Mesoscopic Urban-Traffic Simulation Based on Mobility Behavior to Calculate NOx Emissions Caused by Private Motorized Transport

Abstract

Motorized transport is one of the main contributors to anthropogenic CO 2 emissions, which cause global warming. Other emissions, like nitrogen oxides or carbon monoxide, are detrimental to human health. A prominent way to understand and thus be able to minimize emissions is by using traffic simulations to evaluate different scenarios. In that way, one can find out which policies, technical innovations, or behavioral changes can lead to a decrease in emissions. Since the effect of CO 2 is on a global scale, a macroscopic model is often enough to find reasonable results. However, NO x emissions can also have a direct, local effect. Therefore, it is interesting to investigate these emissions on a mesoscopic scale, to gain insight into the local distribution of this pollutant. In this study, we used a traffic model that, contrary to most other state-of-the-art traffic simulations, does not require an origin–destination matrix as an input, but calculates it from mobility behavior extracted from a survey. We then generated agents with realistic mobility behavior that perform their daily trips and calculate key features like congestion and emissions for every edge of the road network. Our approach has the additional advantage of allowing to investigate technical, juridical, as well as behavioral changes, all within the same framework. It is then possible to identify strategies that minimize NO x emissions caused by private motorized transport. Evaluation showed good agreement with reality in terms of local and temporal resolution. Especially when looking at the sum of emissions, the main feature for evaluating policies, and deviations between our simulation and available statistics were negligible. We found that, from all scenarios we investigated, the ban of old diesel cars is the most promising policy: By replacing all diesel cars built in 2005 or earlier with petrol cars of the same age, NO x emissions could drop by roughly a third.

Full text

atmosphere
Article
Mesoscopic Urban-Traffic Simulation Based on
Mobility Behavior to Calculate NOx Emissions
Caused by Private Motorized Transport
Simon Plakolb 1, Georg Jäger 1,*, Christian Hofer 2and Manfred Füllsack 1
1
Institute of Systems Sciences, Innovation and Sustainability Research, University of Graz, 8010 Graz, Austria;
simon.plakolb@uni-graz.at (S.P.); manfred.fuellsack@uni-graz.at (M.F.)
2Information Technologies & Business Informatics, CAMPUS 02 University of Applied Sciences, 8010 Graz,
Austria; christian.hofer@campus02.at
*Correspondence: georg.jaeger@uni-graz.at
Received: 29 March 2019; Accepted: 21 May 2019; Published: 28 May 2019


Abstract:
Motorized transport is one of the main contributors to anthropogenic CO
2
emissions, which
cause global warming. Other emissions, like nitrogen oxides or carbon monoxide, are detrimental
to human health. A prominent way to understand and thus be able to minimize emissions is by
using traffic simulations to evaluate different scenarios. In that way, one can find out which policies,
technical innovations, or behavioral changes can lead to a decrease in emissions. Since the effect of
CO
2
is on a global scale, a macroscopic model is often enough to find reasonable results. However,
NO
x
emissions can also have a direct, local effect. Therefore, it is interesting to investigate these
emissions on a mesoscopic scale, to gain insight into the local distribution of this pollutant. In this
study, we used a traffic model that, contrary to most other state-of-the-art traffic simulations, does not
require an origin–destination matrix as an input, but calculates it from mobility behavior extracted
from a survey. We then generated agents with realistic mobility behavior that perform their daily
trips and calculate key features like congestion and emissions for every edge of the road network.
Our approach has the additional advantage of allowing to investigate technical, juridical, as well
as behavioral changes, all within the same framework. It is then possible to identify strategies
that minimize NO
x
emissions caused by private motorized transport. Evaluation showed good
agreement with reality in terms of local and temporal resolution. Especially when looking at the sum
of emissions, the main feature for evaluating policies, and deviations between our simulation and
available statistics were negligible. We found that, from all scenarios we investigated, the ban of old
diesel cars is the most promising policy: By replacing all diesel cars built in 2005 or earlier with petrol
cars of the same age, NOxemissions could drop by roughly a third.
Keywords:
traffic simulation; agent-based model; spatial networks; congestion; emissions; nitrogen
oxides; incomplete combustion
1. Introduction
The impact of emissions coming from motorized traffic is severe, not only on a global scale but
also locally. Especially in urban environments where congestion plays a significant role and population
density is high, local pollutants cause detrimental effects on human health [
1
]. Particularly problematic
are nitrogen oxides (NO
x
). These have many different negative effects: they increase the formation
of ozone in the troposphere, contribute to acid rain, and reduce air quality [
2
]. Those direct negative
effects are especially pronounced for citizens living in an urban area [
3
,
4
]. In 2018, the road-transport
sector had the highest impact on NO
x
emissions in the EU-28 with 39%. [
5
] A case study investigating
Atmosphere 2019,10, 293; doi:10.3390/atmos10060293 www.mdpi.com/journal/atmosphere

Atmosphere 2019,10, 293 2 of 14
the city of Lyon [
6
] found that traffic was the main contributor, responsible for over 50% of total NO
x
emissions. When considering citizen exposure to NO
2
in urban areas, the relative contribution of the
road sector is even bigger [
5
]. This extent of emissions is not only caused by the higher population
density of an urban environment, but also by high congestion. Congested roads lead to an increase
in traffic emissions and thus health risks for people in these areas [
4
]. In order to quantify those
health risks, emission inventories created by coupled traffic and emissions models are then fed into
meteorological and atmospheric chemistry transport models to yield their effect on air quality [
7
].
Subsequently, human exposure models link the concentration of pollutants with human factors [
8
].
There is not necessarily a linear relation of the concentration to health effects. Thus, together with data
on other adverse substances, the health hazard can finally be modeled [9].
In order to combat the negative effects of traffic-related emissions, infrastructural and policy
changes in a city’s road network are necessary. A big problem in planning such endeavors is estimating
the possible gain of specific policies. Where experiments in the real world are hard if not impossible
to conduct, one has to rely on models. With these, one can simulate certain scenarios of interest to
investigate the change they could bring [
10
]. Traffic systems exhibit complex dynamics induced by
nonlinear behavior through internal and external factors [
11
], and thus need sophisticated modeling
approaches. They can be subsumed into two categories: the bottom–up approach of microscopic
modeling, and the top–down variant of macroscopic modeling [
10
,
12
]. In between these categories,
mesoscopic models have emerged as a means to devise detailed simulations without extensive
computational cost [13–15].
State-of-the-art models include commercial ones like VISSIM [
16
], and noncommercial ones like
MATSIM [
17
] and SUMO [
18
]. A significant challenge for most of these is to find data about the origin
and destination of travelers (OD data). OD data are difficult to find, whether it is through surveys [
19
]
or more novel techniques like Bluetooth [
20
,
21
] and mobile-phone-network data [
22
,
23
]. A novel
approach by Reference [
24
] remedied this shortcoming. Drawing on an empirical foundation of data
about mobility behavior, OD data are calculated within the simulation. Furthermore, by dropping the
traditional car-following paradigm of Reference [
25
] in favor of a less detailed link-by-link evaluation,
computation becomes less expensive and can be parallelized. This approach also reduces the risk for
pitfalls of too-high or -low resolution [
26
]. Thus, this model is especially targeted toward the full-scale
assessment of urban areas in the interest of comparing diverse scenarios with different parameters.
Recently, hybrid-modeling advances have emerged. They combine macroscopic traffic with
instantaneous emissions models [
14
,
27
]. In Reference [
13
], the method subsequently provided a
framework to assess CO
2
emissions in an urban environment. Additionally, the model provides spatial
resolution that enables the estimation of pollutants with more local effects as well. In this study, we
adapted the model to investigate NO
x
emissions with local resolution. Usually, both microscopic and
macroscopic emission models considerably overestimate the NO
x
emissions [
28
,
29
], and microscopic
models often lack the required spatial detail on pollutants [
26
]. Hence, we hope that a hybrid approach
would lead to better results.
In order to showcase the traffic model presented here, we investigate the city of Salzburg, which
is an Austrian city struggling with meeting the EU regulations regarding NO
x
[
30
]. Additionally, it is
one of the most congested cities in Europe [
31
]. In this study, we investigate different scenarios that
could help alleviate traffic-related problems. We simulated technical as well as societal or juridical
changes. The used traffic model enabled us to study all scenarios within the same framework. It thus
became possible to quantitatively compare the results and identify promising solutions.
This manuscript is organized as follows: The used model, as well as the investigated scenarios,
are introduced in Section 2. Results of the baseline simulation and for the scenarios are presented and
evaluated in Section 3. Section 4concludes with a discussion of the results, and presents limitations
and possible expansions of the model.

Atmosphere 2019,10, 293 3 of 14
2. Materials and Methods
The goal of the presented model is to calculate the NO
x
emissions coming from private motorized
transport (PMT) in an urban environment. However, the input of the model is not an origin–destination
matrix like in most conventional traffic models. Instead, it uses data about the mobility behavior
of the citizens [
24
]. That way, the model is capable of investigating scenarios that affect the origins
and destinations of people, like juridical or social changes, making us able to compare various
scenarios within the same framework. The drivers populating the model interact via their reaction
and contribution to congestion [
32
]. They are thus depicted as agents who serve as distinct entities;
agent-based models are comprised of [
33
]. While there is no clear definition of what such an agent
is, the literature seems to agree in various points. Agents share an environment in which they are
individually located and clearly identifiable. They have an initial state, react autonomously to their
environment depending on its state and their own state, and interact with each other [33–36].
Note that the scope of the model is limited to emissions from private motorized car traffic. Agents
also use other modes of transport (public transport, bikes, etc.), but only emissions from private cars
use are calculated and stored.
The description of the model follows [13]. The main steps of the simulation are:
•Creating a traffic network from map data;
•generating the ways from mobility-behavior data; and
•calculating congestion and emissions.
These steps are illustrated in Figure 1and they are explained in detail in the following sections.
during congestions
avoidance iterations
finally
Figure 1. Schematic diagram of model’s individual steps.
2.1. Creating a Traffic Network
First, the investigated traffic system needs to be translated from a map to a spatial traffic
network. This is done by interpreting all intersections and similar points of interest as nodes of
a network. The road segments connecting them are the edges of the network. We used map data from
OpenStreetMap (OSM) [
37
], which we transformed using the OSMnx library [
38
]. It can extract not
only nodes and edges, but also relevant information about the roads (like speed limit and number of
lanes). These are stored as edge attributes so they can be accessed during the simulation. A comparison
between the map representation and the network representation of the city of Salzburg is shown in
Figure 2.
While some of the relevant information is already included in the map itself, other features need
to be calculated. The most important ones are the fastest paths [
39
] between all nodes in the network.

Atmosphere 2019,10, 293 4 of 14
They are used by the agents every time they make a decision about their path. Here, we could utilize
the fact that the fastest paths are independent of the agent itself, and rather an intrinsic network
property. By calculating them before the simulation and storing them as a node property, we can save
a significant amount of computation time. If computer memory is an issue, it is also possible to not
save the full information about all fastest paths. Calculating lists of nodes in reach within certain
distances of each node is sufficient. For example, one could use driving distances of 100, 200, and
300 m. This facilitates computational speed. During the simulation, agents often search for a possible
target node that is a certain distance away from them. By storing these possibilities as node properties,
we minimize the amount of fastest path calculations that need to be performed. Note that fastest paths
are calculated here as weighted shortest paths, where the weight is the average time necessary to move
over the road represented as a network edge.
Figure 2.
Comparison between the (
left
) network representation and (
right
) map of the center of the
investigated city of Salzburg.
2.2. Generating Ways
A novel aspect of the presented model is that ways are not taken from origin–destination matrices,
but calculated within the model based on mobility-behavior data. We imported data from “Österreich
Unterwegs 2013/2014” [
40
], a survey containing information about the mobility behavior of 18,000
Austrian respondents. Each respondent reported on two randomly selected days, leading to a dataset
of roughly 36,000 days, distributed around the year. For each day, they described the number, length,
and starting times of all trips they performed, which method of transportation they used, which car
they used, and the reason for the trip (commute, leisure, or shopping). Information about the used
cars includes the size of the car, divided into the three categories small (<1155 kg), medium, and large
(>1550 kg), the year during which it was built, and the type of engine it uses (petrol, diesel, or other).
This information, in combination with statistical data about the average emissions of cars built in a
certain year [
41
], was used to approximate the average fuel consumption of each car. Details on this
process can be found in Reference [
13
]. Note that the start and end point of the trips are not precisely
known, but we do have access to demographic information about the respondents, like their age or
their place of residence.
Using these empirical data as foundation, we generated mobility-behavior archetypes.
Each archetype contains information about the performed trips, the used car, and an age bracket, but
no specific start or end points. When investigating urban systems, it is viable to restrict the used
archetypes to those originating from respondents living in the urban area in the scope, which was
done for this study.

Atmosphere 2019,10, 293 5 of 14
Once the network and the archetypes are set up, the actual agent-based simulation begins. Since
we were interested in a full-scale model, we generated one agent for each inhabitant of the investigated
city. They were positioned in the traffic network according to the population density of the area. If such
information was available for the investigated system, the agents were separated into age brackets.
Then they were assigned a random archetype corresponding to their age. They performed all trips
given by the archetype with the respective mode of transportation. Choosing a target node for each
trip is done by utilizing previously stored information about the distances between all nodes. If the
agent, for example, needs to perform a trip of 1000 m, it can obtain a list of possible destinations with
that distance from the node at which it is currently positioned. It then chooses a random viable node
and loads or calculates the fastest path to that node. Each edge it crosses is used to store the relevant
data of the trip, e.g., type of engine, consumed fuel, and time. Saving this information as an edge
property rather than as agent property gives us the local resolution about resulting emissions. This is
especially relevant for emissions with local effects, like NOx.
In addition to the inhabitants of the investigated city, we also needed to include commuters
coming from outside the city. They enter the city road network through the main and secondary roads
connecting it with the surrounding network. Thus, nodes of the network where these connections
would be placed in the total network were implemented as sources where commuters emerge.
The number of commuters and their geographical origin are extracted from Reference [
42
]. However,
not all of these enter the city at the same time. This was modeled by a Gaussian distribution around
7:00 for the time of entry into the network. Consequently, only 68.27% of commuters enter the city
at 7:00, while at 6:00 and at 8:00, 15.865% of them enter. Departure is carried out in a similar manner.
The peak of distribution was set to 18:00, with 68.27% of agents exiting, and 15.865% exiting at 17:00
and 19:00.
The choice of which node to utilize for entering or exiting the city is based on the node properties
and the coordinates of the commuter agent’s origin. Among the node properties were the speed limit
at the connecting road, their connection to the other nodes, and their ‘highway’ tag in OSM. The latter
served as an adjustment parameter [
43
] for the free speed. Connected to the commuter’s origin and
destination, influencing factors are directness and travelled distance. These serve as a proxy for the
perceived difference in appeal for route choices.
The resulting paths, both from commuters and noncommuters, are only the first step in an
iterative process. During the path selection of each agent, no information about congestion is available.
Consequently, there is no direct interaction between agents. This interaction is later included in an
iterative way by calculating the resulting congestion from the initial simulation run and using this
information during the path selection in a second run. In effect, some agents avoid heavily congested
roads. Previous work [
32
] showed that most realistic results can be obtained by using just one such
iterative step, with 10% of agents changing their path due to congestion, which is comparable to the
value found by other studies of this behavior [44].
Once realistic paths exist, we can then calculate the resulting congestion, as well as emissions
with a local and temporal resolution.
2.3. Calculating Emissions
During the generation of the paths, the average fuel consumed by each car is already stored as an
edge property based on the age and the size of the car, and the year in which it was built. Microscopic
effects like speed dependency, ambient temperature, and cold start are only included through averages,
not for individual cars. Other effects, like the state of traffic flow, cannot be included using a
macroscopic perspective, and we need to calculate them in detail in each simulation run. Emissions
of CO
2
, NO
x
, and others scale very differently in free-flowing, congested, or stop-and-go traffic [
45
].
Thus, we need the state of the traffic flow on the respective road segment at the respective time.
In a first step, we need to calculate the capacity of each road segment. Hourly traffic capacity
Ch[46] can be calculated via [47]

Atmosphere 2019,10, 293 6 of 14
Ch=Leff 750 cars
hour , (1)
with
Leff
being the effective number of lanes.
Leff
is calculated starting from the actual number of
lanes. If this number is unknown, it is possible to approximate values of 2.6 for roads wider than 7.5 m,
2.0 for roads between 7.5 and 5.5 m, and 0.8 for roads narrower than 5.5 m [
46
]. For roads that are not
one-way streets, this number of lanes is then halved, since only half of the traffic flow moves in the
relevant direction there. The hourly capacity is then compared to the actual number of cars using this
road for each hour. This leads to road-utilization factor a:
a=
Vh
Ch(2)
The road-utilization factor can then be used to obtain the stage of traffic flow:
a
smaller than 0.75
can be seen as free-flowing traffic,
a
between 0.75 and 0.9 is congested traffic flow, and
a
larger than 0.9
corresponds to stop-and-go traffic [46].
For finding NO
x
emissions, we start out from fuel consumption calculated by the model. Using
the European exhaust-emission standard [
48
], we had access to estimates of NO
x
emissions for different
engine types. Because of the mesoscopic perspective of the model, which is required for fast calculation,
there is no precise information about the age of each car during the simulation run, and only their
fuel consumption and type of engine are known. Therefore, we need to calculate NO
x
emissions
from fuel consumption. This is not trivial. Due to the difference in EURO norms and the used engine
type, there is no linear dependency. We based this calculation on the most common standard in the
dataset (EURO 5) and the finding that especially diesel engines emit more nitrogen oxides in real-life
conditions than they do under lab conditions. A study [
49
] that investigated this phenomenon found
that, under realistic conditions, diesel engines in compliance with the EURO 5 norm [
48
] on average
emit 3.5 times the emission limit. This gives us the required factors for EURO 5 diesel and petrol cars.
Cars with higher fuel consumption gain an additional term for NO
x
emissions, linear to their increased
fuel consumption. This approximation is sufficient for cars built in 2000 and earlier, but leads to errors
for older cars. Fortunately, the investigated dataset contains only few such cars, so that the final result
is still close to the measured value (see Section 3).
Combining these estimates with factors that describe additional emissions caused by congestion or
stop-and-go traffic [
45
], we arrived at NO
x
emissions for every edge of the investigated traffic system.
2.4. Investigated Scenarios
In a baseline scenario, we depict the system as it currently is. Details about the investigated
system are given in Table 1.
Table 1. Key data of the investigated traffic system.
Key Property Value
Citizens 136,000
Commuters 33,500
Average daily trips per citizen 2.8
Average daily car trips per citizen 1.0
Average car trip length 15.1 km
Average car trip length (within city) 5.5 km
In addition to the baseline scenario, we also investigated other scenarios in order to assess the
potential of possible policies to reduce NOxemissions.
As a first scenario, we looked into the replacement of a certain percentage of cars by electric cars.
We scanned a large parameter space (2.5% to 20%), since the number of electric cars in the near future
is currently unknown and difficult to predict. This scenario also serves as a method for evaluation
purposes: We expect that replacing, e.g., 10% of cars with electric cars should lead to a decrease of

Atmosphere 2019,10, 293 7 of 14
nitrogen oxides by roughly 10%. Furthermore, this scenario can be used as a comparison for the
effectiveness of other scenarios.
In a second scenario, we investigated a societal change: We increased the amount of
telecommuting, a possible effect of digitalization. To represent this change in the model, we
could remove a certain percentage of the trips that are performed for the reason of commuting.
This information is already contained and calibrated to empirical data within the model. By removing
up to 50% of work-related trips in increments of 10%, five subscenarios were simulated. These changes
primarily affect rush hour, where most roads are congested. Thus, a nonlinear effect of emission
reduction could be possible. Heavily used roads could also locally benefit more from this change.
Scenario 3 investigated a possible policy often discussed in the context of urban air quality. Here,
we calculated the change resulting from replacing old diesel cars with cars that use a petrol engine.
As variants of this scenario, we used the years 1995, 2000, 2005, and 2010, and substituted all diesel
cars built prior. Additionally, a simulation with all diesel cars being replaced showed what a complete
ban on these could yield. While possibly having a negative effect on CO
2
emissions, such a change has
the potential of significantly reducing NOxemissions.
3. Results
The baseline scenario leads to daily NO
x
emissions of 746
±
7 kg. Their local distribution is
shown in Figure 3. Emissions mainly originate from the city center, and roads leading to and from the
city center. We can also see bottlenecks where two roads merge into one, creating higher congestion
and thus higher emissions.
Figure 3.
Results for baseline scenario. Emissions are primarily concentrated in the city center, and the
roads leading to and from the city center. One can also observe bottlenecks, where two streets merge
into one, leading to higher congestion and thus higher emissions. Color indicates NO
x
emissions in
kg
day·m.

Atmosphere 2019,10, 293 8 of 14
In order to evaluate the results of the baseline scenario, we used statistical data about NO
x
emissions in Salzburg. The most recent data with regional resolution are available from the year
2006 [
50
]. There, we found daily NO
x
emissions of roughly 2000 kg, of which about 680 kg were caused
by PMT, and are thus within the scope of our model. Using the increase of cars in the city as a rough
estimate for the increase in NO
x
emissions, we arrived at a value of 748
±
37 kg per day (approximating
the resulting error with 5%). Our simulations yielded 746 ±7 kg, which is a satisfying result.
For evaluating the temporal and spatial resolution of the model, one has to rely on congestion
data, since no high-resolution data about NO
x
emissions are available. Such a comparison is shown in
Figure 4. It shows simulated congestion during morning rush hour, compared to congestion data from
Google Maps. A quantitative evaluation of the differences between real and simulated congestion of
the used model [32] showed average deviations of ±3%.
Figure 4.
Comparison between simulated (
left
) and real congestion (
right
) during morning rush hour
(8:30), taken from Google Traffic.
In addition, we could compare the results of the baseline scenario (Figure 3) to an air-quality
map from Reference [
51
]. Note that this map shows NO
x
concentration, while we calculated emissions.
Nevertheless, both maps show increased NO
x
on the highway, around the city center, and on roads
connecting the highway and the city center.
Using this baseline result, we could then compare the different scenarios in terms of overall
effect and localization of the effect. Results for Scenario 1, where we investigated the effects of an
increase in the percentage of electric cars, are presented in Figure 5. The left panel shows the overall
reduction effect of this scenario for different intensities. The right panel shows an emission map
of the investigated scenario with the highest intensity. As expected, the decrease in emissions was
linear to the percentage of electric cars. This effect is evenly distributed throughout the city, since
the replaced cars were chosen randomly. Results of this scenario cannot only be used to evaluate the
model. They also give a point of reference in order to judge the effect of other scenarios.

Atmosphere 2019,10, 293 9 of 14
Figure 5.
Results for Scenario 1. By increasing the percentage of electric cars, emissions decrease
linearly throughout the city.
In Scenario 2, we simulated an increase in telecommuting, effectively removing a certain
percentage of commuting trips from the system. Results of this investigation are presented in Figure 6.
The effect is nonlinear: removing a certain amount of commuting trips reduces emissions more than
randomly removing the same amount or trips without regarding the purpose of the trip. This has to
do with the fact that commuting trips are very likely to occur on streets and during times where there
is much congestion. Thus, removing those trips significantly alleviates congestion, leading to lower
NOxemissions.
Figure 6.
Results for Scenario 2. Reducing the number of commuting trips is quite efficient. It removes
trips from congested roads at times where there is a high chance of stop-and-go traffic, leading to
reduced overall emissions.
Older cars with a diesel engine produce more NO
x
per km than cars using a petrol engine.
A viable way of reducing NO
x
emissions is hence to replace older diesel cars with petrol cars. Results
of simulations where diesel cars built before a certain year were replaced by petrol cars of the same age
are presented in Figure 7. When this policy only affects cars built before 2000, the effects are relatively
small (roughly 5% decrease in emissions). However, when the intensity of this policy is increased
by also banning cars built in 2010 and older, the effect on NO
x
emissions is drastic. Banning diesel
cars built before 2010 leads to a reduction of more than 50%. In a hypothetical case where all cars
are replaced by petrol cars, NO
x
emissions would drop below 200 kg/day. Such a policy has also
significant downsides. While the emission of NO
x
would decrease, CO
2
and other emissions might
rise significantly. Even though the CO
2
emissions of petrol engines depend on many different factors,
for example, engine size, on average they tend to emit more CO
2
than their diesel counterparts. Even

Atmosphere 2019,10, 293 10 of 14
though carbon dioxide has no direct local effects on air quality, it contributes to global warming and
should therefore be minimized in a sustainable traffic system.
Figure 7.
Results for Scenario 3. While the effect of replacing very old diesel cars (built before 2000)
with petrol cars of the same age is small, this scenario shows a drastic decrease in NO
x
emissions when
affecting all cars built before 2010. Replacing all diesel cars with cars powered by a petrol engine would
reduce the NOxemissions coming from private motorized transport (PMT) to below 200 kg/day.
In terms of computation time, the model can operate significantly faster than real time: Simulating
one day of traffic in Salzburg (over 100,000 citizens) takes approximately 20 min when utilizing 30
threads on a 3.4 GHz CPU. In contrast, other hybrid models optimized for quick computation calculate
their results in real time [15].
4. Discussion
In this study, we used a hybrid traffic model based on mobility-behavior data to calculate NO
x
emissions resulting from PMT. We applied the model to the city of Salzburg and investigated several
scenarios that included technological, sociological, and juridical changes to the traffic system, all within
the same framework.
Evaluating the baseline scenario, which found daily NO
x
emissions of 746
±
7 kg, showed good
agreement with statistical data. The local distribution of the emissions also yielded realistic results.
Emissions were concentrated in the city center and on large roads leading to it. On road segments with
high congestion (bottlenecks), NO
x
emissions were also significantly higher. This is also in agreement
with real-life measurements [52].
Scenario 1 found that increasing the number of electric cars linearly decreases the resulting NO
x
emissions. However, achieving a high percentage of electric cars is currently not feasible [
53
–
55
],
even though there are many policies that try to promote using electric cars in an urban environment [
56
,
57
].
Furthermore, the overall environmental benefit of electric vehicles is difficult to assess [
58
] and largely
depends on the way the electrical energy is produced. It is thus possible that electric cars decrease
local emissions at the cost of increasing global emissions.
Scenario 2 showed that telecommuting has great potential to decrease urban NO
x
emissions.
The nonlinear effect can be attributed to the fact that commuting trips primarily occur in places
and at times where and when there is a lot of congestion. Removing them does not only remove
direct emissions, but also effectively reduces emissions by other cars. As a further advantage of
telecommuting, its adoption only needs minimal infrastructure or other forms of investment. It is
already beginning to gain popularity due to digitization. Further promoting telecommuting might
thus be a feasible way to reduce urban emissions in general.
Scenario 3 showed the biggest improvement in terms of NO
x
emissions. It replaced old diesel
cars with cars of the same age that use a petrol engine. However, this decrease in NO
x
emissions came

Atmosphere 2019,10, 293 11 of 14
at the cost of increased CO
2
emissions. The decrease in NO
x
was smaller than the increase in CO
2
.
Nevertheless, it is difficult, if not impossible, to answer the question of whether a local benefit can
outweigh a global drawback.
The presented model has various advantages compared to traditional state-of-the-art traffic
models. Its foundation in mobility-behavior data makes it very flexible, so that diverse scenarios can
be investigated. The used hybrid approach gives a mix of detail and calculation speed that is perfectly
fitted for scenario evaluation. However, there are also some limitations to the model and possible
expansions that could improve its predictive power. Since OD data are based on mobility behavior,
we have no access to the target points of commuting trips, but have to generate them randomly at
the correct distances. This has only a minor effect on overall emissions (since distances are correct),
but might lead to an error in the local distribution of emissions. This could be remedied by including
information about the workplace density of the investigated system. Unfortunately, such information is
not available for every city. Including it would, hence, drastically limit the scope of the model. Further
improvement could be to approximate these data from available OSM data. While the mesoscopic
approach leads to fast computation times, it also has its downsides. For example, we lose some
information about individual cars, like their exact age, during the simulation. For this investigation,
it was possible to find a way to approximate it based on fuel consumption, which lead to satisfying
results for NO
x
emissions. This worked because the used dataset mainly contained cars in the EURO
5 and EURO 4 category. When investigating a different system, where the distribution of car ages is
different, it might be beneficial to forgo some computational speed in order to microscopically calculate
NOxemissions. This could be a further improvement of the model.
In conclusion, the presented hybrid traffic model proved to be very well-suited for investigating
NO
x
emissions resulting from PMT. Its local resolution gives more insight than macroscopic models. It
still remains fast enough to simulate whole cities on a 1:1 scale quicker than real time. Since it does not
rely on OD data, it is possible to investigate scenarios that not only change certain properties of the
cars, but also produce changes in the trips that people take. This provides us with a powerful tool to
holistically investigate urban-traffic scenarios.
Author Contributions:
conceptualization, S.P., G.J., C.H., M.F.; methodology, S.P., G.J., C.H., M.F.; software,
S.P., G.J., C.H.; validation, S.P., G.J., C.H.; investigation, S.P., G.J., C.H.; data curation, S.P., G.J., C.H.;
writing—original-draft preparation, S.P., G.J.; writing—review and editing, S.P., G.J., C.H., M.F.; visualization, S.P.,
G.J.; supervision, M.F.
Funding: This research received no external funding.
Acknowledgments: The authors would like to thank all contributors of OpenStreetMap for their efforts.
Conflicts of Interest: The authors declare no conflict of interest.
Abbreviations
The following abbreviations are used in this manuscript:
CO2carbon dioxide
NO2nitrogen dioxide
NOxnitrogen oxides
OD origin–destination
OSM Open Street Map
PMT private motorized transport
References
1.
Krzyzanowski, M.; Kuna-Dibbert, B.; Schneider, J. Health Effects of Transport-Related Air Pollution; World
Health Organization: Geneva, Switzerland, 2005.

Atmosphere 2019,10, 293 12 of 14
2.
Verstraeten, W.W.; Boersma, K.F.; Douros, J.; Williams, J.E.; Eskes, H.; Liu, F.; Beirle, S.; Delcloo, A.
Top-Down NO
x
Emissions of European Cities Based on the Downwind Plume of Modelled and Space-Borne
Tropospheric NO2Columns. Sensors 2018,18, 2893. doi:10.3390/s18092893.
3.
Nicolai, T.; Carr, D.; Weiland, S.; Duhme, H.; von Ehrenstein, O.; Wagner, C.; von Mutius, E. Urban traffic and
pollutant exposure related to respiratory outcomes and atopy in a large sample of children.
Eur. Respir. J.
2003,21, 956–963. doi:10.1183/09031936.03.00041103a.
4.
Zhang, K.; Batterman, S. Air pollution and health risks due to vehicle traffic. Sci. Total Environ.
2013
,
450–451, 307–316. doi:10.1016/j.scitotenv.2013.01.074.
5.
Air Quality in Europe—2018 Report; Technical report; European Environment Agency: Copenhagen,
Denmark, 2018.
6.
Nguyen, C.; Soulhac, L.; Salizzoni, P. Source Apportionment and Data Assimilation in Urban Air Quality
Modelling for NO2: The Lyon Case Study. Atmosphere 2018,9, 8. doi:10.3390/atmos9010008.
7.
Matthias, V.; Arndt, J.A.; Aulinger, A.; Bieser, J.; van der Gon, H.D.; Kranenburg, R.; Kuenen, J.; Neumann, D.;
Pouliot, G.; Quante, M. Modeling emissions for three-dimensional atmospheric chemistry transport models.
J. Air Waste Manag. Assoc. 2018,68, 763–800. doi:10.1080/10962247.2018.1424057.
8. EPA. Human Exposure Modeling—Overview; EPA: Washington, DC, USA, 2018.
9.
Künzli, N.; Kaiser, R.; Medina, S.; Studnicka, M.; Chanel, O.; Filliger, P.; Herry, M.; Horak, F.;
Puybonnieux-Texier, V.; Quénel, P.; et al. Public-health impact of outdoor and traffic-related air pollution: A
European assessment. Lancet 2000,356, 795–801. doi:10.1016/s0140-6736(00)02653-2.
10.
Kotusevski, G.; Hawick, K. A Review of Traffic Simulation Software. Res. Lett. Inf. Math. Sci.
2009
,13, 35–54.
11.
Zegeye, S.; De Schutter, B.; Hellendoorn, J.; Breunesse, E.; Hegyi, A. Integrated macroscopic traffic flow,
emission, and fuel consumption model for control purposes. Transp. Res. Part C Emerg. Technol.
2013
,
31, 158–171. doi:10.1016/j.trc.2013.01.002.
12.
Bert, E.; Chung, E.; Dumont, A.G. Exploring the use of DTA for origin-destination matrix estimation.
In Proceedings of the 6th STRC Swiss Transport Research, Monte Verita, Switzerland, 16–18 May 2006.
13.
Hofer, C.; Jäger, G.; Füllsack, M. Large scale simulation of CO
2
emissions caused by urban car traffic: an
agent-based network approach. J. Clean. Prod. 2018,183, 1–10.
14.
Thonhofer, E.; Palau, T.; Kuhn, A.; Jakubek, S.; Kozek, M. Macroscopic traffic model for large scale urban
traffic network design. Simul. Model. Pract. Theory 2018,80, 32–49. doi:10.1016/j.simpat.2017.09.007.
15. Jamshidnejad, A.; Papamichail, I.; Papageorgiou, M.; De Schutter, B. A mesoscopic integrated urban traffic
flow-emission model. Transp. Res. Part C Emerg. Technol. 2017,75, 45–83. doi:10.1016/j.trc.2016.11.024.
16.
Fellendorf, M.; Vortisch, P. Microscopic Traffic Flow Simulator VISSIM; Springer: New York, NY, USA, 2011;
Chapter 2, pp. 63–93. doi:10.1007/978-1-4419-6142-6\_2.
17.
Horni, A.; Nagel, K.; Axhausen, K. The Multi-Agent Transport Simulation MATSim; Ubiquity Press: London,
UK, 2016.
18.
Lopez, P.A.; Behrisch, M.; Bieker-Walz, L.; Erdmann, J.; Flötteröd, Y.P.; Hilbrich, R.; Lücken, L.; Rummel,
J.; Wagner, P.; Wießner, E. Microscopic Traffic Simulation using SUMO. In Proceedings of the 21st IEEE
International Conference on Intelligent Transportation Systems, Maui, HI, USA, 4–7 November 2018.
19.
Thériault, M.; Vandersmissen, M.H.; Lee-Gosselin, M.; Leroux, D.F. Modelling Commuter Trip Length and
Duration Within GIS: Application to an OD Survey. J. Geogr. Inf. Decis. Anal. 1999,3, 41–55.
20.
Carpenter, C.; Fowler, M.; Adler, T.J. Generating Route-Specific Origin—Destination Tables Using Bluetooth
Technology. Transp. Res. Rec. J. Transp. Res. Board 2012,2308, 96–102. doi:10.3141/2308-10.
21.
Barceló, J.; Montero, L.; Bullejos, M.; Serch, O.; Carmona, C. Dynamic OD Matrix Estimation Exploiting
Bluetooth Data in Urban Networks. In Proceedings of the 14th International Conference on Automatic
Control, Modelling & Simulation, Saint Malo, France, 2–4 April 2012; World Scientific and Engineering
Academy and Society (WSEAS): Stevens Point, WI, USA, 2012; ACMIN’12; pp. 116–121.
22.
Caceres, N.; Wideberg, J.; Benitez, F. Deriving origin—Destination data from a mobile phone network.
IET Intell. Transp. Syst. 2007,1, 15. doi:10.1049/iet-its:20060020.
23.
Larijani, A.N.; Olteanu-Raimond, A.M.; Perret, J.; Brédif, M.; Ziemlicki, C. Investigating the Mobile Phone
Data to Estimate the Origin Destination Flow and Analysis; Case Study: Paris Region. Transp. Res. Procedia
2015,6, 64–78. doi:10.1016/j.trpro.2015.03.006.

Atmosphere 2019,10, 293 13 of 14
24.
Hofer, C.; Jäger, G.; Füllsack, M. Generating Realistic Road Usage Information and Origin-Destination
Data for Traffic Simulations: Augmenting Agent-Based Models with Network Techniques. In International
Workshop on Complex Networks and Their Applications; Springer: Cham, Switzerland, 2017; pp. 1223–1233.
25.
Gipps, P. A behavioural car-following model for computer simulation. Transp. Res. Part B Methodol.
1981
,
15, 105–111. doi:10.1016/0191-2615(81)90037-0.
26.
Fontes, T.; Pereira, S.; Fernandes, P.; Bandeira, J.; Coelho, M. How to combine different microsimulation
tools to assess the environmental impacts of road traffic? Lessons and directions. Transp. Res. Part D Transp.
Environ. 2015,34, 293–306. doi:10.1016/j.trd.2014.11.012.
27.
Jiang, Y.Q.; Ma, P.J.; Zhou, S.G. Macroscopic modeling approach to estimate traffic-related emissions in
urban areas. Transp. Res. Part D Transp. Environ. 2018,60, 41–55. doi:10.1016/j.trd.2015.10.022.
28.
Borge, R.; de Miguel, I.; de la Paz, D.; Lumbreras, J.; Pérez, J.; Rodríguez, E. Comparison of road traffic
emission models in Madrid (Spain). Atmosp. Environ.
2012
,62, 461–471. doi:10.1016/j.atmosenv.2012.08.073.
29.
Smit, R.; Ntziachristos, L.; Boulter, P. Validation of road vehicle and traffic emission models—A review and
meta-analysis. Atmosp. Environ. 2010,44, 2943–2953. doi:10.1016/j.atmosenv.2010.05.022.
30.
Spangl, W.; Nagl, C. Jahresbericht der Luftgütemessungen in Österreich 2017; Technical Report;
Umweltbundesamt: Wien, Austria, 2018.
31.
TomTom International BV. TomTom Traffic Index. 2016. Available online: https://www.tomtom.com/en_
gb/trafficindex/ (accessed on 21 February 2019).
32.
Hofer, C.; Jäger, G.; Füllsack, M. Including traffic jam avoidance in an agent-based network model. Comput.
Soc. Netw. 2018,5, 5.
33.
Ringler, P.; Keles, D.; Fichtner, W. Agent-based modelling and simulation of smart electricity grids and
markets—A literature review. Renew. Sustain. Energy Rev. 2016,57, 205–215. doi:10.1016/j.rser.2015.12.169.
34.
Chen, L. Agent-based modeling in urban and architectural research: A brief literature review. Front. Arch.
Res. 2012,1, 166–177. doi:10.1016/j.foar.2012.03.003.
35.
Grimm, V. Pattern-Oriented Modeling of Agent-Based Complex Systems: Lessons from Ecology. Science
2005,310, 987–991. doi:10.1126/science.1116681.
36.
Macal, C.M.; North, M.J. Agent-based modeling and simulation: Desktop ABMS. In Proceedings of the 2007
Winter Simulation Conference, Washington, DC, USA, 9–12 December 2007. doi:10.1109/wsc.2007.4419592.
37.
Open Street Map. 2019. Available online: https://www.openstreetmap.org/ (accessed on 21 February 2019).
38.
Boeing, G. OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street
networks. Comput. Environ. Urban Syst. 2017,65, 126–139.
39.
Zhan, F.B.; Noon, C.E. Shortest path algorithms: An evaluation using real road networks. Transp. Sci.
1998
,
32, 65–73.
40.
Tomschy, R.; Herry, M.; Sammer, G.; Klementschitz, R.; Riegler, S.; Follmer, R.; Gruschwitz, D.; Josef, F.;
Gensasz, S.; Kirnbauer, R.; et al. Österreich unterwegs 2013/2014: Ergebnisbericht zur österreichweiten
Mobilitätserhebung Österreich unterwegs 2013/2014. 2016. Available online: https://www.bmvit.gv.at/
verkehr/gesamtverkehr/statistik/oesterreich_unterwegs/ (accessed on 21 February 2019).
41.
Pötscher, F. CO2-Monitoring 2016 der Neuzulassungen von Pkw; European Environment Agency: Copenhagen,
Denmark, 2016.
42.
Statistik Austria. Registerzählung 2011. Statistik Austria Pendlerstatistik 2011. Available online: https:
//www.statistik.at/web_de/frageboegen/registerzaehlung/index.html (accessed on 21 February 2019).
43.
Zilske, M.; Neumann, A.; Nagel, K. OpenStreetMap for Traffic Simulation. In Proceedings of the 1st
European State of the Map—OpenStreetMap conference, Vienna, Austria, 15–17 July 2011, pp. 126–134.
44.
De Palma, A.; Rochat, D. Understanding individual travel decisions: Results from a commuters survey in
Geneva. Transportation 1999,26, 263–281.
45.
Infras, A. Handbuch Emissionsfaktoren des Straßenverkehrs; Version 3.1; INFRAS AG: Bern, Switzerland, 2010.
46.
Höfler, F. Verkehrswesen-Praxis-Band 1: Verkehrsplanung; National Academies of Sciences, Engineering, and
Medicine: Washington, DC, USA, 2004.
47.
FGSV. Richtlinien für die Anlage von Straßen (RAS) Teil: Querschnitte (RAS-Q); FGSV-Verlag: Köln, Germany,
2006.

Atmosphere 2019,10, 293 14 of 14
48.
Parlament, E.; Rat, E. Verordnung (EG) Nr. 715/2007 des Europäischen Parlaments und des Rates vom
20. Juni 2007 über die Typgenehmigung von Kraftfahrzeugen hinsichtlich der Emissionen von leichten
Personenkraftwagen und Nutzfahrzeugen (Euro 5 und Euro 6) und über den Zugang zu Reparatur-und
Wartungsinformationen für Fahrzeuge. Amtsblatt der Europäischen Union
2007
,29. Available online:
https://www.eea.europa.eu/policy-documents/regulation-ec-no-715-2007 (access on 21 February 2019)
49.
Weiss, M.; Bonnel, P.; Hummel, R.; Manfredi, U.; Colombo, R.; Lanappe, G.; Le Lijour, P.; Sculati, M. Analyzing
on-Road Emissions of Light-Duty Vehicles with Portable Emission Measurement Systems (PEMS); JRC Scientific
and Technical Reports; European Commission: Ispra, Italy, 2011; Volume 24697.
50.
Salzburg, L. Semiakat Tables. 2006. Available online: https://www.salzburg.gv.at/umweltnaturwasser_
/Seiten/semikat-tabellen.aspx (accessed on 21 February 2019).
51.
Ausbreitungskarten Stickstoffdioxid—Bericht Salzburg Stadt, 2010. Available online: https://www.salzburg.
gv.at/umweltnaturwasser_/Documents/no2_karte_stadt.pdf (accessed on 21 February 2019).
52.
R., G.; Foelsche-Trummer, E.; Sommer, A.; Sperka, G. SEMIKAT-Salzburger Energie- und Emissionskataster.
2000. Available online: https://www.salzburg.gv.at/umweltnaturwasser_/Documents/pdf-semikat98.pdf
(accessed on 21 February 2019).
53.
Egbue, O.; Long, S. Barriers to widespread adoption of electric vehicles: An analysis of consumer attitudes
and perceptions. Energy Policy 2012,48, 717–729.
54.
Al-Alawi, B.M.; Bradley, T.H. Review of hybrid, plug-in hybrid, and electric vehicle market modeling studies.
Renew. Sustain. Energy Rev. 2013,21, 190–203.
55.
Zhou, Y.; Wang, M.; Hao, H.; Johnson, L.; Wang, H. Plug-in electric vehicle market penetration and incentives:
A global review. Mitig. Adapt. Strateg. Glob. Chang. 2015,20, 777–795.
56.
Sierzchula, W.; Bakker, S.; Maat, K.; van Wee, B. The influence of financial incentives and other
socio-economic factors on electric vehicle adoption. Energy Policy 2014,68, 183–194.
57.
Helveston, J.P.; Liu, Y.; Feit, E.M.; Fuchs, E.; Klampfl, E.; Michalek, J.J. Will subsidies drive electric vehicle
adoption? Measuring consumer preferences in the US and China. Transp. Res. Part A Policy Pract.
2015
,
73, 96–112.
58.
Hawkins, T.R.; Singh, B.; Majeau-Bettez, G.; Strømman, A.H. Comparative environmental life cycle
assessment of conventional and electric vehicles. J. Ind. Ecol. 2013,17, 53–64.
c
2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).