Francesc Soriguera

Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

4-FUNDAMENTALS OF TRAFFIC FLOW MODELING

Contents

1. Introduction to traffic flow modeling ............................................................................................................. 2

2. Variables defining a traffic flow: Macroscopic & Microscopic perspective.................................................... 3
3. The fundamental equation of traffic (𝑞 = 𝑘𝑣) ............................................................................................... 9
4. The vehicles' conservation equation ............................................................................................................ 11
5.1. Relative flow seen by a moving observer ............................................................................................ 13
5. Traffic diagrams............................................................................................................................................ 15
6.1. Greenshields, Greenberg and Edie's 𝑘 − 𝑣 models ............................................................................. 15
6.2. Traffic engineering manuals: a warning .............................................................................................. 21
6. Macroscopic modeling of traffic flow: LWR - Kinematic Wave Theory ........................................................ 22
7.1. Speed of a traffic wave ........................................................................................................................ 25
7.2. The fundamental diagram of traffic .................................................................................................... 27
7.3. Shocks and waves ................................................................................................................................ 30
7.4. Simplifications ..................................................................................................................................... 34
7. Example of application of the LWR theory: Incident on a freeway ............................................................. 37
8. Limitations of the LWR macroscopic traffic flow theory .............................................................................. 42

------

-1-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

1. Introduction to traffic flow modeling

Traffic flow consists in the movement of a large number of vehicles along a particular infrastructure. Although
the individual movement of these vehicles may seem random, the overall behavior of the traffic flow is highly
predictable. Take as an example the morning traffic rush accessing big cities, like Barcelona. In normal conditions
(i.e. similar demand, no incidents) congestion appears always at same locations, at similar times and evolves in
a similar fashion and with similar queue lengths. In this context, it seems reasonable to think that there must
exist some physical behavioral laws that steer traffic evolution. The knowledge of these laws would allow to
develop traffic flow models, able to predict traffic evolution, congestion and queue lengths. Such models would
be very useful in order to assess planning, management and control of road traffic facilities.
This idea of analyzing traffic flow as a science and in quantitative terms appeared in the 1930's with the pioneer
works of the first traffic scientist ever, Mr. Bruce Greenshields (https://www.ite.org/about-ite/history/honorary-
members/bruce-d-greenshields/). In spite of the early works of Greenshields, it was not until the 1950's that
traffic science was popularized, and that the first dynamic models of traffic flow appeared. From the very
beginning, two different approaches to flow modelling could be differentiated.
On the one hand, traffic could be seen as a flow, neglecting the fact that it is composed of individual "particles"
(i.e. the vehicles). This fluid like approach took researchers to apply hydrodynamic models to traffic, setting the
foundations of macroscopic traffic flow modeling. Two physicists in the UK, M.J. Lighthill and G.B. Whitham
developed in 1955 the first macroscopic traffic flow model, by comparing "traffic flow on long crowded roads"
with "flood movements in long rivers". A year later in the USA, P.I. Richards (1956), physicist and applied
mathematician, developed independently the same model, introducing the concept of "shock-waves on the
highway" completing the so-called LWR model (i.e. Lighthill-Whitham-Richards model). Since then, this model
has received multiple names, like the Continuous Traffic Flow Model, Shock-Wave Theory or Kinematic-Wave
Theory. They all refer to the same original and fundamental macroscopic model.
On the other hand, traffic flow could be seen as the aggregated movement of multiple vehicles, so that if
individual trajectories are predicted, the overall traffic behavior can be obtained by integration. This trajectory-
based approach constitutes the microscopic traffic flow modelling approach. In its most simple form,
characterized by the unidimensional movement of vehicles in a single lane, models are referred to as car-
following models. The concept is that in dense traffic conditions each vehicle follows the trajectory of the vehicle
in front, and reacts to its speed changes. By modeling this car-following behavior of vehicles, for all the vehicles
in a traffic stream, we could obtain the overall traffic evolution. Seminal car-following models were developed
by L. A. Pipes (1953) based on the California driving code and T.W. Forbes (1958) introducing the concept of the
"reaction time", both proposing a linear relationship between the distance between two consecutive vehicles
(i.e. the vehicular spacing) with respect to the traveling speed. Building on these initial works, and others
developed over the years, a series of models developed by R.E. Chandler, D.C. Gazis, R. Herman, E.W. Montroll,
R.B. Potts and R.W. Rothery at the General Motors (GM) Research Laboratories (Detroit, USA) in the period 1958-
61 have received maximum attention over the years. These car-following models were the first to incorporate
the stimulus-response structure, where the stimulus is a differential speed between the leader and the follower
vehicles and the response of the follower is an acceleration or braking maneuver. It has taken more than half a
century to evolve from these first analytical car-following models at the GM research labs to the traffic
microsimulators we have today (like this one: https://www.youtube.com/watch?v=k_KjM3l295M or this other:
https://www.youtube.com/watch?v=OtYby7QnyAE). Obviously, in addition to a car-following model, we need
many other components (e.g. lane-changing model, controllers behavior, O/D matrixes, ...) plus an advanced
-2-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

digital graphics animation. In spite of being very fancy, calibrating these microsimulators requires many
parameters (e.g. of the order of 30 maybe), some of them without any physical interpretation. This is 10 times
more than the 3 parameters required to calibrate the fundamental diagram in the LWR theory (as you will see
next in this lecture). This means that microscopic models are much less robust with respect to LWR, and that all
results should be validated with the later.
After this brief introduction to traffic flow modeling and its history, this lecture aims to just present the most
fundamental concepts of traffic flow modeling. This includes the definition of traffic variables and the
fundamental equation, and the postulates of LWR macroscopic model (i.e. the conservation equation and traffic
diagrams). We will end the lecture with an example application of this theory.

2. Variables defining a traffic flow: Macroscopic & Microscopic perspective

Recall that a trajectory is the representation of the movement of one vehicle in space and time (i.e. 𝑥, 𝑡).
However, when dealing with traffic flow modeling, we are more interested in the variables defining the
movement of large amounts of vehicles. Figure 1 and Figure 2 show the trajectories of several consecutive
vehicles in a single lane, from where we can derive the fundamental variables of traffic flow.

x
𝑖=1 𝑖=2 𝑖=3 𝑖=4 𝑖=5

𝑥!
ℎ"

Headway

𝑡! 𝑇 t
𝑡"
Figure 1. Headway definition on a time – space diagram.

The time interval between the passage of two consecutive vehicles at a given location 𝑥! is defined as the
headway, ℎ, in units of time (e.g. [s]). In Spanish the headway is named "intérvalo". Note that the headway is
-3-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

measured taking the same reference point of consecutive vehicles (e.g. the rear bumper in Figure 1), so that, in
addition to the empty time gap between vehicles, it includes the time necessary for the passage of the length of
the vehicle. Also note that in order to measure the headway, we need to measure at a fixed location continuously
in time. This kind of measurement is called a temporal measurement. Consider that we take this temporal
measurement during a long period of time, 𝑇, and that we measure the passage of 𝑚 vehicles (e.g. in Figure 1,
𝑚 = 5). Then we can define the traffic flow, 𝑞, as:
𝑚
𝑞= [𝑣𝑒ℎ/𝑡𝑖𝑚𝑒]
𝑇

Note that the headway is a traffic variable affecting individual vehicles (i.e. a microscopic traffic variable) while
the flow is an aggregated or average variable (i.e. a macroscopic traffic variable). Both variables (i.e. ℎ and 𝑞)
represent different perspectives (i.e. micro or macro) of the same concept, and they can be related as:

𝑚 𝑚 1 1
𝑞= = $ = =9
𝑇 ∑#%" ℎ# 18𝑚 ∑#%" ℎ# ℎ
$

So that, the flow is equal to the inverse of the average headway.

𝑗=1 𝑗=2 𝑗=3 𝑗=4 𝑗=5

𝑥!

𝑠"
𝐿

Spacing

𝑥"

t
𝑡!
Figure 2. Spacing definition on a time – space diagram.

-4-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

The same analysis could be done but considering a spatial measurement region. This is the simultaneous
measurement of all the vehicles in a length 𝐿 of the infrastructure at a particular instant of time, 𝑡! (see Figure
2). This allows defining the vehicular spacing, 𝑠, as the distance between the same point of consecutive vehicles
(e.g. the rear bumper in Figure 2). The spacing, is the second fundamental microscopic traffic variable. Note that,
as before, 𝑠 includes the vehicle length plus the empty space gap between vehicles. In Spanish the spacing is
named "espaciado". The macroscopic variable equivalent to the spacing is the traffic density 𝑘, defined as the
number of vehicles per unit distance observed at a given instant in a particular infrastructure. The traffic density
𝑘 can be obtained from the measurement of the existing vehicles, 𝑛, on a given infrastructure length, 𝐿, at a
particular instant of time, 𝑡! . Note that in Figure 2, 𝑛 = 2 (i.e. vehicles 𝑗 = 1 and 𝑗 = 2). Then, traffic density is
defined as:
𝑛
𝑘= [𝑣𝑒ℎ/𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒]
𝐿

Similarly, the spacing (microscopic) and the density (macroscopic) variables are related, as one is the aggregation
of the other. Namely:

𝑛 𝑛 1 1
𝑘= = ' = =
𝐿 ∑&%" 𝑠&
18𝑛 ∑&%" 𝑠& 𝑠̅
'

So that the traffic density can be obtained as the inverse of the average spacing.

The third and last of the fundamental variables is the speed. The microscopic version of the variable is simply the
vehicles' individual speed, 𝑣# (i.e. the slope of the vehicles' trajectory at a given point in time and space).
However, the macroscopic average, 𝑣̅ , is a bit more problematic. The average speed is defined as the arithmetic
average of the individual speed of vehicles, the problem is that this average depends on which vehicles are
considered. Amongst all possible selections of the space-time measurement region, the temporal region (e.g.
(𝑥( , 𝑇) and the spatial region (𝐿, 𝑡! ), define two particular cases, leading to the time-mean speed, 𝑣̅) , and to the
space-mean speed, 𝑣̅* , respectively (see Figure 3). Time-mean and space-mean definitions, could be applied to
any property of the travelling vehicles in addition to the travelling speed.

-5-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

x
Space-mean
(fixed 𝑡)

𝐿 Time-mean
𝑥! (fixed 𝑥)

𝑡! t
𝑇
Figure 3. Time vs Space mean averages.

Then, according to the notation in Figure 1 and Figure 2, we define:

∑$
#%" 𝑣#
𝑣̅) =
𝑚
∑'&%" 𝑣&
𝑣̅* =
𝑛

Because faster vehicles are overrepresented when seen by a stationary observer (i.e. in a temporal measurement
region (𝑥( , 𝑇)) for any prevailing traffic state, 𝑣̅) ≥ 𝑣̅* , and they are equal only when the speeds of all vehicles
are constant. More precisely:

𝜎+,-!
𝑣̅ ) = 𝑣̅* +
𝑣̅*

where 𝜎+,-! is the variance of the individual speeds measured over the spatial measurement region (𝐿, 𝑡! ). Recall
that the variance is always positive, and only zero when the variable (i.e. the individual speed, 𝑣# ) is constant.
The previous relationship is named as the Wardrop relationship, in honor of J.G. Wardrop, the distinguished
English mathematician and transport analyst who formulated it the first time in 1952 in his famous publication
about "some theoretical aspects of road traffic research".

-6-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Time means can be related with flow and space means to density. Note that any traffic state can be thought of
being composed of 𝑙 different families of vehicles (see Figure 4), where within each family traffic is stationary
(i.e. constant vehicular speeds, headways and spacings). This is not limiting in any sense, as any non-stationary
traffic state can be formulated in these terms (e.g. even a traffic state where all vehicles travel at different
speeds, can be thought in terms of "families", where each family is composed of a single vehicle). Then we can
define 𝑞. , 𝑘. and 𝑣. , as the flow, density and speed of traffic within family 𝑙. Because the flow and density are
additive magnitudes, the total flow, 𝑞, and the total density, 𝑘, can be obtained as:

𝑞 = H 𝑞. ; 𝑘 = H 𝑘.
. .

Also, we can state that:

∑$
#%" 𝑣# ∑. 𝑚. 𝑣. J18𝑇K ∑. 𝑚. 𝑣. ∑. 𝑞. 𝑣.
𝑣̅) = = = =
𝑚 𝑚 J18𝑇K𝑚 𝑞

where 𝑚. is the number of vehicles of family 𝑙 in the temporal region (𝑥( , 𝑇). This shows that time-means can
be obtained as weighted averages, where the weights are the relative flows of each family of vehicles.

Similarly:

∑'&%" 𝑣# ∑. 𝑛. 𝑣. J18𝐿K ∑. 𝑛. 𝑣. ∑. 𝑘. 𝑣.
𝑣̅* = = = =
𝑛 𝑛 J18𝐿K𝑛 𝑘

where 𝑛. is the number of vehicles of family 𝑙 in the spatial region (𝐿, 𝑡( ). As before, this shows that space-means
can be obtained as weighted averages, where the weights are the relative densities of each family of vehicles.

Figure 4 exemplifies these concepts. Consider three vehicles travelling on a circular track with a length of 30 km
at different speeds. This can be seen as 3 families of vehicles:

𝑙 = 1 → 𝑣" = 90 𝑘𝑚/ℎ
𝑙 = 2 → 𝑣- = 60 𝑘𝑚/ℎ
𝑙 = 3 → 𝑣/ = 30 𝑘𝑚/ℎ

From the spatial perspective, in the 30 km of the track there are:

1 vehicle @ 90 km/h (𝑛" = 1)

1 vehicle @ 60 km/h (𝑛- = 1)
1 vehicle @ 30 km/h (𝑛/ = 1)

So that the space mean speed is simply the arithmetic average of their speeds, 𝑣̅* = 60 𝑘𝑚/ℎ, since the relative
density of the 3 families is the same (i.e. 1/3).

-7-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

In contrast, from the temporal perspective, one observer standing at any fixed location at the side of the track
during 𝑇 = 1ℎ, would observe:

3 vehicles @ 90 km/h (𝑚" = 3)

2 vehicles @ 60 km/h (𝑚- = 2)
1 vehicle @ 30 km/h (𝑚/ = 1)

Then, the time mean speed computed as a weighted average considering the relative flows is 𝑣̅) = 70 𝑘𝑚/ℎ.
Note that vehicles travelling at higher speeds are overrepresented with respect to their density.

30 km/h
90 km/h
1
3
𝐿 = 30 𝑘𝑚
2
60 km/h
x

90 km/h 60 km/h 30 km/h

30 km

𝑥!

𝑡! 1h t

Figure 4. Time vs Space mean speed.

Table 1 provides a summary of what we have seen so far. Additionally, it includes two additional pieces of
information. The "Simulators" column shows computer implementations of the different traffic modeling
approaches. The Cell Transmission Model (originally developed by Prof. C.F. Daganzo) is a finite difference
-8-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

implementation of the LWR macroscopic model. This, can be coded into any computing language, and several
implementations can be found as freeware. On the other hand, microscopic traffic simulators tend to be
commercial software under license, and only SUMO represents an open-source alternative developed by the
academic community. Table 1 also provides the macro-micro equivalence for a couple of models. For instance,
the Greenberg macroscopic 𝑘 − 𝑣 is the macroscopic equivalent of the 3rd generation of the General Motors car-
following theories. Take this only as informative, as we are not going to discuss the details of these models in
this chapter.

Table 1. Macro and micro approaches to traffic modelling – Variables, models and simulators

Variables Models Simulators

Flow Density Average Speed Continuous theories Cell Transmission

MACRO
(𝑞) (𝑘) (𝑣̅ ) LWR – KW (shock wave theory) Model

Instantaneous
Headway Spacing e.g. AIMSUN, VISSIM,
MICRO speed Car following theories
(ℎ) (𝑠) PARAMICS, SUMO
(𝑣! )
e.g. Greenberg (macro 𝑘 − 𝑣) – 3rd General Motors (micro
MACRO-MICRO 1 1 ∑"! 𝑣! car following)
𝑞= 𝑘= 𝑣̅ =
Relationships ℎ, 𝑠̅ 𝑛 e.g. Newell Triangular diagram (macro 𝑞 − 𝑘)– Forbes min.
safety distance (micro car followig)
Aggregation (𝑥, 𝑇) → 𝑣̅#
(𝑥, 𝑇) (𝐿, 𝑡)
types (𝐿, 𝑡) →𝑣̅$

T)
3. The fundamental equation of traffic (𝒒 = 𝒌𝒗
The three fundamental variables of traffic (i.e. flow 𝑞, density 𝑘, and average speed, 𝑣̅ ) are related by the so
called "Fundamental Equation of Traffic", which states that the flow is equal to the traffic density times the
average speed:

𝑞 = 𝑘𝑣̅

Obviously, the fundamental equation can also be formulated in terms of the average microscopic variables:

𝑠̅ = ℎ9𝑣̅

The fundamental equation of traffic directly results from the previous definitions of the variables, meaning that
it is true "by definition". This implies that it holds everywhere, for all kind of infrastructures and for all possible
traffic states. In spite of this, for stationary traffic (i.e. constant vehicular speeds, headways and spacings) it is

-9-
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

particularly easy to prove (see Figure 5). If we carefully define a time-space measurement region (𝐿, 𝑇) so that
𝐿⁄𝑇 = 𝑣̅ , where 𝑣̅ = 𝑣# ∀𝑖 because traffic is stationary, then:
𝑚 𝑚
𝑞= ;𝑘=
𝑇 𝐿
𝑞 𝐿
= = 𝑣̅
𝑘 𝑇

𝐿 𝑚 vehicles
𝑣# =
𝑇

t
𝑇
𝑚 vehicles
Figure 5. Stationary traffic on a time – space diagram.

In spite of the previous derivation of the fundamental equation of traffic for a stationary traffic state, recall that
the equation holds even for non-stationary traffic (i.e. when the vehicular speeds are not constant). Then, an
additional issue appears, because the average speed, 𝑣̅ , is not well defined. Should we consider the time-mean
speed, 𝑣̅) in the fundamental equation? or should we consider the space-mean speed 𝑣̅* ? Note that, because
they are different in non-stationary traffic, only one of the options can be true.

To give an answer to this question, think again of a non-stationary traffic composed of 𝑙 different families of
vehicles. Within each family, traffic is stationary with 𝑞. , 𝑘. , 𝑣. , and because we have already proved that the
fundamental equation holds in stationary traffic, within each family we have:

𝑞. = 𝑘. 𝑣.

Then, by simply working with the algebra que can state that:

- 10 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

∑. 𝑘. 𝑣.
𝑞 = H 𝑞. = H 𝑘. 𝑣. = 𝑘 W X = 𝑘𝑣̅*
𝑘
. .

Recall that a weighted average of vehicles speeds, where the weights are the relative densities, yields the space-
mean speed. So, this proves that in non-stationary traffic, the fundamental equation holds only if the speed is
obtained as a space-mean. This is:

𝑞 = 𝑘𝑣̅*

4. The vehicles' conservation equation

One of the postulates of all models of traffic flow is that vehicles are "conserved". This means that they cannot
"disappear", which seems a quite reasonable postulate. The formulation of the vehicles' conservation law is
equivalent to any other of the existing conservation laws (mass, energy, etc.) and it states that in any closed
system the number of vehicles entering minus the number of vehicles exiting, must be equal to the difference in
the number of vehicles stored inside the system.

x 1 2
(A) 2
𝑥2 2
3

𝐿 3 4
4
1
𝑥1 (B)
𝑇

N
𝑡1 𝑡2 t

Figure 6. Derivation of the vehicles' conservation equation from a trajectories diagram.

There are several ways of analytically formulating the vehicles' conservation equation, but using a space-time
diagram is quite simple and convenient. In the context of traffic flow, a "closed system" means a closed space-
time region, where all the entrances and exits to the infrastructure are monitored. Figure 6 shows a closed (𝑥, 𝑡)
region with dimensions (𝐿, 𝑇) and some vehicle trajectories. The infrastructure considered in Figure 6 could be
a section of a one-way street so that vehicles can only enter through location 𝑥" and exit through 𝑥- . This section
of the street is observed in the time period between 𝑡" and 𝑡- . In this context, note that the vehicles entering
- 11 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

the section of the street through 𝑥" during the period 𝑇, (i.e. crossing border 1 of the closed space-time region)
are vehicles 3 and 4 (i.e. two vehicles). The notation for this variable will be 𝑚" = 2. The number of vehicles
exiting through 𝑥- during the same period (i.e. crossing border 2) are vehicles 1 and 2. In this case 𝑚- = 2. Also,
you can see from Figure 6 that, initially (i.e. at 𝑡" ), there were two vehicles inside the section of street of length
𝐿. These are vehicles 1 and 2. Because these are spatial borders, the notation used in this case for the number
of trajectories crossing border 3 is 𝑛/ = 2. Finally, the number of vehicles "stored" inside the section of street at
the end of the observation period, 𝑡- (i.e. trajectories crossing border 4) is 𝑛0 = 2 (i.e. vehicles 3 and 4). Given
these definitions, the conservation of vehicles can be simply formulated by imposing that the number of
trajectories entering the closed (𝑥, 𝑡) region must be equal to the number of trajectories exiting1. This is:

𝑚" + 𝑛/ = 𝑚- + 𝑛0

Note that the physical meaning of the previous equation is exactly that of the conservation law:

𝑒𝑛𝑡𝑒𝑟𝑖𝑛𝑔(𝑚" ) + 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 𝑎𝑐𝑐𝑢𝑚𝑢𝑙𝑎𝑖𝑜𝑛(𝑛/ ) = 𝑒𝑥𝑖𝑡𝑖𝑛𝑔(𝑚- ) + 𝑓𝑖𝑛𝑎𝑙 𝑎𝑐𝑐𝑢𝑚𝑢𝑙𝑎𝑡𝑖𝑜𝑛(𝑛0 )

And working with the algebra, allows obtaining a more familiar form of the vehicles' conservation equation:

𝑚- − 𝑚" = 𝑛/ − 𝑛0
𝑚- − 𝑚" 𝑛/ − 𝑛0
=
(𝑥- − 𝑥" )(𝑡- − 𝑡" ) (𝑥- − 𝑥" )(𝑡- − 𝑡" )

𝑚" + 𝑛/ 𝑚- + 𝑛0
=
𝐿𝑇 𝐿𝑇

Note that 𝑚's divided by 𝑇 are flows, while 𝑛's divided by 𝐿 are densities. Therefore:

𝑞- − 𝑞" 𝑘/ − 𝑘0
=
(𝑥- − 𝑥" ) (𝑡- − 𝑡" )

∆𝑞 ∆𝑘
=−
∆𝑥 ∆𝑡

Which says that the variation of flow with respect to space is equal to the minus variation of the density with
respect to time, representing an equivalent formulation of the conservation equation. Considering that typically
we deal with many vehicles and that in this case the vehicle count function can be considered continuous, the
previous equation can be rewritten as:

1
This is equivalent to imposing that the change in the cumulative vehicle count, 𝑁, between two points in the time-space
plane (e.g. points A and B in Figure 7) must be the same independently of the followed path (e.g. Borders 1 and 3 or Borders
2 and 4).
- 12 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝜕𝑞 𝜕𝑘
=−
𝜕𝑥 𝜕𝑡

which represents the most typical expression for the vehicles' conservation equation.

4.1. Relative flow seen by a moving observer

As a side note, and as an application of the conservation equation, it is interesting to determine the relative flow
seen by an observer that is moving with traffic. The relative flow can be interpreted as the number of overtakings
per unit time seen by the moving observer. Note that this depends on the prevailing flow and on speed of the
observer (e.g. if the observer travels at the same speed as the average traffic, the relative flow would be zero).

𝒙
Observer
B

𝑢!

𝐿 = 𝑢' 𝑇

𝑇 𝒕

Figure 7. Derivation of the relative flow seen by a moving observer from the conservation equation.

In order to derive an analytical expression for this relative flow, look at Figure 7, where in the time – space plane
we define a triangular control area of sizes 𝐿 and 𝑇. By applying the conservation of vehicles in this closed area,
we can say:

𝑞! 𝑇 + 𝑘𝐿 = 𝑞𝑇

Where 𝑞! is the relative flow seen by an observer moving at speed 𝑢! , and 𝑞 and 𝑘 are respectively the flow and
density of the prevailing traffic state. Note that 𝑞! 𝑇 represents the number of times the observer is overtaken
by other vehicles in the duration 𝑇 (i.e. black full dots in Figure 7), 𝑘𝐿 is the number of vehicles on 𝐿 at the
beginning of 𝑇 (i.e. squares in Figure 7), and 𝑞𝑇 is the total number of vehicles leaving 𝐿 during 𝑇 (i.e. empty dots

- 13 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

in Figure 7). By realizing that 𝐿 = 𝑢! 𝑇 , and substituting in the previous equation, we obtain (after simplyfing for
𝑇):

𝑞! + 𝑘𝑢! = 𝑞

And finally:

𝑞! = 𝑞 − 𝑘𝑢!

The previous equation states that the relative flow seen by a moving observer is equal to the prevailing flow
minus the prevailing density time the speed of the observer. Note that, according to the fundamental equation
of traffic, 𝑞 = 𝑘𝑣, where 𝑣 is the average speed of traffic. This means that:

𝑞! = 𝑘(𝑣 − 𝑢! ) = 𝑘𝑣12.

Where 𝑣12. = 𝑣 − 𝑢! is the relative speed between traffic and the observer. Note that if the relative speed is
zero, the relative flow it is also zero. In addition, if the speed of the observer is larger than that of traffic, the
relative flow is negative. This means that the observer overtakes the other vehicles in the traffic stream, and
seen from the moving observer perspective, traffic is like running backwards (see Figure 8). Finally, consider that
this expression can also be applied if the observer travels in the opposite direction of traffic. In such case the
speed of the observer is negative, and the relative flow would be larger than the regular flow seen by a static
observer.

Case 1. Vehicles slower than observer Case 2. Vehicles faster than observer
𝒙 𝒙

𝑢! 𝑢!

𝑣 𝑣

𝐿 𝐿

𝑇 𝑇
𝒕 𝒕

Figure 8. Moving observer on a trajectories diagram.

- 14 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

5. Traffic diagrams
Along with the vehicles' conservation equation, the second postulate of macroscopic traffic flow models is the
existence of an equation of state. In other words, this means that the state of the system can be univocally
determined by applying one equation to a state variable. Typically, this equation of state is plotted in a relevant
coordinate axis, defining what is known as a traffic diagram. In the context of traffic flow modelling, there are 3
fundamental variables, which univocally define the traffic state (i.e. 𝑞, 𝑘, 𝑣̅ ). This means that if we want only one
degree of freedom (i.e. one state variable), we need two equations between these variables. We already know
one, the fundamental equation of traffic (i.e. 𝑞 = 𝑘𝑣̅ ). Still, we need another equation between any pair of the
variables.

5.1. Greenshields, Greenberg and Edie's 𝒌 − 𝒗 models

The possibility of deriving a relationship between a pair of traffic variables appeared from observation. The first
traffic observations were performed by the civil engineer B. D. Greenshields, as early as 1933-35. He carried out
experiments to measure traffic density and speed using photographic measurement methods for the first time.
Measuring traffic with the equipment available at the 1930's, was not an easy task, so that obtaining a single
measuring point was challenging (see http://www.krbalek.cz/For_students/mds/clanky/Greenshields.pdf) for a
more detailed description of the Greenshields experiments). In spite of this, Greenshields managed to obtain 6
measurement points, which he represented in a density - speed coordinate axis (see Figure 9). Observing the
fairly linear arrangement of the measurement points, Greenshields proposed a linear relationship between
speed and density. This linear 𝑘-𝑣 model was the first proposal for the equation of state, and the first traffic
diagram ever.

From this historical note on the appearance of traffic diagrams, two important properties are already apparent:

• Traffic diagrams are empirical. They are obtained from observation (not by definition as in the fundamental
equation). This means, that they are valid for the infrastructure and the drivers/vehicles observed. So, care
must be taken when extrapolating traffic diagrams to infrastructures and drivers significantly different from
where they were measured.

• Traffic diagrams are obtained from regression to data points. This means that they are true on average, but
that it is possible that some measurement point lay significantly apart from the average regression.

- 15 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Fig. 5: Speed Density Relation V (Greenshield 1934)

Figure 9. The original Greenshields 𝒌-𝒗 model

The regression of a functional form to data points for obtaining the traffic diagram (i.e. the calibration of the
diagram) yields its parameters. The typical parameters of traffic diagrams are:

• 𝑣3 , the free flow speed => This is the maximum average speed that drivers feel safe and comfortable to drive
at when the density is very low. A typical value for a freeway lane is 𝑣3 = 100 − 120 km/h, depending on
the physical layout, and although this is influenced by the prevailing speed limit.

• 𝑘& , the jam density => This is the maximum density at the infrastructure, when the vehicles are in a gridlock
jam, completely stopped. A typical value for a freeway lane is 𝑘& = 125 − 150 veh/km.

• 𝑞$45 , the maximum flow, referred as the capacity => This is the maximum throughput that the infrastructure
can hold. In turn, 𝑣! and 𝑘! represent the corresponding optimal speed and density, respectively, for which
the capacity point is obtained. A typical value for a freeway lane is 𝑞$45 = 2000 − 2200 veh/h, 𝑘! = 20 −
25 veh/km and 𝑣! = 70 − 90 km/h.

For instance, the Greenshields linear 𝑘-𝑣 model, requires the calibration of two parameters, 𝑣3 and 𝑘& to obtain
the linear functional form as:

𝑘
𝑣 = 𝑣3 `1 −Diagram
Fig. 6: The first Fundamental a as v-q Diagram
𝑘 &

In this model some traffic flow characteristics are expressed well. It shows a maximal
Figuretraffic flowall the
10 shows withprevious
the related optimal
parameters trafficGreenshields
in the linear density. In
𝑘-𝑣the q-v-diagram exist two
diagram.
regimes, that means it’s possible to have two speeds at the same traffic flow. By this
the traffic flow is classified in a stable and - 16 -an unstable regime. Greenshields linear
relation would be called an univariate model, because both regimes are calculated
with the same formula.
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝒗
Free-flowing Congested

𝑣3
𝑘
𝑣 = 𝑣3 `1 − a
𝑘&
𝑣" Capacity
𝑣̅ ! =
2

𝑞$45

𝑘& 𝐤
𝑘0 = 𝐾𝑗
2

Figure 10. Greenshields 𝒌 – 𝒗 linear model (1934).

Note, that the variable not directly represented in the coordinate axis (i.e. the flow in this case) can be obtained
by applying the fundamental equation, or figuring out the graphical representation on the diagram. In a 𝑘-𝑣
diagram the flow is obtained as the area defined by the coordinate points of a given traffic state. The capacity
point in the Greenshields diagram illustrates that, although this model is of historical importance and
academically used (because of its simplicity), it is not accurate. Note for instance that the optimal speed, 𝑣!
results to be 𝑣3 ⁄2, while in reality is significantly larger, and 𝑘! = 𝑘& ⁄2, while in reality is much smaller.

The capacity point divides the diagram in two parts. For densities higher than 𝑘! , the flow is reduced with growing
densities. This is the most precise definition of congested traffic. So, the right-hand side of the diagram Figure 10
corresponds to congested traffic states. In contrast, the left-hand side corresponds to free-flowing traffic states,
where an increase of the density is translated into an increase in the flow.

Since the first linear 𝑘-𝑣 model proposed by Greenshields, and given his limitations to reproduce the empirical
data that became available with the passage of time, many other functional forms were (and still are) proposed
to describe traffic diagrams. In the early days, dealing with data scarcity, it could be said that there were as many
functional proposals for traffic diagrams as traffic databases. In spite of this, some functional relationships have
stood the pass of time, and it is worth referring to them here:

• The Greenberg 𝑘-𝑣 model, proposed by Harold Greenberg (1959) who considered traffic as continuous
compressible fluid (an acceptable assumption for congested traffic states, but not in light traffic conditions)
and derived analytically a logarithmic relationship between density and speed, as shown in Figure 11.

- 17 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Greenberg validated his logarithmic functional form by calibrating the two parameters of his model (i.e. the
optimal speed, 𝑣! , and the jam density, 𝑘& ) with data measured at the Lincoln Tunnel in New York city and
at the Merritt Parkway in Connecticut, which included for the first time a significant number of
measurements of congested traffic. The model cannot be applied directly for low densities (i.e. free-flow
traffic) as speed tends to infinity when density tends to zero. One possible solution is to truncate the model
at the free-flow speed, 𝑣3 , adding a third calibration parameter to the model and defining what is called a
two-regime diagram (i.e. different functional forms for free-flowing and congested traffic).

Not accurate in this

zone
𝑣𝑓
𝑘&
𝑣0 𝑣 = 𝑣! . 𝑙𝑛 ` a
𝑘

𝑘𝑗
𝑘0 =
𝑒

𝑘0 𝑘𝑗 𝐤

Figure 11. Greenberg model (1959) – logarithmic approach to 𝒌 − 𝒗.

• The Underwood 𝑘-𝑣 model, proposed by Robin T. Underwood (1961) aimed to overcome the limitation of
Greenberg's model for free-flowing traffic and put forward an exponential model as shown in Figure 12.
Underwood calibrated the two parameters of the model (i.e. the free-flow speed, 𝑣3 , and the optimal
density, 𝑘! ) at two locations: the Merritt Parkway in Connecticut (same as the calibration location for the
Greenberg model) and at the Princess Highway in Victoria (south Australia). In Underwood's model, speed
becomes zero only when density reaches infinity. Some academics consider this as a drawback of this model,
while others postulate that actually, even when traffic density reaches the jam value, 𝑘& , the average speed
is not zero as traffic continues moving forward in a stop and go fashion. Hence, using Underwood's model
for predicting speeds at high densities is still acceptable.

- 18 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

6787
𝑣 = 𝑣3 . 𝑒 #

𝑣𝑓

Discussion
about accuracy
in this zone

𝑘𝑂 𝑘𝑗 𝐤

Figure 12. Underwood model (1961) – exponential approach to k-v.

6787
𝑣𝑓 𝑣 = 𝑣3 . 𝑒 #

𝑘&
𝑣𝑂 𝑣 = 𝑣! . ln ` a
𝑘

𝑘0 𝑘𝑗 𝐤
Underwood Greenberg

Figure 13. Edie k –v model (1961).

• Leslie C. Edie proposed for the first time a two-regime model, using the Underwood model for free-flow
conditions, and the Greenberg model for congested traffic, resulting in a discontinuous exponential form
(see Figure 13). Edie's proposal requires the calibration of 4 parameters: the free-flow speed, 𝑣3 , the jam
- 19 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

density, 𝑘& , and the breakpoint between free-flowing and congested traffic (i.e. the optimal point at capacity,
𝑣! and 𝑘! ). Note that the calibration of "optimal" values at capacity is more cumbersome than those of
parameters at jam or light traffic conditions, because establishing that a particular traffic state corresponds
to capacity conditions is not evident and easily determined. This is the reason why 𝑣! and 𝑘! are usually
determined from statistical regression, without much support from its physical interpretation. This can be
seen as a drawback (e.g. of Greenberg, Underwood or Edie's models) with respect to other models (e.g.
Greenshields) relying only of free-flow or jam parameters.

As an end for this section devoted to traffic diagrams, it should be understood that from the functional form of
one diagram (e.g. a 𝑘-𝑣 model) all the other derived diagrams (i.e. all the possible 2-dimensional plots of different
combinations of variables) can be obtained using the fundamental equation of traffic. It is a good exercise for
practicing to derive the different type of diagrams that would result, for instance, from the linear Greenshields
𝑘-𝑣 model. For illustrative purposes, Figure 14 shows the conceptual generic form of several possible diagrams.

𝑣 𝑣 𝑞
𝑣$ 𝑣#
𝑣2 𝑣2 1
𝑣- 1 𝑞/01
1
𝑣'
𝑣' 𝑞- 𝑤
1/𝑘! = 𝑠!

𝑞/01
𝑠"
𝑘 𝑞 𝑘
𝑘- 𝑘. 𝑞/01 𝑘.
𝑝 𝑣
𝑝
1

𝑞/01
1/𝑣' 𝑡𝑡' 𝑣2
𝑘- 1 1/𝑞%&'
𝑡𝑡2
1/𝑞!
𝑡𝑡2 1

𝑞/01 𝑞 𝑘. 𝑘 𝑠. = 𝑘.,- 𝑠 = 𝑘 ,-

Figure 14. Generic representation of different traffic diagrams

Note: 𝑝 is defined as the pace, 𝑝 = 1⁄𝑣, which represents the travel time (𝑡𝑡) per unit distance.

- 20 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

5.2. Traffic engineering manuals: a warning

Manuals on traffic engineering, and specifically the Highway Capacity Manual (HCM), do have a long tradition in
the analysis of traffic flows. The first edition of the Highway Capacity Manual was released in 1950 as a result of
a collaborative effort between the Transportation Research Board (TRB) and the Bureau of Public Roads
(predecessor to the Federal Highway Administration) in the USA. This first edition of the HCM contained 147
pages, presenting concepts, guidelines, and procedures for computing the capacity and quality of service of
highway facilities. Today, 70 years and 6 editions later, the HCM is a vast publication (mainly online) which serves
as a reference for evaluating the multimodal operation of streets, highways, freeways, arterial roads,
roundabouts, signalized and unsignalized intersections, rural highways, and the effects of mass transit,
pedestrians, and bicycles on the performance of these systems.

Despite the HCM expansion to multiple types of infrastructures, and the incorporation of computational
procedures, the essence of the HCM is still the same as in its origins. It consists on a methodology to define
different levels of service (LoS) on “traffic” diagrams measured in the different types of infrastructures (see Figure
15). LoS are categorized between A to F, where A is the best LoS possible, E corresponds to capacity and F to
congested states in the infrastructure. The typical application of the HCM consists in the evaluation of the LoS in
a given infrastructure for a particular demand input.

A
𝑣3
B
C
D
E

F
𝑞4 = Equivalent flow in
ideal conditions

𝒒
1 A𝒒𝒆

Figure 15. HCM levels of service on a 𝒗– 𝒒 diagram.

- 21 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

In line of our previous discussion of traffic diagrams, two warnings should be risen in relation to the usage of
traffic engineering manuals:

• Traffic diagrams result from statistical regression of empirical data to some selected functional form.
Many errors could exist in this calibration process (e.g. wrong specification of the functional form, low
correlation of the function to data, small statistical significance due to the lack of data, …) which are not
evident from the utilization of the manual. This implies that the utilization and results of this type of
manuals give a false sense of precision.

• Traffic diagrams are valid for the particular infrastructure and driver types for which they were
measured. Diagrams may not hold for a particular application where some of the boundary conditions
are different.

In conclusion, the warning is that results of applying traffic engineering manuals are not precise and may be
unreliable. So, do not make important decisions made only on their outputs, without taking your own
measurements and making comparative assessments.

6. Macroscopic modeling of traffic flow: LWR - Kinematic Wave Theory

The objective of macroscopic traffic flow modeling is to predict the evolution of traffic variables in time and
space. The adjective "macroscopic" means that the focus is not on individual vehicles, but on the prediction of
the aggregate variables (i.e. 𝑞, 𝑘 and 𝑣̅ ). This is especially interesting when queued traffic conditions prevail so
that the theory could help in answering questions like: when does queue arrive to a point of interest? (e.g.
important to prevent blockages), how far does the queue grow? or when does the queue dissipate. Note that
traffic flow theory deals with the evolution of traffic states, and it is very well suited to answer questions related
to the physical extension of queues. However, there are easier ways of answering questions about delays (e.g.
use cumulative count curves).

The first and fundamental macroscopic approach to traffic flow modelling is the LWR (Lighthill-Whitham-
Richards) model, also named the continuous model of traffic flow, the kinematic wave theory, or the shock-wave
theory. All refer to the same original model by LWR. This is a traffic flow model for heavy traffic (i.e. dense,
queued traffic) where an increase of the traffic density implies a reduction in the average travelling speed, and
vice versa (see Figure 16).

- 22 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝒗 𝒗

2
2

𝑞- = 2𝑞"
1
1
𝑞"

𝑘. 𝒌 𝒒

Figure 16. Congested traffic states on a 𝒗 – 𝒒 diagram from a 𝒌 – 𝒗 model.

Note that the causality for heavy traffic can be twofold. One possibility is that it appears as a result of a flow
restriction (Case a; see Figure 17). When the demand is large, approaching or going above the maximum flow
that can go through the restriction, the density of vehicles upstream of the bottleneck grows leading to a speed
reduction to ensure safety. Another option is that heavy traffic appears as a result of a speed limitation or a
moving bottleneck on the infrastructure (Case b; see Figure 18). In such case, drivers need to adapt to the lower
restricted travelling speed, accepting a higher density of vehicles.

Traffic state 1 Traffic state 2

𝑞" ≥ 𝑞$45
𝑞- = 𝑞$45 𝑞$45

𝑞$45
𝑣- =
𝑘(𝑞$45 )

Figure 17. Case (a): Flow restriction

- 23 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Traffic state 1 Traffic state 2

𝑉-
𝑣' < 𝑣-
𝑉1

𝑞- 𝑞5 , 𝑘5 , 𝑣5 = 𝑣'
𝑘5 = 𝑘(𝑣' )

𝑞5 = 𝑘5 𝑣'

Figure 18. Case (b): moving obstruction.

Causality

Case (b): moving obstruction;

speed limit
Causality
Case (a): Flow restriction

Figure 19. Two possible causes of heavy traffic on a 𝒒 – 𝒗 diagram.

In any case, whenever a vehicle joins or leaves a congested traffic state, there is a transition to the new traffic
state, which involves a deceleration or an acceleration process. The LWR traffic flow theory for dense traffic aims
to predict the evolution of these transitions. The theory is based on the hydrodynamic analogy and considers
that traffic state transitions are propagated as waves. Whenever there is a change in the traffic state, a traffic
wave is generated. A wave is a transition, an interphase between different traffic conditions (i.e. different 𝑞, 𝑘
or 𝑣̅ ). The crossing of a wave informs drivers that they need to adapt to the new traffic conditions. Traffic waves
evolve in time and space and they have their own trajectories. It is not the trajectory of any vehicle, but the
- 24 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

trajectory of information, the trajectory which locates in time and space the changes in traffic conditions. As an
example, Figure 20 shows (in red) the traffic waves generated when vehicles decelerate to stop in a traffic signal
for accelerating later on, when the signal turns green. Note that if we could predict the evolution of these waves,
we could predict the evolution of traffic states and their transitions, which is precisely the objective of a
macroscopic traffic flow model. In summary, LRW theory deals with the prediction of the trajectories of the traffic
waves. The following link illustrates quite well the concept of a traffic wave (https://youtu.be/Fn3HMAaEfcQ).

Delay

Red

𝑢
∆𝑞
𝑢=
∆𝐾

Figure 20. Example of traffic waves (interphases) in a traffic signal.

Given the previous context, the LWR theory assumes two postulates:

• The conservation of vehicles

• The existence of an equation of state (e.g. 𝑞(𝑘), the flow as a function of the density, which acts as the
state variable). This is the existence of a traffic diagram.

With these two postulates, it is possible to predict the speed of a traffic wave between any two given traffic
states.

6.1. Speed of a traffic wave

LWR theory is based on the prediction of the evolution of traffic waves. So, it is fundamental being able to
determine the speed of a wave from the knowledge of attributes of the "colliding" traffic states. To this end,
consider two different traffic states: 𝑈(𝑞A , 𝑘A ) upstream and 𝐷(𝑞B , 𝑘B ) downstream. In between there needs
to exist a transition, a wave. If traffic states 𝑈 and 𝐷 are stationary (i.e. they do not change neither in time nor

- 25 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

in space), the trajectory defined by the traffic wave is linear (i.e. constant speed). The objective is to determine
the speed, 𝑢, of this traffic wave given the traffic variables defining 𝑈 and 𝐷 (i.e. (𝑞A , 𝑘A ) and (𝑞B , 𝑘B ).

𝒙
𝐷(𝑞 # , 𝑘 # ) Interface
2 B
𝑥2
Traffic states:
• Downstream (𝐷)
• Upstream (𝑈)
𝐿 3
𝑢 4

𝑥1 𝑈(𝑞 $ , 𝑘 $ )
A 1

𝑇 𝒕
𝑡1 𝑡2

Figure 21. Derivation of the speed of a traffic wave from the conservation equation.

To this end, the conservation equation can be used. Look at Figure 21 and recall the derivation process for the
conservation equation. Then, it should be clear that we can formulate the following:

𝑚" + 𝑛/ = 𝑚- + 𝑛0

Where 𝑚 and 𝑛 are the number of vehicle trajectories crossing the borders of the control area in time – space.
Subscripts refer to the particular borders considered.

We can rearrange the terms according to if they belong to a border of the region in 𝑈 or 𝐷. This is:

𝑚" − 𝑛0 = 𝑚- − 𝑛/

Then, express the number of trajectories in terms of flows and densities:

𝑞A 𝑇 − 𝑘A 𝐿 = 𝑞B 𝑇 − 𝑘B 𝐿

𝐿 𝐿
𝑞A − 𝑘A = 𝑞B − 𝑘B
𝑇 𝑇

And note from Figure 21 that, by definition, 𝑢 = 𝐿⁄𝑇, so that:

- 26 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝑞A − 𝑘A 𝑢 = 𝑞B − 𝑘B 𝑢

And finally:

𝑞A − 𝑞B Δ𝑞
𝑢= =
𝑘A − 𝑘B Δ𝑘

The conclusion is that the speed of the shockwave between two stationary traffic states can be determined by
the increase in the flow over the increase in the density of the colliding traffic states. Note that 𝑢 could be positive
(i.e. in the same direction of traffic) or negative (i.e. against traffic), depending on the properties of the colliding
traffic states. This solution is of great importance as it allows to predict the evolution of the interphases between
different traffic states. Note for instance that if 𝑈 was a free-flowing traffic state and 𝐷 a congested one, the
interphase between them would be the end of the queue, and the previous expression would allow to predict
the evolution of the end of the queue (i.e. the queue extension).

6.2. The fundamental diagram of traffic

Different traffic analysis disciplines prefer to use different traffic diagrams in their studies. For instance, traffic
engineers dealing with infrastructural capacity and level of service prefer the 𝑞 – 𝑣 diagram, because this is the
one appearing in the Highway Capacity Manual. Transportation planners use the 𝑡𝑟𝑎𝑣𝑒𝑙 𝑡𝑖𝑚𝑒 – 𝑞 diagram,
usually in the form of the BPR function2. Traffic researchers working on microscopic modeling of traffic often use
the spacing – speed diagram (𝑠-𝑣) because of its direct implications on car-following, and academics working
with macroscopic traffic flow modelling prefer the density – flow diagram (𝑘-𝑞), because it results particularly
easy to derive some relevant properties of traffic dynamics by using this graphical representation. Possibly,
amongst all the possible diagrams, 𝑘-𝑞 is the most widely used and the one that reveals more information by
simple inspection. This is why this diagram is called the Fundamental Diagram of traffic.

2
The Bureau of Public Roads (BPR) function is a classical relationship between the travel time (𝑡𝑡) in a network link as a
function of its travelling flow (𝑞). The traditional expression of the BPR function is:
𝑞 "
𝑡𝑡 = 𝑡𝑡2 H1 + 𝛽 K L M
𝑞/01
Where 𝑡𝑡2 is the free flow travel time and 𝑞/01 the capacity of the link. 𝛽 and 𝑛 are two calibration parameters, whose
classical values are 𝛽 = 0.15 and 𝑛 = 4, but today often varying by link type. Note that the BPR function implies that 𝑡𝑡 is
continuously increasing with 𝑞. While this is valid for free-flowing traffic, it does not represent congested conditions, where
𝑡𝑡 increase for decreasing 𝑞 (see the 𝑝 - 𝑞 diagram in Figure 14). So, the BPR function should be applied with much caution
in oversaturated networks.
- 27 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

1- = 𝑠𝐷
𝐾𝐷 Traffic states:
• Downstream (𝐷)
• Upstream (𝑈)
𝐷(𝑞 # , 𝐾 # )
1- = 𝑠𝑈 −𝑢
𝐾𝑈
𝑈(𝑞 $ , 𝐾 $ )
Interface

Figure 22. Representation of a traffic wave with negative speed (i.e. against traffic direction).

In particular, on the 𝑘-𝑞 diagram, it results particularly easy to visualize the speed of a traffic wave between two
different traffic states. Take for instance the traffic states 𝑈 and 𝐷 shown in Figure 22. Recall that the speed, 𝑢,
of the traffic wave determining the transition between them is obtained as ∆𝑞 ⁄∆𝑘, which in the 𝑘-𝑞 plane
represents the slope of the straight line joining traffic states 𝑈 and 𝐷 (see Figure 23). Note that all “slopes” on
the 𝑘-𝑞 plane have units of speed, since according to the fundamental equation of traffic, 𝑣 = 𝑞 ⁄𝑘.

𝒒
(𝑘 B , 𝑞B ) 𝑞A − 𝑞B Δ𝑞
𝑢= A =
𝑘 − 𝑘B Δ𝑘

∆𝑞 𝑢

∆𝑘
𝑣B (𝑘 A , 𝑞A )

𝑣A

Figure 23. Traffic wave on a flow (q) – density (k) diagram.

- 28 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Then, in the density – flow diagram represented in Figure 24 we can determine:

• The infrastructure capacity, 𝑞$45 , and the related optimal density, 𝑘! .

• The jam density, 𝑘& .
• The free flow speed, 𝑣3 , as the slope of the tangent to the diagram through the origin.
• The speed, 𝑢, of the traffic wave between any two traffic states (e.g. 𝑈 and 𝐷) as the slope of the line
joining both traffic states.

From conservation:
𝑞𝑚𝑎𝑥 ∆𝑞
𝑢"- =
𝑈 ∆𝑘
∆𝑘
𝑞𝑈
𝑢"- ∆𝑞
𝑞𝐷
𝐷
𝑣A
𝑣3 𝑣B

𝑘0 𝒌
𝑘𝑈 𝑘𝐷 𝑘𝑗

Figure 24. Density – flow diagram: the fundamental diagram

Also, the relative flow, 𝑞! , seen by a moving observer travelling at speed 𝑣! has a direct representation on the
fundamental diagram (see Figure 25). It is just needed to draw an auxiliary straight line with slope 𝑣! through
the origin. Then, for any prevailing traffic state 𝐴, the relative flow seen by the observer is identified in the
diagram as the vertical separation between the auxiliary line and the diagram. Note that this construction would
also hold for traffic states 𝐵, where 𝑣C < 𝑣! , and that in such case the relative flow is negative, as the distance
between the auxiliary line and the diagram follows the opposite direction of the coordinate axis for 𝑞.

- 29 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝐴
𝑞𝐴

Observer
𝑞!
𝑞! = 𝑞E − 𝑘E . 𝑣!
𝑣E
𝑣F 𝐵
𝑘E . 𝑣!

𝑘𝐴 𝒌

Figure 25. Relative flow seen by a moving observer on a fundamental diagram.

6.3. Shocks and waves

The concept behind LWR traffic flow theory is simple: at every change in the traffic state a traffic wave is
generated. Then, it is only needed to track these waves in time and space to predict the traffic evolution. This is
the main concept, although its application results more complex as we are going to discuss next.

Only two possible situations can arise in a traffic state change: i) an acceleration transition (i.e. vehicles moving
to a faster traffic state), or ii) a deceleration transition (i.e. vehicles moving to a slower traffic state; joining a
queue for instance). First, consider case i) and assume a group of vehicles travelling behind a slow truck (i.e. a
moving obstruction) until 𝑡" , when the truck exits the road (see Figure 26). Just after 𝑡" , the vehicle immediately
after the truck will start accelerating to recover his free-flow speed. In his acceleration, this vehicle will travel at
all the intermediate speeds between the speed of the truck, 𝑣C (i.e. the one at the start of the acceleration) and
the free-flow speed, 𝑣D (i.e. the one at the end of the acceleration process). In every differential increase of
speed, the traffic state changes, and a traffic wave is generated to inform the following vehicles of this transition.
When these waves sequentially hit the following vehicles, they start accelerating and modifying their speeds
accordingly. Note that the speed of the traffic wave generated by a differential increase in the speed while in
state 𝐵, is determined by the slope of the line tangent to the diagram at 𝐵 (see Figure 27). Given the concave
shape of the fundamental diagram, it is not difficult to visualize that the infinite number of traffic waves
generated by the leading vehicle when accelerating from 𝑣C to 𝑣D always grow in speed (i.e. from very negative,
to zero at capacity, and turning positive afterwards), so that, when represented in the 𝑥 - 𝑡 plane they create a
"fan" of waves. Keep in mind that waves arise from bottlenecks or any other type of disturbance. This is
represented in Figure 26 and Figure 27 showing only a selection of the infinite traffic states generated (i.e. traffic
states C, D, E, F and G). You can see in Figure 26 how the acceleration process is sharp for the first vehicles after
the truck and it is more and more smooth for vehicles further behind Probably you have experienced this
phenomenon while driving in a similar context.
- 30 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝐻
𝑣D
𝐵 Infinite number of
𝑥-
interphases for an
𝑣C infinite number of
different speeds

Fan of
waves

𝑡- 𝒕

Figure 26. Fan of traffic waves in an acceleration process.

𝒒 G F

E
H
D
𝑢=>

C
𝑢?=

𝑣= B
𝑣?

𝒌
Figure 27. Traffic states in an acceleration process.

The trajectories of traffic waves on the 𝑥 - 𝑡 plane define the regions where the different traffic states prevail. In
particular, we know the average speed that vehicles will experience at any point in 𝑥 - 𝑡. This accomplishes our
traffic flow modeling objective, because we can predict the trajectory of any vehicle given its initial position in
the 𝑥 - 𝑡 plane (see Figure 28).

- 31 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝒙 Predicted trajectory

Disturbance
G
𝑥-

D
C
B

𝑡- 𝒕

Figure 28. Solution from a map of traffic waves (accelerating).

The second possible type of traffic transition is a deceleration process. Just consider a situation where vehicles
need to slow down when joining a queue, from free-flow, 𝑣E , to the slow speed in the queue, 𝑣G . As in the
acceleration case, an infinite number of traffic waves are generated at every differential reduction of the
travelling speed. The main difference in this case is that the speeds of the generated waves decrease along the
deceleration process. This can be seen by looking at the slopes of the tangents to the fundamental diagram when
moving from traffic state A to F (see Figure 29). When representing these traffic waves on the 𝑥 - 𝑡 plane, they
do not fan out as in the acceleration case, but they tend to concentrate and collide (see Figure 30). When two
traffic waves collide, the traffic state between them vanishes and a new wave is generated splitting, now, two
very different traffic states, because the existing smooth transition between them has vanished (see Figure 31).
This new strong wave which results from the collision of two "soft" waves receives the name of a "shock", or
more specifically of a "shockwave". So, a shockwave is a strong traffic wave which determines a change of traffic
state not passing for all the neighboring states. Still, the speed of a shockwave is determined by 𝑢 = ∆𝑞⁄∆𝑘, the
difference in flow with respect to the difference in density of the colliding traffic states. By looking at Figure 31
you can realize that traffic waves and shockwaves will keep colliding until all the intermediate states vanish.
Finally, only one shockwave remains, 𝑢EG , splitting traffic states A and F. This means that, while first drivers may
experience a soft deceleration process, later ones will find a sharp change of speed when joining a queue. Again,
this reflects what we experience in real traffic.

- 32 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝑢𝐶𝐷
𝑢?= C D
B Shockwaves' speed
𝑢A?
A E

𝑣@

Figure 29. Traffic states in a deceleration process.

𝒙
Leader

𝑣𝐸 𝑣𝐹
𝑣𝐷 The waves' speed is
determined according to
𝑣𝐶
∆𝑞 ⁄∆𝑘 on the 𝑘-𝑞 diagram
𝑣𝐵
Waves
𝑢A?
Spacing according to 𝑘-𝑞
𝑣𝐴

Figure 30. Solution from a map of traffic waves (decelerating).

- 33 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝒙
𝑢
>
F

𝑣𝐸 𝑣𝐹 𝑢=>
𝑣𝐷 𝑢A> 𝑢AF
𝑣𝐶
𝑢 ?= 𝑢A=
𝑣𝐵
𝑢A@ 𝑣𝐹
𝑢 A? Finally:
𝑢A?
𝑣𝐴

Figure 31. Shockwave propagation.

6.4. Simplifications
It is clear that the LWR theory as it is, with infinite number of shocks and waves, is not convenient for being
applied in real practice. To solve these inconveniences two assumptions (or simplifications) are made. These are:

• Consider acceleration / deceleration processes as instantaneous

• Assume a triangular fundamental diagram

Instantaneous accelerations
Assuming instantaneous acceleration / deceleration processes allows simplifying the application of LWR theory,
especially in the deceleration processes where an infinite number of waves collide and create new shocks. Note
that if the deceleration process is neglected, and therefore vehicles' trajectories are piecewise linear (as in Figure
32), then only one shock is created (i.e. the shock between the initial and the final traffic states). This means that
the smooth deceleration transition, experienced by some of the initial drivers after the disturbance, is neglected.
Actually, this does not imply a severe drawback, as this transition between multiple traffic states take place only
on a very limited 𝑥 - 𝑡 region. The evolution of the traffic states that persist in time (e.g. tracking the evolution
of queues) is not affected much by this simplification, which in contrast simplifies a lot the application of the
theory.

- 34 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝒙
Vehicle “𝑖”

Convert trajectories to
piecewise linear functions

𝑢𝐴𝐹

Shock

∆𝑡
Small 𝒕

Figure 32. Assumption of piecewise linear trajectories.

Triangular fundamental diagram

The assumption of instantaneous accelerations simplifies a lot the application of LWR theory, especially for
decelerations. However, the same solution is not valid for accelerations, because waves do not collide, but fan
out covering a vast region in 𝑥 - 𝑡. It would not be realistic to simplify this growing fan into a single shock.

In the mid 1990's, prof. Gordon F. Newell at the University of California, Berkeley proposed to use a triangular
diagram in the application of the LWR theory. Newell justified that the triangular diagram is not only easier to
work with, but actually it is more realistic to reproduce the behavior of today's real highway traffic. In fact,
diagrams that are calibrated today with many available data points, resemble the triangular shape, possibly with
the only exception of not having such a sharp vertex at capacity, which typically is smoother.

The benefit of using a triangular fundamental diagram is that between any pair of congested traffic states, the
wave speed is the same, 𝑤 (see Figure 33 and Figure 34). 𝑤 is a parameter of the triangular fundamental diagram
named as the "characteristic wave speed" between congested traffic states. The same happens between any
two free-flowing traffic states. The free-flowing branch of the diagram is linear (and with slope 𝑣3 ), meaning that
any traffic state on this branch will travel at the free-flow speed 𝑣3 , and actually, any wave between free-flowing
traffic states will also travel at 𝑣3 . This implies that free-flowing waves do not cross the trajectory of any vehicle,
but travel with them (see Figure 35).

In conclusion, with the simplification of instantaneous accelerations and with the assumption of a triangular
fundamental diagram, the application of the LWR theory with pen and paper to simple traffic flow problems is
straightforward, as we are going to see in the next example.
- 35 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝑞/01

2
D
1
𝑣> B

𝑣? C
𝑣2
𝑣= 𝑤

𝑘. 𝒌

Figure 33. Triangular fundamental diagram.

𝒙
Vehicle “𝑖”

𝑣𝐷

𝑣𝐶 𝑤

𝑤
𝑣𝐵

Figure 34. Shockwaves between three different congested traffic states (B, C and D) assuming a triangular
fundamental diagram.

- 36 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝒙 Wave

ℎ- ℎ5

𝒕
Figure 35. Shockwave between two free flowing traffic states (1 and 2) assuming a triangular fundamental
diagram.

7. Example of application of the LWR theory: Incident on a freeway

A traffic flow theory problem is posed by knowing the initial conditions (i.e. the initial traffic states), the contour
conditions (i.e. the traffic demand during the analysis period), and the fundamental diagrams for all the sections
of the infrastructure under analysis. Given these inputs, the objective is to predict the evolution of shockwaves
and traffic states in in 𝑥 - 𝑡, paying special attention to the evolution and physical extension of queues. The
procedure of solving a traffic problem with pen and paper using the LWR model is illustrated with an example in
this section.

Consider a 𝑞E = 6000 veh/h traffic flow travelling on a 3 lane freeway. At time 𝑡 = 0 this traffic flow is disrupted
by a vehicle breakdown at the location 𝑥 = 0. This incident blocks one lane during 30 min. (i.e. from 𝑡! = 0 until
𝑡" = 30 𝑚𝑖𝑛). Figure 37 represents the flow (𝑞) – density (𝑘) diagram for one lane of the freeway.

Queue evolution 𝑥=0

𝑞E
𝑞B , 𝑘B , 𝑣B
𝑘E Queue 𝑞B < 𝑞E

𝑣E
𝑞C = 𝑞B 30 min. duration
𝑘C
𝑣C
Figure 36. Incident on a three lane freeway.
- 37 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

𝒒 (veh/h)

2200

22 km/h
110 km/h

𝒌 (veh/Km)
20 120
Figure 37. One lane fundamental diagram (veh/km)

And the questions to answer are:

• Does this incident generate queues? Why?

• How far does the queue reach?
• When does the queue dissipate?
In order to answer these questions, the first thing to do is to obtain the fundamental diagrams for all the sections
in the problem. Note that we have a 3-lane freeway, except at the location of the incident that we have 2-lanes.
So, we need the diagrams for the 3-lane and for the 2-lane sections. From the 1-lane fundamental diagram
provided, we can construct the other diagrams assuming that the diagram for a 𝑛-lane section exhibits the same
speed for 𝑛 times the density. This is a common assumption that allows us to draw the three and the two-lane
sections' diagram on the same plot (see Figure 38).

𝒒 (veh/h)
6600

4400 3 lanes

2 lanes

𝑤 = 22 km/h
110 km/h

𝒌 (veh/Km)
40 60 240 360
Figure 38. Two and three lane fundamental diagrams (veh/km)

- 38 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Note that the problem considers triangular fundamental diagrams. As commented, this is the common
simplification in the application of LWR theory, initially proposed by prof. G.F. Newell in the 1990's, that not only
simplifies the problem, but also triangular diagrams are in general more accurate that the previously used
parabolic shapes.
The solution to the problem consists in predicting the evolution of traffic states in (𝑥, 𝑡). It is recommended to
locate the 𝑥 = 0 reference at a relevant point in the problem, which allows to visualize upstream and
downstream traffic (i.e. somewhere in the middle of the (𝑥, 𝑡) diagram, as in Figure 39).

30 min.
t0 t1 𝒕

(ve

Figure 39. 𝒙 - 𝒕 diagram where to draw the solution to the problem.

Next step consists in identifying the initial and contour conditions. This is critical, as it will determine the solution.
The identification of the intervening traffic states can be done on the fundamental diagrams, as in Figure 40.

𝒒 (veh/h)

C
6600
6000 A

D B
4400

W = 22 km/h
110 km/h

𝒌 (veh/Km)
40 60 240 360
Figure 40. Traffic states identification on the fundamental diagram.
(veh/km)

- 39 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Traffic state 𝐴 is the traffic demand. This applies before and after the incident for the whole freeway section.
Note that the flow of state 𝐴 is 𝑞E = 6000 veh/h, larger than the capacity of 2 lanes, of 4400 veh/h. This implies
that the incident will create a queue. The bottleneck at the incident location will discharge the maximum possible
flow (i.e. the capacity of 2-lanes, 4400 veh/h). Therefore, the queue that will appear in the 3-lane freeway
upstream of the 2-lane bottleneck will be represented by state 𝐵 in the previous figure. This is a congested traffic
state with the flow equal to 𝑞C = 4400 veh/h (i.e. same as the capacity of 2-lane bottleneck). Downstream of
the bottleneck, the flow still needs to be the same (i.e. conservation), but traffic will be free-flowing on a three
lane section. This is represented by state 𝐷, with 𝑞B = 4400 veh/h. Finally, one additional traffic state needs to
be considered. This is the discharge flow when the full capacity of the 3-lane is recovered (i.e. at 𝑡 = 30𝑚𝑖𝑛).
Queues always discharge at the full available capacity, in this case 6600 veh/h. This is represented by traffic state
𝐶. Once the intervening traffic states have been identified, they can be located on the previous (𝑥, 𝑡) diagram
(see Figure 41).

𝒙
D A
A C
30 min.

t0 t1 𝒕
B C A
A
(ve
Figure 41. Traffic states location on the 𝒙 - 𝒕 diagram.

Finally, it is necessary to identify the shockwaves delimiting each traffic state, which will allow to determine the
traffic evolution in the (𝑥, 𝑡) diagram. This is done with the help of the fundamental diagram and the graphical
derivation of the speeds of the shockwaves (see Figure 42).

𝒒 (veh/h)
∆𝑞 6000 − 4400
C 𝑢A? = = = −15,2𝑘𝑚/ℎ
6600 ∆𝑘 54,5 − 160
6000 A
6000 4400
D B 𝑘A = = 54,5 ; 𝑘? = 360 − = 160𝑣𝑒ℎ/ℎ
4400 110 22

W = 22 km/h
110 km/h

40 60 240 𝒌 (veh/Km)
360
Figure 42. Shockwave identification. (veh/km)

- 40 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

Capacity: Max sustained flow that

can pass through the link when
there is a queue upstream feeding it
𝒙 𝑣$ =110 km/h
and no restriction downstream
D
constraining the flow.
A C
𝑣$ =110 km/h

𝑡𝑞 A
30 min.

t0 t1 t2 𝒕
B w C
A uAB (ve
𝑣$ =110 km/h

𝑥G A

Figure 43. Solution to the traffic problem on the 𝒙 - 𝒕 diagram.

And this is the final solution of the problem, as it allows to predict the evolution of the queued state 𝐵. Note that
the queue grows against traffic at a speed 𝑢EC (i.e. creation shockwave) and it dissolves when the incident is
cleared and also from the incident location and against traffic at a speed 𝑢CH = 𝑤 (i.e. dissolving shockwave).
When these to waves collide, they determine the end of the queued state 𝐵. So, you can observe that the queue
has reached the location 𝑥I upstream of the incident and lasts a duration 𝑡I since the start of the blockage. Also
note that the queue does not discharge completely until 𝑡- , when the effects of the incident at 𝑥! finally end. If
you wish, you could predict the average trajectory of any vehicle, as you know the prevailing speeds in each
traffic state zone. One possible trajectory in blue is shown as an example.
Below, you can find the necessary calculations to determine these times and locations.

𝑢EC · 𝑡I = 𝑤 · J𝑡I − 0,5K

15.2 · 𝑡I = 22J𝑡I − 0,5K

15.2 · 𝑡I − 22 · 𝑡I = −11

𝑡I = 1.6ℎ

𝑥I = 15.2 · 1.6 = 24.5𝑘𝑚

-0.K
𝑡! = 0 𝑡" = 0.5ℎ 𝑡I = 1.6ℎ 𝑡- − 𝑡 = = 0.22 → 𝑡- = 1.8ℎ
""!

- 41 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

8. Limitations of the LWR macroscopic traffic flow theory

The LWR macroscopic traffic flow theory is a powerful and very robust theory to model the evolution of traffic
states. Note that its calibration requirements are minimal, as it is only necessary to calibrate the 3 parameters
of the triangular fundamental diagram. The information provided by the model and its level of accuracy is
admirable given its simplicity and few input parameters.

In spite of this, the LWR model has its limitations, which are going to be described next:

1. The LWR model does not consider differences between vehicles. All the vehicles are assumed to have the
same properties (e.g. same free-flow speed). Density, for instance, is an absolute variable and it is not
stratified by vehicle types with different free-flow speeds. This has some implications in the results, and the
theory does not predict, for instance, that when traffic is opened after a road closure (or red signal)
downstream of the restriction first are seen the faster vehicles. Also, LWR theory cannot predict overtakings
or relative flows. This is a common drawback of macroscopic models, although some attempts have been
done to develop macroscopic traffic flow models with different vehicle classes (e.g. the slugs and rabbits
model by prof. Carlos F. Daganzo). In summary, LWR theory is a bad model for analyzing light traffic when
differential speeds play a role. In contrast, it is good for dense and congested traffic states, when the speeds
of all vehicles are approximately the same.

2. Theory neglects "waves" and only works with "shocks". This is the price to pay for simplifying the model by
neglecting acceleration and deceleration processes. The implications of this simplification in the results are
that shockwaves should not be interpreted as precise trajectories. They should be interpreted as time-space
zones where traffic changes state (i.e. a shock wave should not be interpreted as a thin line, but as a wide
line or region). You cannot predict the evolution of a queue to the precision of a few meters, but only
hundreds of meters. Figure 32 should be interpreted as in Figure 44, where the shock defines a transition
zone.

𝒙
Vehicle “𝑖”

Shockwave transition
zone
∆𝑡
Small 𝒕

Figure 44. Shockwave interpretation as transition zones.

- 42 -
 Francesc Soriguera
Associate Professor

Jordi Girona 1-3, Office B1-114

08034 Barcelona
Tel. (+34) 93 401 72 66
< francesc.soriguera@upc.edu >
< fsoriguera.com >

3. Traffic is always stable according to LWR. The continuous theory of traffic does not predict stop&go behavior
inside congested traffic states. It just predicts the average speed for the congested state, as given by the
fundamental diagram. However, in reality this average speed results from periods when the vehicle is
stopped and periods when the vehicle travels faster. This unstable behavior is not reproduced in the LWR
model. This implies that it is a bad model to analyze anything which is significantly affected by stops and
accelerations. For instance, pollution emissions of traffic depend largely on stops and accelerations. If you
use LWR to estimate emissions, these would be largely underestimated, because constant speeds, which
much smaller emissions, are assumed inside queues instead of stop&go.

In light of these limitations the overall conclusion could be that:

• Macroscopic modeling, and specifically LWR model, is a perfect and robust tool to predict the evolution
of congested traffic states to the precision of hundreds of meters.

• However, it is a bad model to be used in light traffic, or to predict the details of what happens inside of
the queues (i.e. stop&go).

These limitations give rise to microscopic traffic flow models, were the objective is to predict the trajectories of
individual vehicles, with all their details.

- 43 -