Dynamic traffic assignment: A review of

the methodological advances for

environmentally sustainable road
transportation applications
Highlights
•
This review summarizes and examines the recent methodological
advances of dynamic traffic assignment (DTA) models in
environmentally sustainable road transportation applications.

•
This review presents the systematic classifications of
emission estimation models and DTA models.
•
This review identifies some research gaps in current studies and
highlights several highly inspiring research directions.
Abstract
The fact that road transportation negatively affects the quality of the
environment and deteriorates its bearing capacity has drawn a wide range
of concerns among researchers. In order to provide more realistic traffic
data for estimations of environmental impacts, dynamic traffic
assignment (DTA) models have been adopted in transportation planning
and traffic management models concerning environmental sustainability.
This review summarizes and examines the recent methodological
advances of DTA models in environmentally sustainable road
transportation applications including traffic signal control concerning
vehicular emissions and emission pricing. A classification of emission
estimation models and their integration with DTA models are accordingly
reviewed as supplementary to the existing reviews. Finally, a variety of
future research prospects of DTA for environmentally sustainable road
transportation research are discussed. In particular, this review also
points out that at present the research about DTA models in conjunction
with noise predictive models is relatively deficient.
1. Introduction
Faced with the economic and social advances of civilization, transportation planners and
traffic engineers no longer solely focus on congestion when solving transportation
problems or optimizing transportation systems. In recent years, the fact that road
transportation negatively affects the well-being of the environment and deteriorates its
bearing capacity has drawn an enormous amount of concern. According to the United
States Environmental Protection Agency (USEPA), road transportation (automobiles,
trucks, and buses) is recognized as the greatest source of carbon monoxide (CO), nitrogen
dioxide (NOx), and sulfur dioxide (SOx) from the burning of fuel, as well as the main
contributor of reactants for producing particulate matter (PM) through chemical reactions in
the atmosphere. Numerous studies have confirmed that these vehicular emissions explicitly
contribute to various health problems including cancer, cardiovascular and respiratory
diseases, and perinatal mortality (Pearson et al., 2000, HEI 2010). Transportation is also a
significant source of greenhouse gas emissions (GHG in short, e.g., carbon dioxide (CO2)),
which leads to irreversible anthropogenic global warming. Scientists believe that the
climate changes and damage costs could be catastrophic concerning the ecological
degradation in a long run (Hansen, 2008). These cautions underscore the need for greater
efforts at enhancing environmental awarenessfrom the transportation sector and make it
imperative to monitor the relationship between transportation and the environment, as well
as to develop transportation systems that ensure economic growth, but at the same time are
sustainable and preserve the ecosystem.
In the last decade, remarkable advances in road traffic emission models have been achieved
with the rapid growth of comprehensiveness, complexity, and accuracy (see Smit et al.,
2008, Smit et al., 2009, Smit et al., 2010, Fontes et al., 2015). These advances provide
transportation planners and traffic engineers a variety of estimation tools for assessing the
effects of vehicular emissions and allow for from second-by-second emission descriptions
to yearly emission inventories. Meanwhile, an increasing number of studies combining road
transportation modeling with emission modeling emerge in the field of environmentally
sustainable road transportation research. Traffic assignment models are always a key
component of the integrated modeling.
The fundamental aim of a traffic assignment process is to reproduce the pattern of vehicular
movements based on certain behavioral rules, and then the outputs of the process are used
as the inputs to emission models. Based on the consideration of the temporal dimension,
traffic assignment models are classified into three categories: dynamic traffic assignment
(DTA), semi-dynamic traffic assignment, and static traffic assignment (STA) models
(Bliemer et al., 2017). A semi-dynamic traffic assignment model can be considered a series
of connected STA models (e.g., Nakayama and Conors, 2014). Unlike STA, a semi-
dynamic traffic assignment model has multiple time periods for route choice and allows the
residual traffic of one period to transfer to the following time periods. A semi-dynamic
traffic assignment model often considers only a single time step for network loading (i.e.,
flow propagation through the network) within each route choice period. This contrasts to a
DTA model in which within each time period for route choice, there exist (smaller) time
steps for network loading (Bliemer et al., 2017). Note that there are quasi-dynamic models,
but they are in fact static capacity constrained STA models (e.g., Bifulco and Crisalli,
1998, Smith, 2013, Bliemer et al., 2014) that are more realistic, mainly regarding adopting
hard capacity constrained and having queues (albeit point queues) to model congestion,
compared with classical capacity restrained type STA models with the use of Bureau of
Public Roads functions for their costs.
With the merit of computational efficiency and mathematical simplicity for encapsulation
in an analytical framework, STA models have been favored by policymakers for strategic
transportation planning among the three classes of models. STA models are also commonly
adopted for the design and management of environmentally sustainable road networks. In
the review by Szeto et al. (2012), the applications of STA models to the design and
management of environmentally sustainable road networks are well summarized. However,
the limitations of STA models are widely recognized, which is the incapability to model
traffic dynamics (e.g., queue spillback, and speed variations). It has been scientifically
proven that traffic dynamics are closely related to vehicular emissions. For example, faster
accelerations and stop-and-go conditions tend to increase emission rates (Rakha and Ahn,
2004, Barth and Boriboonsomsin, 2008, Litman, 2015). DTA models are acknowledged to
be able to capture more realistic traffic flow characteristics such as shock waves, expansion
waves, and queue spillback (Szeto and Lo, 2006), which shows high potential for
improving the accuracy of emission estimations. Hence, the implementations of DTA
models in solving transportation problems or optimizing transportation systems with
environmental considerations are beginning to receive greater attention. However, at the
time of this writing, no review on the methodological advancements for solving these
problems or optimizing the systems can be found.
In this paper, a review is given on the methodological advancements of the incorporation
of environmental sustainability concerns into various road transportation problems,
focusing on the use of DTA models. We aim to discuss and summarize the significance of
the implementation, the suitability, and the feasibility of the DTA models integrated with
emission models, as well as the trade-off between the computational complexity and
accuracy of integrated models, by surveying and categorizing a variety of successful
applications in the literature. Great attention is also given to the identification of the
potential research directions in this rapidly developing research field.
The main contributions of this paper are as follows.

This paper provides a critical and comprehensive literature review of the

methodological advances for environmentally sustainable road transportation
applications, focusing on the use of DTA models.

This review presents a systematic classification of emission estimation models and

their integration with DTA models, providing guidance for the future model
development of dynamic and emerging environmentally sustainable transportation
applications.
•

This review identifies some research gaps in current studies and highlights several
highly inspiring research directions with practical relevance for the future studies of
dynamic and environmentally sustainable transportation research.
The remainder of this paper is organized as follows. The second section provides a brief
review of the commonly used road traffic emission models, with a focus on their
applications integrated with traffic assignment models. Section 3 reviews the fundamentals
of DTA models from the two major components, namely the network loading model and
the travel choice principle, and discusses the recent advances in incorporating
environmental concerns into these components. Section 4 discusses the integrated studies of
DTA and emission models from two main categories: traffic signal timing optimization and
emission road pricing. Finally, some potential future research directions are suggested in
the last section.
2. Vehicle emission modeling
Szeto et al. (2012) pointed out that environmental impacts from road transportation
primarily comprise emissions and noise emitted. To mitigate the impacts, various measures
have been implemented, including actions on transportation infrastructure and traffic
management measures. To quantify the impacts and benefits of these measures, the
integration studies of noise and DTA models have not been prevailingly carried out
compared to those of emissions. Hence, this review mainly focuses on the limited
integrated studies of emissions and DTA models.
The effort to specify the issues of automobile emissions can be traced back to 1960s (Soltau
and Campbell, 1968, Jackson et al., 1969). A vast variety of gaseous pollutants and fine
particulate matter from fuel combustion are emitted from vehicles on roads, among
which carbon monoxide (CO), nitrogen oxides (NOx), and particulate matter (PM) have
been listed as criteria air pollutants by most environmental
protection agencies/departments (e.g., U.S., EU, Australia, and Hong Kong) because of
their severe adverse effects on humans and the ecosystem. As indicated
by Cappiello (2002) and Boulter et al. (2007), the levels of emissions are influenced by
many variables and parameters, including vehicle technical specifications, vehicle status,
vehicle operating conditions, and external environment conditions (such as air conditions
and road characteristics). It is therefore helpful to present a summary of the emission
models in this section for the discussion of their integration with DTA models in the studies
to be reviewed in the later sections.
Emission models provide an emission estimate by multiplying a determinative emission
factor expressing the mass of pollutants emitted per unit distance, time, or mass fuel burnt
by the corresponding vehicle activity data (i.e., vehicle kilometers or miles travelled (VKT
or VMT), total time spent in particular driving conditions, or fuel consumption) (Smit et al.,
2010). Emission models are becoming more sophisticated, reflected in the level of
resolution, the level of details for input data, the extent of incorporating driving behaviors,
etc. (Smit et al., 2009). Researchers have actively contributed to the literature regarding the
effects of the increased model complexity on accuracy, as well as recommendations on
different applications/scales through meta-analysis and classification (e.g., Rakha et al.,
2004, Mellios et al., 2006, Smit et al., 2010). Although the classification framework varies
among studies (e.g., Cappiello, 2002, Smit, 2006, Boulter et al., 2007, Smit et al.,
2008, Smit et al., 2009, Smit et al., 2010, Wismans et al., 2011b, Szeto et al., 2012, Demir
et al., 2014, Fontes et al., 2015), two key factors are spotted. One is the level of aggregation
over space and time and another is the driver behavior aggregation in the modeling process.
To connect emissions to congestion more directly and provide appropriate suggestions on
integrated implementations, in this review, the taxonomy used for traffic assignment
models, namely static and dynamic modeling (Cappiello, 2002), is adopted for categorizing
the commonly applied emission models. Table 1 summarizes the main characteristics of
these models.
--------img------------

2.1. Static emission models

According to Cappiello (2002), for the development of this type of emission model, the
exhaust emissions are measured as the total generated during the driving cycle in a
dynamometer test. As discussed before, to assess the environmental impacts of road
traffic using these static emission models, vehicle activity data on the aggregate level, such
as national statistics (e.g., public road and street mileage), field measurements (e.g., traffic
flow data gathered using video cameras), or traffic volume predicted from STA models, are
mostly adequate when a high resolution is not required. Although these models cannot give
accurate emission estimates, they are commonly used for transportation planning purposes
due to their relative simplicity. Following the classification framework from Smit et al.
(2008), these static models are further categorized into the following types with increasing
complexity:
2.1.1. Aggregated models
Examples of this type of model are the UK National Atmospheric
Emissions Inventory (NAEI, 2012) and the energy workbook for transport (AGO, 2006).
The resolution of this type of model in both spatial and temporal dimensions is the lowest.
The emission factors are calibrated by a number of measurements over given driving cycles
and are usually in terms of mass of pollutant per vehicle per unit distance (g/VKT or
g/VMT) or mass of pollutant per unit of fuel consumed (g/L). A constant emission factor is
used to estimate the emissions from a particular type of vehicle and a general type of
driving without an explicit definition. Hence, the estimation requires inputs only on
network-wide fleet composition and VKT (or VMT or fuel consumption).
Little information on vehicle operation is taken into account by these models, let alone
traffic dynamics. Owing to the simplicity of the emission factors, these models are easy to
apply and most suitable for estimating large-scale emissions inventories, such as national or
regional emission inventories, on a strategic level (Cappiello, 2002, Wismans et al., 2011b).
Indeed, a number of integrated studies on a strategic level (e.g., network design problem)
have implemented this type of model in combination with STA models in the early stage
(e.g., Tzeng and Chen, 1993, Nagurney, 2000a, Nagurney, 2000b, Qiu and Chen, 2007, Jia
et al., 2009; and Chen and Xu, 2012).
2.1.2. Average-speed models
Examples of this type of model include COPERT (EEA, 2000), MOBILE (USEPA, 2002),
and EMFAC (The California Air Resources Board, CARB, 2014). This type of model is
based on the principle that the average emission factor varies according to the average
speed during a trip (Boulter et al., 2007). In this type of emission model, the emission
factor (in g/VKT or g/VMT) for each category of vehicle is a continuous function of
average speed, sometimes corrected with different factors taking account of different
conditions. In essence, the inputs are trip-based or road section-based average speed and
VKT (or VMT), which are relatively easy to obtain from static traffic assignment models or
field measurements.
Since these models are developed based on actual driving pattern data with vehicle average
speed being the sole descriptor of driving patterns, congestion is implicitly accounted for in
these models on different levels (Smit et al., 2008). However, the average speed is not an
adequate congestion indicator (e.g., both a free-flow condition of an arterial and a stop-and-
go situation on a freeway give the same average speed), which leads to significant errors in
estimations of emission factors. Nevertheless, most of them, such as the estimation model
as part of the TRANSYT-7F programme developed by Penic and
Upchurch (1992) and Wallace et al. (1998), are usually algorithmically tractable and can be
solved efficiently. With appropriate interfacing with STA models, their estimations can be
obtained down to the link level, which enables wider applications in planning, such as
tolling design (e.g., Yin and Lawphongpanich, 2006, Li et al., 2012). However, the
emissions are modeled to occur uniformly along the length of the link without further
differentiation under various traffic conditions, which is deemed insufficiently realistic.
2.1.3. Traffic situation models
Examples of this type of model include Handbook Emission Factors for Road
Transport (HBEFA) (INFRAS, 2010) and Assessment and Reliability of Transport
Emission Models and Inventory Systems (ARTEMIS) (Boulter and McCrae, 2007). This
type of model advances in terms of capturing cycle dynamics by correlating the average
emission rates with various driving cycle parameters, which in turn are referenced to
specific traffic situations. These models define traffic situations as a combination of several
dimensions, including area type, road type, speed limit, and level of service. For different
vehicle categories, the computation of the emission factor (in g/VKT or g/VMT) for each
traffic situation relies on representative speed-time curves. Hence, this type of model
requires qualitative variables (e.g., congestion level, road type, and area type) or
quantitative traffic situation variables (e.g., average speed, traffic volume, and link length)
that represent traffic situations, as well as the VKT (or VMT) data per driving situation as
inputs.
Either through qualitative (textual) or quantitative descriptions, the definitions of traffic
situations may result in inconsistencies in interpretation as Boulter et al. (2007) pointed out.
Despite this shortcoming, traffic situation models tend to be best suited to local applications
(e.g., emission estimations for individual links, street atmospheric dispersion modeling,
etc.) among all types of static emission models, thanks to their capability of providing
emission estimations with more detailed traffic characteristic considerations on the link
level.
2.2. Dynamic emission models
In contrast to the static modeling approach, the dynamic modeling approach measures
emissions from driving cycles in a dynamometer test continuously (typically second-by-
second). Hence, other than the standard traffic related inputs on links such as VKT/VMT
and fleet composition, the operational conditions of a vehicle are taken into account to a
high resolution when emissions are recorded simultaneously. As pointed out
by Cappiello (2002), with continuous measurements, both modal and instantaneous
analysis and modeling are adopted for developing dynamic emission models, based
respectively on more aggregated modal variables or on instantaneous vehicle kinematic
variables (Boulter et al., 2007). Hence, most classification frameworks (e.g., Boulter et al.,
2007, Wismans et al., 2011b) have further divided them into three groups: regression-based
models, modal models, and instantaneous models.
2.2.1. Regression-based models
Examples include VERSIT+ (Smit et al., 2007) and the Virginia Tech Microscopic Energy
and Emission Model (VT-Micro1) (Ahn et al., 2002). This type of model employs
regression approaches to modeling emissions with a substantial set of descriptive variables
characterizing driving cycles. Detailed driving pattern data at high temporal resolution
(e.g., instantaneous speed, acceleration levels, or speed-time profile data) are required for
obtaining average emission factors (g/s, g/VMT or g/VKT) based on the regression
equations, in which the emission factors can be considered either discrete or continuous,
depending on whether categorical variables are used as inputs. This type of model can
estimate emissions on a highly detailed level in conjunction with microscopic data from
GPS equipment. This type of model offers a great amount of freedom to implement in
conjunction with DTA models. However, due to the use of a large number of explanatory
variables, this type of model is susceptible to over-fitting calibration data (Cappiello et al.,
2002).
2.2.2. Modal models
An example of this type of model is the Mobile Emissions Assessment System for Urban
and Regional Evaluation (MEASURE) (Fomunung et al., 1999). In this type of model, an
emission factor (g/s, g/VKT, g/VMT) for a certain vehicle type is expressed as a discrete
function of different modes of vehicle operating conditions (e.g., starts, idle, acceleration,
deceleration, cruise, etc.). In principle, both vehicle characteristics (e.g., odometer readings,
catalytic converter type, fuel injection type, etc.) and cycle characteristics (e.g., the
percentage of the cycle exceeding various thresholds, acceleration, deceleration, cruise
speed, etc.) are required for the data sets of emissions tests for model development. These
models typically required detailed information on vehicle movements (e.g., instantaneous
data on speed, acceleration, and road gradient) as inputs to determine the emission factor
(Smit et al., 2010). The total emission during a trip or a road section is computed by
weighting each modal emission factor by the time spent or distance traveled in the
corresponding mode. Similar to regression-based models, this type of model is also capable
of modeling the effects of congestion on emissions, provided that congestion is reflected in
the input driving patterns. Traffic dynamics are incorporated indirectly by distinguishing
various operating modes. Compared with situation-based models, this type of model
requires more detailed vehicle characteristics and operations in order to produce more
accurate emission estimates.
2.2.3. Instantaneous models
Examples include PHEM (Passenger car and Heavy duty Emission Model)
(Hausberger et al., 2003), Comprehensive Modal Emission Model (CMEM2) (Scora and
Barth, 2006), and Motor Vehicle Emission Simulator (MOVES3) (The United States
Environmental Protection Agency, 2014). This type of model is indeed a more detailed type
of modal model (Boulter et al., 2007). Emission factors (g/s) are estimated from the
measurements of a driving cycle or an engine test at the highest resolution (typically one
second) using speed and acceleration, or by using the description of the engine power
requirement. Since these models inherently and directly take the dynamics of driving cycles
into account, the estimation of emissions is the finest in theory. Usually in these models,
various components model the different processes in the vehicle related to emissions,
requiring both dynamic operating variables (e.g., second-by-second speed, road grade, and
vehicle accessory use) and static parameters (to characterize the vehicle tailpipe emissions
for the appropriate vehicle/technology category) as inputs (Scora and Barth, 2006).
However, this type of model is relatively difficult to calibrate and more data-intensive
compared with regression-based and modal models. The merits of instantaneous models
can only be exploited with the detailed and precise measurements of vehicle characteristics,
operations, and locations as inputs, which leads to the difficulty in the combined use with
DTA models because of data availability.
3. DTA for integration with emission models
Traffic assignment models are crucial for traffic flow and travel timeforecasting in long-
term transportation planning and project appraisal, as well as in short-term traffic operation
management and control. These models determine the flow on each link and capture the
interaction between demand and supply. If a (departure) time dimension is introduced to the
assignment module (and traffic flow module), the formulation can be generalized from an
STA to a DTA context.
In the literature, the outputs of STA models, including traffic volume, traffic composition,
and the level of service (i.e., volume-to-capacity ratios, average travel speeds), etc., are
widely used to estimate traffic emissions (see Szeto et al., 2012). However, the limitations
associated with STA models are widely recognized as its ignorance of realistic traffic
dynamics (i.e., dynamic queuing and spillback phenomena), as well as the spatial and
temporal vehicular interactions. For example, simplified congestion functions (i.e., Bureau
of Public Roads travel time function) are mostly used in classical capacity restrained type
STA models to relate average travel times to traffic flows on a link level but do not
consider speed variations and stop-and-go traffic conditions, particularly in congested
networks. Given the time-varying nature of traffic flow, to truly assess traffic emissions, a
DTA model should be considered. Using the outputs from DTA models not only deliver
greater compatibility with emission estimation models but also have high potential in
improving emission estimates with higher temporal and spatial resolutions. As indicated
in Section 2, noise estimation models are not that widely combined with DTA models
compared to emission models. Hence, this review mainly focuses on the limited integration
of emission models and DTA models.
Both static and dynamic traffic assignment models consist of two main components,
namely the travel choice principle and the network loading model. In the past decade,
extensive comprehensive reviews on DTA models or their components have been
conducted (e.g., Hoogendoorn and Bovy, 2001, Peeta and Ziliaskopoulos, 2001, Szeto and
Lo, 2005, Szeto and Lo, 2006, Jeihani, 2010, Mun, 2007, Rakha and Tawfik, 2009, Szeto
and Wong, 2012, Bliemer et al., 2017). For the detailed mathematical formulations,
formulation and solution approaches, classifications, and implementations of DTA models
or components, readers are welcome to consult the wealth of literature listed above. This
section mainly focuses on the compatibility of DTA models with emission models, through
discussion on the two major components, i.e., the network loading model and the travel
choice principle.
3.1. Network loading model
The network loading model depicts how traffic propagates inside a road network and hence
governs the network performance in terms of travel time. It is sometimes called a traffic
flow model or component and can be further divided into two sub-models: link and node
models. The outputs of the network loading model in DTA heavily affected by the types of
link model and node model chosen. Based on the level-of-detail with the representation of
traffic, network loading models are typically classified into macroscopic, mesoscopic, and
microscopic (Hoogendoorn and Bovy, 2001, Rakha and Tawfik, 2009). A summary of
commonly utilized network loading models in the literature is provided in Table 2. Note
that the traffic flow simulation software can be regarded as a network loading model of
DTA (Szeto and Wong, 2012).

--table2—

Macroscopic traffic flow models describe traffic from the viewpoint of the collective
vehicular flow, using macroscopic variables (e.g., link flows). Starting from single link
loading, the Vickrey model (Vickrey, 1969), exit-function-based traffic flow model (M–N
model) (Merchant and Nemhauser, 1978), and performance function based network loading
model (Janson, 1991, Ran and Boyce, 1996; Chen and Hsueh, 1998) have been used to
capture congestion on a link in DTA models in the early stage, but yet they fail to describe
some fundamental traffic dynamics such as queue spillbacks. To address this issue, a first-
order kinematic wave traffic flow model, the LWR model, was captured in DTA models
(e.g., Kuwahara and Akamatsu, 2001, Lo and Szeto, 2002, Friesz et al., 2013b, Long et al.,
2015) and solved by Daganzo, 1994, Daganzo, 1995) solution scheme (i.e., Cell
Transmission Model (CTM)), Newell's (1993)solution scheme, or the Link Transmission
Model (LTM) solution procedure (Yperman, 2007). More advanced solution schemes have
been developed based on LTM, including the General Link Transmission Model (GLTM)
solution procedure (Gentile, 2010) and the event-based solution procedures for continuous
time LTM for triangular fundamental diagrams (Raadsen et al., 2016) and nonlinear
fundamental diagrams (Bliemer et al., 2018), respectively. Recently, the analytical
reformulations of the LTM and the Vickrey model are also emerging to allow the
development of novel solution approaches to solve DTA models with macroscopic network
loading models (e.g., Long et al., 2015, Han et al., 2013a, Han et al., 2013b, and 2013c).
Although the first-order kinematic wave models have been used extensively in DTA
models compared to second- or higher order kinematic wave models owing to the merits of
the analytical tractability of the first-order models, they are not capable of explicitly
capturing complex waves of vehicular traffic (e.g., start-stop waves, oscillatory congested
traffic, and capacity drop or traffic hysteresis patterns), and hence lead to the inaccuracy in
the estimates of higher-order traffic variables (e.g., acceleration/deceleration and
perturbation). To overcome these limitations, higher-order traffic flow models have been
proposed, including the Payne–Whitham model (Payne, 1971, Payne, 1979; Whitham,
1974) and its further development by Aw and Rascle (2000) and Zhang (2002). In contrast
to the one-to-one relationship between traffic flow and density in first-order kinematic
wave models, they explicitly describe the scattered density-flow pairs using a set-valued
fundamental diagram. More recently, the PTM has gained increasing popularity for their
capability of representing complex waves while keeping the LWR structure for light traffic
(Piccoli et al., 2015). One of the significant advantages of the second-order traffic flow
models is their capability of capturing higher-order variations in velocity and acceleration
and hence producing more accurate emission rate estimates (Piccoli et al., 2015).
While link models compute dynamic traffic flow propagation along road segments, node
models determine flows over the intersections and affect the congestion dynamics in the
adjacent links. Recent advances on macroscopic node models have been made to pose
formulation requirements, including Tampère et al. (2011), Flötteröd and
Rohde (2011), Gibb (2011), and Corthout et al. (2012). For an overview of the state-of-the-
art macroscopic node models, interested readers are referred to the study of Smits et al.
(2015).
In contrast to macroscopic traffic flow models which deal with the collective traffic flow
dynamics, microscopic traffic flow models track each vehicle at a high temporal resolution.
A microscopic traffic flow model usually consists of sub-models that describe assorted
human driver behavior, including gap-acceptance, speed adaptation, lane-changing,
overtakes, and car-following (Olstam and Tapani, 2004). Most previous research on
microscopic traffic flow theory has been focused on the car-following logic, among which
the introduction of Gazis–Herman–Rothery models (Gazis et al., 1961), safety-distance
models (Kometani and Sasaki, 1959), and psycho-physical car-following models
(Wiedemann, 1974) have laid the theoretical foundation for the continuing development of
micro-simulation models or software packages. Several commonly used micro-simulation
models are provided in Table 2, such as AIMSUN (Barcelo et al., 1997) embedding a
safety-distance type model, MITSIMLab with a Gazis–Herman–Rothery-type car-following
model, and VISSIM incorporating a psycho-physical model. Microscopic traffic flow
models can provide outputs in terms of the position, instantaneous speed, and acceleration
of each vehicle at each time step. For a comprehensive review of the existing micro-
simulation models, please refer to Algers et al. (1997) and Olstam and Tapani (2004).
Mesoscopic traffic flow models are hybrids of macroscopic and microscopic traffic flow
models, which describe the traffic entities at a high level of detail, while their behavior and
interactions are traced at a lower level of detail. The modeling approaches can vary from
model to model. For example, in the cluster models under this scope (e.g., CONTRAM),
traffic flow propagation is based on the macroscopic traffic flow theory wherein the traffic
is represented by small packets or groups of vehicles rather than individual vehicles.
Alternatively, individual traffic movements are modeled in some other mesoscopic traffic
flow models (e.g., DYNASMART, MATSIM) and the behavior of vehicles traversing
roadways is determined using a macroscopic speed/density function. Mesoscopic traffic
flow models achieve a balance between the computational cost and the simulation
accuracy.

To conclude, non-microscopic traffic flow models always require less computational effort
and are relatively easier to calibrate. Though they do not allow for a detailed representation
of traffic behavior as microscopic traffic flow models do, such as lane changing, they can
capture many traffic flow phenomena which are important for emission estimations.
Therefore, macroscopic or mesoscopic traffic flow models are better suited for modeling
larger networks (e.g., regional modeling) than microscopic traffic flow models
(Wismans et al., 2011b). On the other hand, microscopic traffic flow models are usually
appropriate for a local area only, as an extensive amount of computation and memory are
required for calibrating and simulating a large transportation network at a level of detail
down to individual vehicles. Nevertheless, a multi-resolution modeling approach has been
exploited by many practitioners in recent years (Zhou et al., 2015), where different types of
traffic flow models can be used to provide multiple levels of modeling details regarding
network dynamics and traveler/driver choices in order to produce satisfactory results in an
integrated fashion. For instance, macroscopic or mesoscopic traffic flow models can be
used to model the whole large network, together with microscopic traffic flow models
generating more detailed traffic flow information on a single intersection therein.
Cappiello (2002) indicates that the integration between network loading and emission
models can be achieved on various levels in term of temporal scale, spatial aggregation, and
vehicle aggregation. In general, the output of the macroscopic network loading models can
be easily fed to macroscopic emission models, so is the integration of microscopic network
loading models and microscopic emission models (Zegeye et al., 2013). This indicates that
the choice of network loading models directly influences the selection of the interfacing
emission models. Moreover, the purpose of integration, the requirements of accuracy, the
constraints of data availability, as well as the computational time requirement are crucial
factors for both model selections. While featuring a good compatibility between emission
and network loading models, the integration always aims to achieve a balanced trade-off
between accuracy and computational complexity.
3.2. Travel choice principle
A travel choice principle models traveler's propensity to travel, and if so, how they select
their routes, departure times, modes or destinations (Szeto and Lo, 2006). Typically, the
travel choice principle of DTA is the dynamic extension of the route choice principle of
STA, including the Dynamic User Optimal (DUO) principle (Friesz et al., 1989) extended
from Wardrop's (1952)first principle, the Dynamic System Optimal (DSO) principle
(Merchant and Nemhauser, 1978) extended from Wardrop's second principle or the
Dynamic Stochastic User Optimal (DSUO) principle (Ran and Boyce, 1996) extended from
the stochastic extension of Wardrop's (1952) first principle, etc. These principles assume
that travelers select their routes and/or departure times to minimize their individual
actual/marginal/perceived travel time/cost respectively, while the environmental
impacts are typically not considered in these principles.
Although a list of choices can be included in a travel choice principle of DTA, the travel
choice principles used in DTA with environmental objectives in the literature only consider
route choice, as listed in Table 3. From this table, we can see that the routing principles
adopted are not restricted to Wardrop's principles. It is because under Wardrop's routing
principles, the emissions produced may not be the most favorable to the surrounding
environment, which has been demonstrated in a variety of studies using static approaches
(e.g., Rilett and Benedek, 1994, Benedek and Rilett, 1998, Sugawara and Niemeier,
2002, Chen and Yang, 2012). Although the theoretical finding can be derived from STA
models, it can often be dangerous to generalize the conclusions from STA models to actual
traffic systems due to the ignorance of traffic dynamics. As Rilett and
Benedek (1994) suggested, it is of vital importance that traffic assignment models produce
comprehensive and accurate outputs for evaluating the environmental impacts of traffic
assignment. To remedy the limitations of STA models, recent efforts have been contributed
to identifying and analyzing the environmental impacts of route choice behavior using DTA
models with a higher spatial and temporal resolution, as listed in Table 3.
---Table 3.
Ahn and Rakha (2008) examined the environmental impacts of route choice behavior using
second-by-second GPS commute data and a micro-simulation study limited to
a highway corridor and an arterial route. The simulation results illustrate that the faster
highway route choice is not always the best from an environmental and energy
consumption perspective. It is suggested that an emission- and energy-optimized traffic
assignment can significantly improve emissions over the standard user equilibrium and
system optimum assignment, which urges environmental concerns to be incorporated in
traffic routing. Moreover, according to the comparison of the estimates from a static
emission model (MOBILE6) and dynamic emission models (VT-micro (ORNL) model and
CMEM model), it is demonstrated that macroscopic emission estimation tools can produce
erroneous conclusions due to their ignorance of transient vehicle behavior along a route. It
is also worth noting that emissions are sensitive to changes in a vehicle's speed and
acceleration profiles. Given the fact that significant increases in emissions occur under high
engine load conditions (e.g., during acceleration), minimizing high-emitting driving
behavior has potential in improving air quality. Recently, Ahn and Rakha (2013) further
investigated the trade-off between emissions and travel time in daily route choices by more
comprehensive microsimulations. The transient vehicle behavior is taken into account in
this study. The results demonstrate the significant impacts of emission-minimized routing
in reducing network-wide fuel consumption and emission levels. The analysis results
further indicate that the reduction is more significant in grid networks compared to freeway
corridor networks, and rises with an increase in the level of network congestion and market
penetration of such routing vehicles. Guo et al. (2013) also simulated and approximated a
“green user equilibrium” (i.e., emission- or fuel consumption-minimized routing) and
conducted a realistic assessment of its impacts. They further proposed a “targeted market
penetration” strategy wherein travelers with the greatest potential for emission reductions
are chosen to follow the green route assignment. Such a strategy is demonstrated to bring
more environmental benefits than the random market penetration strategy, which assumes
that the travelers who choose to follow the green route assignment are randomly distributed
throughout the study area. Moreover, the heterogeneity in emission reduction is spotted
among different types of vehicles for each pollutant. All these studies concluded that the
flow pattern based on minimizing route travel time does not necessarily correspond to the
flow pattern based on the least total emissions or the least fuel consumption in the traffic
assignment, based on field data or simulation results. They also demonstrated that there are
tangible environmental benefits associated with emission-minimized routing.
On the other hand, the consideration of emissions in route choice decisions has been
analytically modeled as well. Aziz and Ukkusuri (2012) integrated an emission-based
component into a DTA framework, leading to a model for system optimal assignment with
regard to minimization of a weighted sum of total system travel time cost and CO emission
cost, and a model for minimizing the total system CO emissions only. Based on the CTM,
the resultant models with the incorporation of an emission model are formulated as two
nonlinear and nonconvex mathematical programs and further approximated for their
solvability with the CPLEX solver. The study demonstrated the significant effect of
integrating an emission-based objective into route choice on reducing emissions.
Nevertheless, there are certain model limitations such as vehicle holding leading to possible
misestimation of travel time and the ignorance of fleet heterogeneity of the
CTM. Long et al. (2016) indicated that the approximated quadratic programming models
in Aziz and Ukkusuri (2012) can be very inaccurate for emission estimations except for a
very congested traffic state. Adopting the LTM as the network loading model, Long et al.
(2016) also solved the system optimal traffic assignment problems that minimize total
system travel time, total system CO emissions, and both. Coupling with the same average
speed based emission function, the two system optimal problems with emission
considerations are both formulated as mixed integer linear programming problems
(MILPs). Although both modeling accuracy and computational efficiency are improved in
this study, their DTA models with environmental objectives are still for single destination
network applications only. Moreover, the existing analytical models can be extended where
emissions other than CO (e.g., NOx, HC, etc.) are considered, since different pollutants
possess varied emission characteristics, leading to different emission functions.
As reflected in Table 3, the studies on DTA models with environmental considerations are
rather limited. It may be because it is hard to solve the resultant models, including
nonlinear and nonconvex programs, MILPs, and those involved complicated
microsimulations, for exact solutions quickly for large network applications. This may also
explain why the networks used in some literature are not large.
As discussed before, the environmental impacts can be greatly reduced by incorporating
emission-based objectives into route choice behavior in the traffic assignment process.
However, it is not practically easy to achieve the environmentally system optimal or
emission-minimized assignment in the real world without the implementation of traffic
or demand managementstrategies, since drivers do not consider environmental impacts to
be minimized at their own initiatives during the route selection process. Nevertheless,
system optimal assignments always provide benchmarks for the design of road pricing
strategies. The environmentally system optimal models in Aziz and
Ukkusuri (2012) and Long et al. (2016) lay the theoretical foundation for developing
emission-based pricing strategies.
In addition to the route choice principle embedded in traffic assignment models for
travelers, environment-related issues in freight routing have also attracted much attention
from the operations research field. The Pollution Routing Problem or Eco-
routing involving the minimization of pollutant emissions, which is a variant of the
traditional vehicle routing problem, has been investigated with different aspects of
operational constraints (e.g., Bektaş and Laporte, 2011, Franceschetti et al., 2013, Nie and
Li, 2013). Interested readers are welcome to consult the recent comprehensive reviews
by Lin et al., 2014, Alam and McNabola, 2014, and Demir et al. (2014) for more
information of this research field.
4. Specific dynamic transportation problems with environmental considerations
Not surprisingly, numerous studies focus on the evaluation of emission
impacts. Wismans et al. (2011b) provided a brief review of the application of microscopic
DTA models for the assessment of emission impacts among a series of externalities to road
transportation networks. Fontes et al. (2015)summarized the most recent integrated
microsimulation studies combining road traffic and emission modeling for the evaluation of
emission impacts and provided best practices on the link representation in the traffic flow
model for minimizing emission estimation errors in future applications. Indeed, the
evaluation of emission impacts is only the first step to mitigate these impacts.
In the literature, the mitigation of environmental impacts from road traffic is mostly
addressed with combating traffic congestion through two broad methods, namely supply
and demand management. The supply management strategies include increasing the
transportation network capacity through improving current transportation infrastructure or
adding new transportation infrastructure (e.g., building new roads, expanding existing
roads) and increasing the utilization of existing infrastructure through the introduction of
traffic management measures (e.g., traffic signal controls, ramp metering, and variable
message sign controls). On the other hand, the demand management strategies emphasize to
reduce travel demand or to redistribute this demand in space or in time, one of which is the
use of road pricing. Szeto et al. (2012) firstly reviewed the transportation planning and
traffic management problems that explicitly consider environmental issues, mainly within
the scope of static traffic modeling. In view of higher fidelity and accuracy of modeling
results, DTA models are playing increasingly important roles in these transportation
planning and traffic management problems. Given that no studies are on dynamic
transportation planning problems with environmental and DTA considerations at the time
of this writing, this section reviews the most current studies on various dynamic traffic
management problems, including traffic signal control and emission pricing.
4.1. Traffic signal control
It is important to note that the magnitudes of vehicle emissions are heavily affected by the
various modal activities of vehicles (Pandian et al., 2009), such as starts, stops,
acceleration, deceleration, etc. Excessive alteration of modal activities is witnessed
especially near intersections since they generally serve as the most restrictive bottlenecks in
urban networks. On the other hand, the optimization of traffic signal settings can be a
positive means of controlling vehicular emissions, owing to its inherent ability to define
vehicle movements throughout a signalized intersection.
Building upon the solid literature foundation of traditional traffic signal optimization
problems (e.g., Improta and Cantarella, 1984, Meneguzzer, 1995, Lo, 1999, Lo, 2001, Wey,
2000, Lin and Wang, 2004, Cascetta et al., 2006), the incorporation of vehicular emission
considerations into the optimization of traffic signal timings has been achieved
progressively. Robertson et al. (1980)made the debut of minimizing fuel consumption in
the optimization of traffic signals. An integrated TRANSYT computer
program incorporating a macroscopic traffic flow model, a macroscopic fuel consumption
model, and a signal optimizer is utilized in this study. Vehicle dynamics at intersections are
poorly taken into account. As pointed out by Rakha et al. (2000), such macroscopic models
do not account for time variations in speed, let alone in acceleration, which significantly
impacts the accuracy of emission evaluation. More recently, Li et al. (2004) proposed a
signal timing optimization model to reduce emissions, in which emissions are calculated
using the emission factors from the MOBILE model. Though the emission measurements
are based on dynamometer testing, the data is too aggregate for the analysis of intersection
emissions. To comprehensively evaluate the emission impacts by traffic signal timing,
studies integrating modal models or instantaneous emissions with dynamic traffic models
gradually come to the fore (see Hallmark et al., 2000, Rakha et al., 2000, Rouphail et al.,
2001, Williams and Yu, 2001, Chen and Yu, 2007, Zhang et al., 2009, Li and Shimamoto,
2011, Guo and Zhang, 2014, Kölbl et al., 2015). The simulation results from these studies
have demonstrated the effectiveness of improved signal timing or coordination in reducing
fuel consumption and vehicular emissions.
The simulation approach has also been widely applied for optimizing mobility and
minimizing environmental impacts of traffic signal timing at intersections as summarized
in Table 4. Two types of integrations are observed. The first type is the utilization of
microscopic traffic flow simulation models, from which the outputs are accurate and
consistently interfacing with modal or instantaneous emissions models (e.g., Park et al.,
2009, Stevanovic et al., 2009, Kwak et al., 2012, Ma et al., 2014, Zhou and Cai,
2014, Stevanovic et al., 2015, Mascia et al., 2017). Such an integrated study can serve as an
effective tool for the comprehensive optimization of the traffic signals for vehicular
emission mitigation from corridors, major arterials, and large-scale networks while it relies
on computationally intensive simulations. Since dynamic emission models demand the
temporal and spatial resolution of data inputs, these integrated optimization problems are
mostly tackled by simulation-based heuristics, e.g., simulation-based genetic algorithm,
which incorporates traffic simulations during the functional evaluation of heuristics.
Table 4. Dynamic traffic signal optimization studies considering emissions.
To achieve a trade-off between computational complexity and accuracy, the second type
aims at cross-resolution integration (e.g., analytical dynamic traffic flow models coupling
microscopic emission models). However, the interfacing process is not straightforward,
which involves the construction of speed profiles and vehicle trajectories (e.g., Lv et al.,
2013), or the aggregation of individual vehicle behavior (e.g., Lin et al., 2013). Such
aggregation or approximation is carried out on limited datasets with simplified
assumptions, whose fidelity has yet to validate. Recently, Osorio and Nanduri (2015a,b)
developed a metamodel (surrogate model), which combines simulated travel time and
emissions with their analytical approximations. Their model has been successfully applied
to a real network problem and solved within a tight computational budget.
In spite of the greater acceptability among real-life deployment, simulations are time-
consuming and the results cannot be easily extended to other cases. In contrast, an
analytical-based approach presents more value in a prescriptive view, since theoretical
insights can be derived and solution properties can be determined. One related study was
by Khalighi and Christofa (2015), who investigated the fundamental problem of green time
optimization towards emission minimization through a combination of previously
introduced analytical models based on traffic flow theory. The emissions are estimated by
modal emission rates with regard to the estimated time spent in each operating mode, which
is analytically derived from the time-space/queuing diagram and is a function of demand,
vehicle arrival times, saturation flow, and signal control parameters. Though computation
times for emission estimations is significantly lowered and sufficient for real-time
operations, such an application is limited to every single cycle of an isolated intersection.
Another analytical-based dynamic signal control study was undertaken by Han et al.
(2016). They developed a mixed integer mathematical programmingapproach to tackle the
significant difficulty and burdens in solution computation caused by nonlinear and
nonconvex nature of emission-related objectives/constraint. Their approach relies on the
development of the macroscopic relationship between the aggregate emission rate and the
vehicle occupancy on the same link. This relationship is obtained based on both regression
analysis and extensive numerical simulations. The relationship is then approximated by
piecewise affine functions and the associated uncertainties are handled explicitly using
robust optimization techniques so that the resultant problem can be formulated as a mixed
integer linear program and solved to global optimality efficiently. Compared to the
traditional mixed integer nonlinear program, this robust optimization formulation for signal
control admits abundant theories for analysis as well as effective and well-developed
solution schemes. Nevertheless, such macroscopic relationship may be not adequate for an
accurate emission estimation. Moreover, extensive numerical simulations or empirical data
are required to calibrate such a reduced model, which can be time-consuming or expensive
to do so.
4.2. Emission pricing
In dynamic route choice modeling within the context of emission considerations, a
significant reduction in vehicular emissions is demonstrated by the studies reviewed
in Section 3.2. However, it is practically not easy to achieve this reduction in reality
without the implementation of traffic or demand management strategies, because drivers do
not consider environmental impact minimization at their own initiatives during route
selection. One way to deal with this is to internalize the environmental impacts or the
corresponding costs by emission pricing.
Emission pricing can be analyzed through static and dynamic modelingapproaches.
According to Ma et al. (2017), albeit the dynamic modeling approach is more complex and
computationally demanding than the static counterpart, it is not deniable that time-varying
tolls can be implemented to achieve a better network performance and more realistic travel
behavior is captured by considering traffic dynamics in the modeling framework.
Moreover, in order to provide more accurate data inputs for the estimation of vehicular
emissions, the dynamic modeling approach tends to be a better choice.
Compared with dynamic congestion tolling studies (see Yao et al., 2012 and Ma et al.,
2017 for a detailed survey), there were limited dynamic emission pricing studies or
dynamic road pricing incorporating emissions in the past two decades. Table 5 lists the
literature on the dynamic road pricing incorporating emissions.
Table 5. Dynamic road pricing studies considering emissions.

Among the studies in Table 5, only two studies adopted the simulation
approach. Kickhöfer and Kern (2016) adopted the simulation-based approach for evaluating
the policy of emission cost internalization for a real-world scenario. An agent-based
microsimulation framework, i.e., MATSim, with a heterogeneous population consideration
is combined with a traffic situation model, HBEFA. The simulated average emission costs
per vehicle kilometer under the first-best emission toll policy are consistent with the
estimated average emission costs per vehicle kilometer in the literature. However, neither
the emission tolls nor the estimates adequately reflect marginal costs with respect to the
damage to human health, since the differentiation of pollution concentration among
exposed populations is not accounted for. Hence, Kickhöfer and Nagel (2015) further
incorporated activity location density in their toll calculations in order to monetize the
effects on human health and investigated the impacts of such emission toll on emission
levels, exposure cost reductions, and system welfare. A potential trade-off was observed
between the minimization of emission exposure and the traditional minimization of vehicle
kilometers traveled. Overall, this proposed emission exposure toll provides a more
comprehensive evaluation of policies that aim at reducing environmental externalities in
urban settings.
Other studies in Table 5 adopted the analytical approach. Indeed, being different from
traffic management measures (e.g., traffic signal control) in nature, road pricing studies
generally aim at providing theoretically derived insights in order to carry out a long-term
strategic decision. Hence, the analytical approach is more commonly adopted in the
emission pricing research area, as shown in Table 5. However, the studies have different
pricing focuses and choice considerations, and also use different pricing schemes, traffic
flow and emission models, and networks for illustration as summarized in this table. The
studies also give different important findings. Zhong et al. (2012) incorporated the emission
impacts through an aggregate emission factor together with a dispersion model into a
pollution charge implementation problem as a side constraint. The necessary optimality
condition for operating the transportation system with environmental constraints is
discussed for both the whole link model and the deterministic queueing model. It is found
that different externality structures of the two models result in different tolling structures to
achieve dynamic system optimal assignment. Friesz et al. (2013a) formulated their multiple
objective dynamic congestion pricing problem into a mathematical program
with equilibrium constraints. The vehicular emissions are estimated based on a macroscopic
function of speed only. The result demonstrates the ability of the optimal tolling scheme
design in reducing both total travel time and emissions. Wen and Eglese, 2016, Whitham,
1974 investigated the effects of different road toll strategies on minimizing total time and
CO2 emissions on two alternative routes through a bi-level pricing model. A macroscopic
fuel consumption function depending on speed is incorporated into the model, in
combination with a simplified simple queueing model for delay estimations together with a
free-flow traffic assumption. Their finding reveals that the multiple step toll pricing
strategy and the time-varying toll pricing strategy outperform the constant toll pricing
strategy in terms of CO2 emission mitigation. With the technology advances in intelligent
transportation systems, the easier implementation of multiple step and time-varying tolling
strategies make their finding more of practical significance. However, such results cannot
be generalized unless a more accurate traffic flow model is adopted and used on a larger
network. By extending the static emission model by Wallace et al. (1998) under a
monotonicity assumption, Ma et al. (2017)sought for and examined the first-best emission
pricing through the existence of a free flow dynamic system optimal solution where the
generalized system cost including total travel times and emissions were minimized. Their
results also show the importance of the free-flow (uncongested) traffic state in developing
network pricing schemes and managing traffic congestion, emissions/fuel consumption, and
other related issues. One may focus on the free-flow traffic state in a network and devise
pricing schemes and other management strategies to sustain such free-flow state. This not
only is beneficial to the network as a whole from the perspective of congestion and
emission control but also leads to simplified dynamic network models (since traffic
dynamics are much simpler under the free-flow traffic state) so that the optimal pricing
scheme or other strategies can be much easily calculated.
It can be concluded from Table 5 that static emission models (e.g., aggregate emission
models, average speed models) are still widely adopted to access the effect of emissions
due to their nice mathematical properties (continuity, monotonicity, etc.) in this research
area. Most of the emission pricing strategies have compromised the accuracy of emission
estimates for mathematical tractability to a large extent. However, whether such emission
estimates without capturing traffic dynamics are adequately accurate for emission pricing
design remains doubtful. Though a balance has to be stricken between mathematical
complexity and the realism of the emission function (Ma et al., 2017), more detailed
emission models for the better representation of traffic dynamics (e.g., mesoscopic
emission models) can still be considered within such an analytical modeling framework.
Moreover, the analytical models or the methodologies in general reviewed are all tested on
either hypothetic or small networks. A road toll pricing model for a larger network has not
been undertaken yet and is of vital significance for generality and applicability.
Emission tolls can be charged by toll facilities. Therefore, a number of studies have
contributed to the analysis and modeling of emission impacts on toll facilities, especially
electronic toll collection (ETC) plazas (e.g., Sisson, 1995, Klodzinski et al., 1998, Saka
et al., 2001, Bartin et al., 2007). Field measurements and simulation results have revealed
that ETC technology is an effective tool in reducing emissions in areas where toll facilities
are deployed, by decreasing the number of stop-and-go situations (Coelho et al., 2005).
Hence, ETC can be viewed as an effective tool for minimizing the deleterious effect of the
stops on pollutant emissions.
5. Future research directions
After reviewing the recent integration studies of DTA and emission models, some gaps in
the literature are discovered. This section proposes a few future research directions from the
aspects of modeling approaches, solution methods, and applications.

5.1. Modeling approaches

5.1.1. DTA modeling
For DTA modeling, three future directions can be identified. First, recent studies
(Geroliminis and Daganzo, 2008, Yildirimoglu and Geroliminis, 2014, Leclercq et al.,
2014) have demonstrated that the low-scattered Macroscopic Fundamental Diagram (MFD)
shows promising results in modeling the complex dynamics of urban congestion. MFDs
have also been incorporated into DUO assignment (Yildirimoglu and Geroliminis, 2014).
Yet its extensions to DSO with respect to system travel time or the combination of system
travel time and emissions has not been found (see Table 3). These extensions provide
useful benchmarks for developing more environmentally friendly real-time traffic
control and demand management strategies. Fruitful future analytical studies can be
performed in this direction.
Second, as reflected in Table 3, existing DTA models with environmental objectives are
developed based on the deterministic setting using classical deterministic traffic flow
models. However, in reality, demand and supplystochasticity are often observed but have
not considered in DTA with environmental objectives yet. Therefore, one direction can be
to encapsulate existing traffic flow models/network loading procedures (e.g., SCTM) or
improve existing traffic flow models/network loading procedures (e.g., LTM) to deal with
supply uncertainty and use existing stochastic programming or chance constraint techniques
to deal with demand stochasticity. Another direction is to develop modeling approaches to
deal with uncertainties associated with MFDs.
Third, though the integration between traffic simulation models and emissions models have
become a fast-evolving research area as can be seen in Table 4, a study by Song et al.
(2012) has identified the deficiency of microscopic traffic simulation models integrating
instantaneous emission models for emission estimations. The observed systemic errors are
primarily due to the fact that the VSP distribution, which is the best explanation of vehicle
emissions, is not sensitive to the common parameters used for calibrating or validating
traffic simulation models (Song et al., 2012). Therefore, future studies are needed for
improving the accuracy of characterizing traffic dynamics by further investigating the
internal mechanism of simulation models such as car-following, deceleration, and
acceleration models, so that the VSP distribution can be more accurately predicted by
integration studies.
5.1.2. Cross-resolution integrated modeling
To get a balanced trade-off between computational complexity and accuracy, the cross-
resolution integration of macroscopic traffic flow models and microscopic emission models
is worth investigating. The interfacing process requires transform of macroscopic traffic
variables (e.g., average density, average flow, and average speed) into
mesoscopic/microscopic ones, or the approximation of a macroscopic relationship between
aggregate emissions rates and macroscopic traffic quantities (e.g., link occupancy). Recent
advances in the development of macroscopic statistical emission models, speed profile
generation models, and trajectories reconstruction models have been successfully achieved
(e.g., Zegeye et al., 2013, Skabardonis et al., 2013, Shabihkhani and Gonzales, 2013, Sun
et al., 2015, Zhou et al., 2015, Chen et al., 2016, Chen et al., 1998). The computational
efficiency makes it particularly appealing in the evaluation or design of various traffic
management strategies concerning emissions. Nevertheless, there are non-negligible errors
associated with such a simplification or approximation. Future research should also be
carried out to further calibrate and validate the proposed models with more real-world
congestion and emission data.
5.1.3. Integration with pollutant concentration models
Currently, the common measure for quantifying the impacts of vehicular emissions is the
weight of pollutants emitted. The measures of the impacts of vehicular emissions can also
be spatial distribution, air quality index, personal exposure, and individual vulnerability.
The social and economic dimensions of such measures are of vital importance when it
comes to meeting air quality goals, the assessment of health risks, and addressing health
equity issues. Few reviewed studies (e.g., Zhong et al., 2012, Zhang et al., 2013) have
incorporated an air dispersion model into their work. Most studies focus on tailpipe
emissions, and either do not calculate or fail to do so at a sufficient temporospatial
granularity the impacts of the emissions on air quality and public exposure. Local pollutant
concentrations are dependent on a number of factors such as the weather, chemical
processes, non-transport sources, and canyon effects. Fontes et al. (2015) confirm the
inadequate modeling efforts found in transportation studies that focus on the pollutant
concentration and health impacts of traffic. Existing methods based on Gaussian plume
dispersion models (e.g., Zhang et al., 2013) rely on the stationarity assumption, and can
only work with annual averages of weather conditions and pollutant concentrations. Hybrid
approaches combining road link dispersion modeling with city-scale chemical transport
modeling (e.g., Stein et al., 2007) have been used to predict air quality index (e.g., Beevers
et al., 2012, Yim et al., 2013). However, emissions from non-transport sectors are not
accounted for in these studies. A potentially viable yet data-intensive approach concerns
with fusing multiple data sources and performs high-dimensional regression. This approach
has been used by Zheng et al. (2013) and should be further explored in the future.
5.1.4. Heterogeneity in modeling
In emission modeling (except the aggregate modeling approach), heterogeneity in vehicle
categories (e.g., diesel- or gasoline-driven vehicles, light-duty or heavy-duty vehicles) and
the emission patterns of pollutants are generally taken into account. However, in most
integrated studies reviewed in Section 4, the driver and vehicle populations are aggregated
to a coarse level of detail (e.g., homogeneous) for simplicity. Since the emission rates of
different types of vehicles differ significantly under the same driving condition, such
negligence of vehicle heterogeneity might result in the non-optimal design of transportation
strategies. The network loading model/procedure in these studies should be modified to, for
example, a multiclass CTM, to capture different vehicle types while taking into account the
effects of spatial queues such as queue spillback. Moreover, a multiclass LTM can be
developed to replace the multiclass CTM to speed up the computation time. For those
integrated studies with the consideration of different vehicle types, they do not take the
cleaner fuel (e.g., natural gas) vehicles into emission modeling, which is gaining higher
penetrations as a sustainable solution to reduce emissions. Such cleaner fuel vehicles
produce significantly fewer PM, CO2, and NOx emissions and their specific populations and
emission rates should be considered for more realistic modeling.
Besides different vehicle types, emissions from different road transportation modes are
worth further investigation. Intuitively, road public transit (i.e., buses) is often considered
as an environmentally friendly alternative to cars in light of its positive impact on the
reduction of emissions with a higher passenger car unit. However, recent microsimulation
studies have indicated that though bus trips are more fuel efficient than car trips on
average, busescould be as polluting as private cars on a per passenger basis under low
ridership situations (Lau et al., 2011). Similar to cars, buses generate higher emissions in
congested networks as well, due to lower speeds, higher frequency of acceleration and
deceleration events, and higher dwell times at stops (Alam and Hatzopoulou, 2014). It is
also pointed by Waraich et al. (2016)that there is a paucity of literature that quantifies and
understands the emissions generated by public transit on the disaggregate level. Hence, it is
important to develop a multi-modal DTA model to take into account the vehicle dynamics
of different public transit modes for more accurate emission estimations and analysis.
However, the integrated study of multi-modal DTA and emission models are hardly found.
Moreover, the realistic mode choice of travelers and passenger flow dynamics should be
considered in that model for improving emission estimation accuracy. The conceptual
modeling approach proposed by Chiabaut (2015) can be a basis for this future research
direction.
Besides heterogeneity in road transportation modes, heterogeneity in route choices in traffic
modeling is worth addressing. Urban transportation systems consist of different vehicle
types, each of which has different routing behaviors for various purposes. For example,
vacant taxis tend to cruise slowly to search for taxi customers whereas occupied taxis tend
to go to their customer destinations as quick as possible. Moreover, car drivers on freeways
travel at their highest speed whenever possible whereas they cruise slowly for searching
parking spaces. Their travel behavior determines travel speed which in turn determines
emissions. In fact, it has been indicated that a great part of taxi emissions are emitted from
non-service cruising for passengers and vary significantly depending on the time of
day (Yu and Peng, 2013). It is therefore important to capture their behavior into a dynamic
integrated model for an accurate prediction of emissions. Yet this line of research has not
been found. Meanwhile, different vehicles interact in the same road space and their traffic
dynamics and interactions must be duly considered. Hence, one future direction is to
develop a dynamic integrated model with the consideration of various routing behaviors
(e.g., taxis cruising for passengers, cars cruising for parking, etc.) and the traffic dynamics
of different vehicles in a network for enhancing environmental sustainability. Using MFDs
to model taxi traffic movements can be a good starting point as taxi traffic is often scattered
over a large area.
5.1.5. Developing traffic control measures towards traffic noise reduction
According to World Health Organization (2011), the noise pollution damages, mostly
from road traffic, make it second only to air pollution in terms of environmental hazards.
The dynamic approach for calculating noise levels and population exposures is of vital
importance due to the within-day dynamics of varying population densities.
Moreover, Chevallier et al. (2009)demonstrated that traffic dynamics have a substantial
impact on noise levels at signalized intersections and roundabouts. Kaddoura et al.
(2017) indicated that implementing traffic control measures, such as reduced speed levels,
turn restrictions or pricing schemes, possibly allow for a reduction in noise exposures.
However, besides the dynamic pricing approach internalizing road traffic noise damages
proposed by Kaddoura and Nagel, 2016, Kaddoura et al., 2017), noise predictive models
are not yet widely integrated with DTA models for designing traffic control measures.
Further work is expected in developing traffic control measures towards traffic noise
reduction, along with other measures of effectiveness simultaneously, such as congestion
and emissions. Such integration studies can be assisted with the critical reviews on
principal noise models by Steele, 2001, Quartieri et al., 2009, December), and Garg and
Maji (2014).
5.2. Solution methods
Because modeling environmental considerations typically result in highly nonlinear and
nonconvex objective or constraint functions (see Section 3.2and Table 4), heuristic methods
have been applied to solve these types of problems (as seen in Table 4). Besides genetic
algorithm, various meta-heuristics, such as chemical reaction optimization (Lam and
Li, 2010), have been proposed and proved to be efficient in tackling nonlinear, nonconvex
problems. The application of hybrid or multi-objective variants of such meta-heuristics to
solve nonlinear, nonconvex problems with environmental considerations may be another
future research direction. Moreover, effective parallel and distributed algorithms or
computing platforms possess the potential of cutting down the runtime and are worth to be
implemented in the future. Furthermore, running time can be shortened by developing
effective surrogate models for time-consuming DTA or simulation models. This solution
approach has only been recently considered in the field of dynamic environmental
sustainability by Osorio and Nanduri (2015a,b). There is a great potential for this line of
research.
5.3. Applications
5.3.1. Dynamic traffic management strategies
Besides dynamic traffic signal control considered in Section 4.1, there are various dynamic
traffic management strategies such as variable message sign (VMS) or dynamic route
information panel (DRIP), variable speed limits, dynamic road marking, and ramp
metering, whose effects on vehicular emissions are also significant. Currently, only limited
research has evaluated the impacts of the emissions imposed by ramp metering (e.g., Csikós
and Varga, 2012, Zegeye et al., 2013, Pasquale et al., 2015), variable speed limits
(e.g., Zegeye et al., 2010), and traffic-calming measures (e.g., Ghafghazi and Hatzopoulou,
2014, Ghafghazi and Hatzopoulou, 2015). However, the design of these dynamic traffic
management strategies has rarely found. Moreover, the average speed emission model is
mostly adopted in all these evaluation studies for simplicity. It was advocated
by Panis et al. (2006) that the analysis of the environmental impacts of any traffic
management and control policies requires not only their impact on average speeds but also
on other aspects of vehicle operation such as acceleration and deceleration. Hence, it is
anticipated that further improvements in both the performance evaluationand design of
dynamic traffic management strategies can be made by incorporating more realistic and
accurate emission models in the aforementioned applications.
5.3.2. Transport demand management
In terms of transport demand management, dynamic emission pricing has been mainly
researched and the studies have been reviewed in this paper. According to these studies,
three future research directions on emission pricing can be identified. First, as stated
in Section 4.2, the analytical models or the methodologies are all tested on either hypothetic
or small networks. More effort should be put to develop a road emission toll pricing model
for a larger, more realistic network so as to apply the model to more general situations.
Second, the reviewed models and methodologies are based on traditional traffic flow
models. MFDs have not been used in this application. In fact, MFDs have been successfully
applied to examine congestion pricing (Zheng et al., 2012, Simoni et al., 2015), parking
(Geroliminis, 2015, Zheng and Geroliminis, 2016), and signal control (Gayah et al., 2014).
MFDs should be a great potential to be used in emission pricing studies.
Third, emissions are just one of the environmental externalities but dynamic environmental
other than emission pricing studies have not been found. To internalize the environmental
externalities more accurately, other environmental externalities, such as traffic noise, must
be duly considered in the pricing strategies. Therefore, one research direction is to extend
the analysis of the current dynamic emission pricing studies to consider other
environmental externalities.
Apart from the dynamic emission pricing studies reviewed, there are a number of transport
demand management strategies playing a prominentrole in achieving vehicular emission
reduction targets. By reducing travel demand or the number of vehicles on roads, emissions
from excessive acceleration and deceleration induced by congestion can significantly
decrease. With the development of communication technology, more platforms for ride
matching are provided to facilitate dynamic and on-demand ridesharing (including
carpooling) systems, leading to a reduction of the number of vehicles on roads. These
problems involve not only the complex optimization of dynamic matching (Agatz et al.,
2012) but also travelers’ behaviors (e.g., route choice, scheduling) in traffic assignment
(Furuhata et al., 2013). It is believed that great research opportunities are provided by this
emerging area of environmentally sustainable transportation. For example, a
comprehensive evaluation of the environmental impacts of such dynamic ridesharing
involving routing and scheduling using a DTA framework has not yet been performed.
5.3.3. Autonomous vehicles
As autonomous driving capabilities (e.g., adaptive cruise control, lane keeping, and self-
parking) become increasingly widespread, the future of fully autonomous vehicles seems
more certain. This technology has the potential to reduce traffic congestion and emissions
and improve safety. Autonomous vehicle technology makes driving patterns more efficient
without the need for frequent brakes or acceleration and averts traffic accidents that cause
severe traffic jams. The revolutionary changes brought by this technology are undeniable,
especially its impacts on the travel behaviors, which will reshape the traffic flow modeling
for future transportation problems. Future study is required to calibrate the associated
roadway parameters (e.g., capacity, free-flow speed, and jam density) and modify the
traffic propagation mechanism (or the network loading model in general) for autonomous
vehicles. To further provide a framework for studying the environmental impacts of
autonomous vehicles on road networks, more development on DTA models for
autonomous vehicles (e.g., Levin and Boyles, 2016) is eagerly anticipated.
5.3.4. Electric/hybrid electric vehicles
Since electric vehicles do not generate pollutants on a local level (e.g., roadside emissions)
and can potentially rely on energy from a selection of renewable sources (Millo et al.,
2014), they are regarded as a valuable solution to environmental sustainability. The hybrid
electric vehicles can also lead to a certain level of the environmental benefits of electric
vehicles for a short distance. Despite continuous development in battery technology, the
charging time and driving range are still considered as barriers to its popularization. The
traffic assignment for electric vehicles will be further constrained by driving range and
availability of charging infrastructures. To our best knowledge, a DTA model with path
distance constraints for electric vehicles remains undeveloped; so do the corresponding
solution algorithms.
5.3.5. Transportation planning
Transportation planning is long recognized as an important issue and involved with the
design and siting of transport facilities (generally streets, highways, bike lanes, and public
transport lines) as well as highway expansion or construction. Such transportation network
design problems have been extensively studied using STA models in the literature
(see Farahani et al., 2013). These problems have also been researched using DTA models
(e.g., Janson, 1995, Waller and Ziliaskopoulos, 2001, Heydecker, 2002, June) but with less
attention. DTA models can account for traffic dynamics and time-variant demand. The
traffic flow modeled by DTA can be consistent with the kinematic wave theory, taking
traffic interaction among adjacent links into account. Hence, DTA models can provide
more realistic traffic flow data as inputs for a more accurate emission estimation.
Moreover, the time-variant demand consideration is crucial for proposing different
emission mitigation measures realistically. Conducting a comprehensive road network
design that combines DTA models with the consideration of environmental issues will be a
natural evolution in the transportation research field. Extensions to multimodal network
design including green transport and public transit modes is also another future research
direction.