speeds), the presence of pedestrian crossings or the effect of pri-
oritization of public transport. Some (theoretical) studies even
suggest that signal coordination has little effect when traffic is
saturated, and as a consequence, a green wave can not be cre-
ated for saturated traffic (Huang and Huang, 2003). Note that
the green wave scheme in this study corresponds to the desyn-
chronized scheme with α = 1 (i.e. the traffic light is always
green). For non-optimal green waves, one could define the ef-
fective green split ˜α < 1, accounting for a certain fraction of
the traffic that has to stop at each signal. Because ˜α can be
measured easily, the results of this study could still be applied
in this case.
Up to now, the influence of signal coordination on the emis-
sions produced by reverse-flow traffic has been neglected. It
can be assumed that signal coordination will only be installed
when traffic flow is dominant in a single direction, such that
effects caused by reverse-flow traffic can be neglected in first
order. Nevertheless, the regression analysis results in this pa-
per can be used to provide a rough estimate for the worst
case scenario, in which the traffic flows in both directions
are the same (Q
1
= Q
2
= Q and α
1
= α
2
= 0.5), and
where the implementation of a green wave signal coordina-
tion scheme in one direction results in the occurrence of a red
wave in the opposite direction. For example, before imple-
menting signal coordination, the total CO
2
emission (for the
section of interest) equals 2 · Q · hCO
tot
2
i
d
(α, Q). After im-
plementing signal coordination, the total emission would equal
Q·[hCO
tot
2
i
g
(α, Q)+hCO
tot
2
i
r
(α, Q)]. When the regression equa-
tions of Section 4.3 are filled in, it is found that, for low traf-
fic flows (<200 vehicles/h), the reduction in CO
2
emissions due
to the green wave is compensated almost completely by the in-
creased emissions caused by the reverse-flow traffic, if the latter
experiences a red wave. For higher traffic flows, the influence
of reverse-flow traffic becomes smaller, e.g. a reduction of the
beneficial effect by 50 % for flows of 400 vehicles/h, to about
10 % for flows of 800 vehicles/h. Similar results can be found
for NO
x
and PM
10
emissions. It can thus be concluded that, in
case the traffic flow is low and the flow in the reverse direction
is not negligible, one should aim to implement the signal coor-
dination scheme in such a way that no red wave is experienced
in the reverse direction. For higher traffic flows, possible effects
on the emission of reverse-flow traffic can be neglected.
Traffic signal coordination decreases travel times and in-
creases road capacity; the effect of facilitating traffic flow may
in the long term induce additional traffic (Hills, 1996; Kitamura,
2009). This side effect potentially offsets the beneficial envi-
ronmental consequences of signal coordination, or even make
the situation worse (Stathopoulos and Noland, 2003). Predict-
ing the amount of induced traffic is not a trivial task, because
it depends on a wide number of intricately interrelated factors
such as land use, accessibility or household’s decisions con-
cerning residence and job location (Kitamura, 2009). On the
10

other hand, the results in this paper allow, to some extent, to
estimate the influence of induced traffic on emissions. For ex-
ample, one can easily calculate the factor f by which the traffic
flow Q may increase in order to compensate for the effect of
lowered CO
2
emissions per trip, caused by the implementation
of a green wave; this will be the case when
Q · hCO
tot
2
i
d
(α, Q) = f Q · hCO
tot
2
i
g
(α, f Q). (15)
Given that we found hCO
tot
2
i
g
to be 40.2 g, independent of Q or
α, one finds that
f = 1 −
∆hCO
tot
2
i(α, Q)
40.2
. (16)
Allowed increases range from 20 % for low initial traffic flows
and high green split, to more than 100 % for high initial traf-
fic flows and low green split (i.e. the cases for which a green
wave would be most favorable). Similar ranges can be found
for NO
x
and PM
10
emissions. Potential increases in traffic flow
are however closely linked to the environmental capacity of the
road (Appleyard et al., 1981). The above analysis only con-
siders the effects of increased traffic on emissions; in practice,
such high increases in traffic flow may not be desired.
6. Conclusions
This paper reported on a study in which the influence of traf-
fic signal coordination on vehicle noise and air pollutant emis-
sions (CO
2
, NO
x
and PM
10
) was investigated. A microscopic
traffic simulation model was used, coupled with the recently
developed Imagine and VERSIT+ emission models, which re-
turn results representative for the Dutch vehicle fleet. A sim-
plified setting was considered, consisting of an urban arterial
road with a limit speed of 50 km/h, and equiped with five con-
secutive traffic signals, spaced at a distance of 200 m from each
other. Through the simulation of a range of scenarios, the in-
fluence of the traffic intensity, the signal coordination scheme
and signal parameters (cycle time and green split) on emissions
was investigated. In first order, the effects of cross-flow and
reverse-flow traffic were not considered.
It was found that, for the considered setting, the introduc-
tion of a green wave could potentially lower the emissions of
the considered air pollutants by 10 % to at most 40 % (if a per-
fect green wave is achieved), a range which is in accordance
with those reported in literature. The traffic intensity and the
green split were found to have the largest influence on the ef-
fective reduction in emission; the largest potential reduction oc-
curs when traffic intensities are close to capacity and the green
split is low. The cycle time was found to have a statistically
significant influence only on travel times, but not on the emis-
sion of the pollutants considered. The introduction of a green
wave resulted in all cases in an increase of the total emitted
noise level, by up to 0.6 dB(A). Sound pressure levels were
found to decrease by up to 1 dB(A) near the traffic signals, but
to increase by up to 1.5 dB(A) between intersections. Often,
efficient coordination of traffic signals is only possible in one
travelling direction, and the potential effects of this measure
on traffic driving in the reverse direction are also of concern.
A more detailed analysis in this work showed that installing
a green wave slightly increased road capacity, but that having
no signal coordination already represented the worst case re-
garding road capacity. This implies that implementing signal
coordination in one direction will not have an adverse effect on
road capacity in the opposite direction. Furthermore, for higher
traffic flows (>200 vehicles/h), possible effects of installing a
green wave on the emission of reverse-flow traffic can also be
neglected. Although applied to the specific case of traffic signal
coordination, the methodology presented in this paper could be
used to study the effects of a wide range of intelligent trans-
portation systems.
Acknowledgments
This study was performed within the framework of the Cen-
ter for Mobility and Public Works (Steunpunt Mobiliteit &
Openbare Werken, Spoor Verkeersveiligheid), which is sup-
ported by the Flemish Government. Bert De Coensel is a post-
doctoral fellow of the Research Foundation–Flanders (FWO–
Vlaanderen); the support of this organization is also gratefully
acknowledged.
References
Ahn, K., Rakha, H., Trani, A., Van Aerde, M., 2002. Estimating vehicle fuel
consumption and emissions based on instantaneous speed and acceleration
levels. J. Transp. Engrg. 128, 182–190.
Appleyard, D., Gerson, M.S., Lintell, M., 1981. Livable streets. University of
California Press, Berkeley, California, USA.
Barkenbus, J.N., 2010. Eco-driving: an overlooked climate change initiative.
Energy Policy 38, 762–769.
B
´
erengier, M.C., 2009. Emission, propagation and control of road traffic noise:
some solutions for the future, in: Proc. Inter·noise, Ottawa, Canada.
B
´
erengier, M.C., Picaut, J., 2008. Methods for noise control by traffic manage-
ment: Impact of speed reducing equipments. Deliverable H.R2. SILENCE
Project, Sixth Framework Programme, European Commission. Brussels,
Belgium.
Botteldooren, D., De Coensel, B., De Muer, T., 2006. The temporal structure
of urban soundscapes. J. Sound Vib. 292, 105–123.
Bretherton, D., Bodger, M., Baber, N., 2004. SCOOT — the future [urban traf-
fic control], in: Proceedings of the 12th International Conference on Road
Transport Information and Control (RTIC), London, UK. pp. 301–306.
Brockfeld, E., Barlovic, R., Schadschneider, A., Schreckenberg, M., 2001. Op-
timizing traffic lights in a cellular automaton model for city traffic. Phys.
Rev. E 64, 056132.
Cameron, G.D.B., Duncan, G.I.D., 1996. PARAMICS—parallel microscopic
simulation of road traffic. J. Supercomput. 10, 25–53.
Can, A., Leclercq, L., Lelong, J., 2008. Dynamic estimation of urban traffic
noise: influence of traffic and noise source representations. Appl. Acoust.
69, 858–867.
Can, A., Leclercq, L., Lelong, J., Defrance, J., 2009. Accounting for traffic dy-
namics improves noise assessment: Experimental evidence. Appl. Acoust.
70, 821–829.
Chen, K., Yu, L., 2007. Microscopic traffic-emission simulation and case study
for evaluation of traffic control strategies. J. Transpn. Sys. Eng. & IT 7,
93–100.
Cheng, S.F., Epelman, M.A., Smith, R.L., 2006. CoSIGN: A parallel algorithm
for coordinated traffic signal control. IEEE Trans. Intell. Transp. Syst. 7,
551–564.
Chowdhury, D., Schadschneider, A., 1999. Self-organisation of traffic jams in
cities: Effects of stochastic dynamics and signal periods. Phys. Rev. E 59,
R1311–R1314.
11

Coelho, M.C., Farias, T.L., Rouphail, N.M., 2005a. Impact of speed control
traffic signals on pollutant emissions. Transport. Res. D 10, 323–340.
Coelho, M.C., Farias, T.L., Rouphail, N.M., 2005b. A methodology for mod-
elling and measuring traffic and emission performance of speed control traf-
fic signals. Atmos. Environ. 39, 2367–2376.
David, G., 2000. Caract
´
erisation physique des sources sonores en milieu urbain
(Physical characterization of sound sources in urban environment). Ph.D.
thesis. University of Maine. France.
De Coensel, B., Botteldooren, D., De Muer, T., Berglund, B., Nilsson, M.E.,
Lercher, P., 2009. A model for the perception of environmental sound based
on notice-events. J. Acoust. Soc. Am. 126, 656–665.
De Coensel, B., Botteldooren, D., Vanhove, F., Logghe, S., 2007. Microsimu-
lation based corrections on the road traffic noise emission near intersections.
Acta Acust. Acust. 93, 241–252.
De Coensel, B., De Muer, T., Yperman, I., Botteldooren, D., 2005. The in-
fluence of traffic flow dynamics on urban soundscapes. Appl. Acoust. 66,
175–194.
Della Ragione, L., Meccariello, G., Punzo, V., Ciuffo, B., 2009. A prelimi-
nary study to evaluate emissions factors by real and micro simulated driving
cycle, in: Proceedings of the 9th International Conference on Engines and
Vehicles, Naples, Italy.
Desarnaulds, V., Monay, G., Carvalho, A., 2004. Noise reduction by urban
traffic management, in: Proc. ICA, Kyoto, Japan.
Ellebjerg, L., 2007. Effectiveness and benefits of traffic flow measures on
noise control. Deliverable H.D1. SILENCE Project, Sixth Framework Pro-
gramme, European Commission. Brussels, Belgium.
Ellenberg, M., Bedeaux, J.F., 1999. Calming waves for safety: A time to rethink
green waves? Traffic Technol. Int. April/May, 55–58.
European Commission, 2007. Towards a new culture for urban mobility. Green
Paper 551. Directorate General for Energy and Transport, European Com-
mission. Brussels, Belgium.
European Commission, 2008. Action plan for the deployment of intelligent
transport systems in Europe. Communication from the Commission 886.
European Commission. Brussels, Belgium.
Fritzsche, H.T., 1994. A model for traffic simulation. Traffic Eng. Contr. 35,
317–321.
Gartner, N.H., Stamatiadis, C., 2002. Arterial-based control of traffic flow in
urban grid networks. Math. Comput. Model. 35, 657–671.
Helbing, D., 2001. Traffic and related self-driven many-particle systems. Rev.
Mod. Phys. 73, 1067–1141.
Hills, P.J., 1996. What is induced traffic? Transportation 23, 5–16.
Huang, D.W., Huang, W.N., 2003. Traffic signal synchronization. Phys. Rev. E
67, 056124.
Int Panis, L., Broekx, S., Liu, R., 2006. Modelling instantaneous traffic emis-
sion and the influence of traffic speed limits. Sci. Total Environ. 371, 270–
285.
Kitamura, R., 2009. The effects of added transportation capacity on travel: a
review of theoretical and empirical results. Transportation 36, 745–762.
Knospe, W., Santen, L., Schadschneider, A., Schreckenberg, M., 2002. Human
behavior as origin of traffic phases. Phys. Rev. E 65.
Li, X., Li, G., Pang, S.S., Yang, X., Tian, J., 2004. Signal timing of inter-
sections using integrated optimization of traffic quality, emissions and fuel
consumption: a note. Transport. Res. D 9, 401–407.
Ligterink, N.E., De Lange, R., 2009. Refined vehicle and driving-behaviour
dependencies in the VERSIT+ emission model, in: Proceedings of the Joint
17th Transport and Air Pollution Symposium and 3rd Environment and
Transport Symposium (ETTAP), Toulouse, France.
Lum, K.M., Fan, H.S.L., Lam, S.H., Olszewski, P., 1998. Speed-flow modeling
of arterial roads in Singapore. J. Transp. Engrg. 124, 213–222.
Madireddy, M., De Coensel, B., Can, A., Degraeuwe, B., Beusen, B.,
De Vlieger, I., Botteldooren, D., 2011. Assessment of the impact of speed
limit reduction and traffic signal coordination on vehicle emissions using an
integrated approach. Transport. Res. D 16, 504–508.
Mensink, C., Cosemans, G., 2008. From traffic flow simulations to pollutant
concentrations in street canyons and backyards. Environ. Modell. Softw. 23,
288–295.
Nagatani, T., 2007. Vehicular traffic through a sequence of green-wave lights.
Physica A 380, 503–511.
Nagatani, T., 2009. Green-light paths in city traffic controlled by signals. Phys.
Lett. A 373, 536–539.
Neunh
¨
auserer, L., Diegmann, V., 2010. Analysis of the impacts of an environ-
mental traffic management system on vehicle emissions and air quality, in:
Proceedings of the 18th International Symposium Transport and Air Pollu-
tion (TAP’10), D
¨
ubendorf, Switzerland.
Osorio, C., Bierlaire, M., 2008. A multiple model approach for traffic signal
optimization in the city of Lausanne, in: Proceedings of the 8th Swiss Trans-
port Research Conference (STRC), Monte Verit
`
a, Ascona, Switzerland.
Pandian, S., Gokhale, S., Ghoshal, A.K., 2009. Evaluating effects of traffic
and vehicle characteristics on vehicular emissions near traffic intersections.
Transport. Res. D 14, 180–196.
Peeters, B., van Blokland, G., 2007. The noise emission model for European
road traffic. Technical Report — Deliverable 11 of the Imagine project
IMA55TR-060821-MP10. M+P Consulting Engineers. Vught, The Nether-
lands.
Rakha, H., Van Aerde, M., Ahn, K., Trani, A.A., 2000. Requirements for
evaluating traffic signal control impacts on energy and emissions based on
instantaneous speed and acceleration measurements, in: Proceedings of the
79th Annual Meeting of the Transportation Research Board, Washington
D.C., USA.
Shinar, D., Bourla, M., Kaufman, L., 2004. Synchronization of traffic signals
as a means of reducing red-light running. Hum. Factors 46, 367–372.
Silva, C.M., Farias, T.L., Frey, H.C., Rouphail, N.M., 2006. Evaluation of nu-
merical models for simulation of real-world hot-stabilized fuel consumption
and emissions of gasoline light-duty vehicles. Transport. Res. D 11, 377–
385.
Smit, R., Brown, A.L., Chan, Y.C., 2008. Do air pollution emissions and fuel
consumption models for roadways include the effects of congestion in the
roadway traffic flow? Environ. Modell. Softw. 23, 1262–1270.
Smit, R., McBroom, J., 2009. Use of microscopic simulation models to predict
traffic emissions. Road Transp. Res. 18, 49–54.
Smit, R., Smokers, R., Rab
´
e, E., 2007. A new modelling approach for road
traffic emissions: VERSIT+. Transport. Res. D 12, 414–422.
Smit, R., Smokers, R., Schoen, E., 2005. VERSIT+LD: Development of a
new emission factor model for passenger cars linking real-world emissions
to driving cycle characteristics, in: Proceedings of the 14th International
Symposium Transport and Air Pollution (TAP’05), Graz, Austria. pp. 177–
186.
Stathopoulos, F.G., Noland, R.B., 2003. Induced travel and emissions from traf-
fic flow improvement projects, in: Proceedings of the 82nd Annual Meeting
of the Transportation Research Board, Washington D.C., USA.
Tindale, S.A., Hsu, P.P., 2005. Crash data and signal coordination: A one-way
pair case study. J. Saf. Res. 36, 481–482.
Unal, A., Rouphail, N.M., Frey, H.C., 2003. Effect of arterial signalization and
level of service on measured vehicle emissions, in: Proceedings of the 82nd
Annual Meeting of the Transportation Research Board, Washington D.C.,
USA.
Varas, A., Cornejo, M.D., Toledo, B.A., Mu
˜
noz, V., Rogan, J., Zarama, R.,
Valdivia, J.A., 2009. Resonance, criticality, and emergence in city traffic
investigated in cellular automaton models. Phys. Rev. E 80, 056108.
Warberg, A., Larsen, J., Jørgensen, R.M., 2008. Green Wave Traffic Optimiza-
tion — A Survey. Technical Report. Department of Informatics and Mathe-
matical Modelling, Technical University of Denmark. Lyngby, Denmark.
Webster, F.V., 1958. Traffic signal setting. Road Research Technical Paper 39.
Her Majesty’s Stationery Office. London, UK.
Zallinger, M., Luz, R., Hausberger, S., Hirschmann, K., Fellendorf, M., 2010.
Coupling of microscale traffic and emission models to minimize emissions
by traffic control systems, in: Proceedings of the 18th International Sympo-
sium Transport and Air Pollution (TAP’10), D
¨
ubendorf, Switzerland.
Zhang, Y., Chen, X., Zhang, X., Song, G., Hao, Y., Yu, L., 2009. Assessing
effect of traffic signal control strategies on vehicle emissions. J. Transpn.
Sys. Eng. & IT 9, 150–155.
Zito, P., 2009. Influence of coordinated traffic lights parameters on roadside
pollutant concentrations. Transport. Res. D 14, 604–609.
12