TRAFFIC STREAM CHARACTERISTICS

BY FRED L. HALL4

4
Professor, McMaster University, Department of Civil Engineering and Department of Geography, 1280 Main Street West,
Hamilton, Ontario, Canada L8S 4L7.
CHAPTER 2 - Frequently used Symbols
k density of a traffic stream in a specified length of road

L length of vehicles of uniform length

ck constant of proportionality between occupancy and

density, under certain simplifying assumptions

ki the (average) density of vehicles in substream I

qi the average rate of flow of vehicles in substream I

Å average speed of a set of vehicles

A A(x,t) the cumulative vehicle arrival function over

space and time

kj jam density, i.e. the density when traffic is so heavy that

it is at a complete standstill

uf free-flow speed, i.e. the speed when there are no

constraints placed on a driver by other vehicles on the
road
2.
TRAFFIC STREAM CHARACTERISTICS

This chapter describes the various models that have been developments in measurement procedures. That section is
developed to describe the relationships among traffic stream followed by one providing detailed descriptions and definitions
characteristics. Most of the work dealing with these relationships of the variables of interest. Some of the relationships between
has been concerned with uninterrupted traffic flow, primarily on the variables are simply a matter of definition. An example is the
freeways or expressways. Consequently, this chapter will cover relationship between density of vehicles on the road, in vehicles
traffic stream characteristics for uninterrupted flow. In per unit distance, and spacing between vehicles, in distance per
discussing the models, the link between theory and measurement vehicle. Others are more difficult to specify. The final section, on
capability is important since often theory depends on traffic stream models, focuses on relationships among speed,
measurement capability. flow, and concentration, either in two-variable models, or in
those that attempt to deal simultaneously with the three
Because of the importance of measurement capability to theory variables.
development, this chapter starts with a section on historical

2.1 Measurement Procedures

The items of interest in traffic theory have been the following:  measurement over a length of road [usually at least 0.5
kilometers (km)];
 rates of flow (vehicles per unit time);  the use of an observer moving in the traffic stream; and
 speeds (distance per unit time);  wide-area samples obtained simultaneously from a number
 travel time over a known length of road (or sometimes the of vehicles, as part of Intelligent Transportation Systems
inverse of speed, “tardity” is used); (ITS).
 occupancy (percent of time a point on the road is occupied
by vehicles); For each method, this section contains an identification of the
 density (vehicles per unit distance); variables that the particular procedure measures, as contrasted
 time headway between vehicles (time per vehicle); with the variables that can only be estimated.
 spacing, or space headway between vehicles (distance per
vehicle); and The types of measurement are illustrated with respect to a space-
 concentration (measured by density or occupancy). time diagram in Figure 2.1. The vertical axis of this diagram
represents distance from some arbitrary starting point along the
A general notion of these variables, based on the intuitive idea road, in the direction of travel. The horizontal axis represents
self-evident from their names, will suffice for the purposes of elapsed time from some arbitrary starting time. Each line within
discussing their measurement. Precise definitions of these the graph represents the 'trajectory' of an individual vehicle, as
variables are given in Section 2.2. it moves down the road over time. The slope of the line is that
vehicle's velocity. Where lines cross, a faster vehicle has
Measurement capabilities for obtaining traffic data have changed overtaken and passed a slower one. (The two vehicles do not in
over the nearly 60-year span of interest in traffic flow, and more fact occupy the same point at the same time.) Measurement at
so in the past 40 years during which there have been a large a point is represented by a horizontal line across the vehicular
number of freeways. Indeed, measurement capabilities are still trajectories: the location is constant, but time varies.
changing. Five measurement procedures are discussed in this Measurement over a short section is represented by two parallel
section: horizontal lines a very short distance apart. A vertical line
represents measurement along a length of road, at one instant of
 measurement at a point; time, such as in a single snapshot taken from above the road (for
 measurement over a short section (by which is meant less example an aerial photograph). The moving observer technique
than about 10 meters (m); is represented by one of the vehicle trajectories, the heavy line in


  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.1
Four Methods of Obtaining Traffic Data (Modified from Drew 1968, Figure 12.9).

Figure 2.1. Details on each of these methods can be found in the One of the more recent data collection methods draws upon
ITE's Manual of Traffic Engineering Studies (Box 1976). video camera technology. In its earliest applications, video
cameras were used to acquire the data in the field, which was
The wide-area samples from ITS are similar to having a number then subsequently played back in a lab for analysis. In these early
of moving observers at various points and times within the implementations, lines were drawn on the video monitor screen
system. These new developments will undoubtedly change the (literally, when manual data reduction was used). More recently
way some traffic measurements are obtained in the future, but this has been automated, and the lines are simply a part of the
they have not been in operation long enough to have a major electronics. This procedure allows the data reduction to be
effect on the material to be covered in this chapter. conducted simultaneously with the data acquisition.

Except for the case of a stopped vehicle, speeds at a 'point' can

2.1.1 Measurement at a Point be obtained only by radar or microwave detectors. Their
frequencies of operation mean that a vehicle needs to move only
Measurement at a point, by hand tallies or pneumatic tubes, was about one centimeter during the speed measurement. In the
the first procedure used for traffic data collection. This method absence of such instruments for a moving vehicle, a second
is easily capable of providing volume counts and therefore flow observation location is necessary to obtain speeds, which moves
rates directly, and with care can also provide time headways. the discussion to that of measurements over a short section.
The technology for making measurements at a point on freeways
changed over 30 years ago from using pneumatic tubes placed Density, which is defined as vehicles per unit length, does not
across the roadway to using point detectors (May et al. 1963; make sense for a point measurement, because no length is
Athol 1965). The most commonly used point detectors are involved. Density can be calculated from point measurements
based on inductive loop technology, but other methods in use when speed is available, but one would have to question the
include microwave, radar, photocells, ultrasonics, and television meaning of the calculation, as it would be density at a point. In
cameras. the abstract, one would expect that occupancy could be
measured at a point, but in reality devices for measuring


  7 5$)),& 675($0 &+$5$&7(5,67,&6

occupancy generally take up a short space on the roadway. recent data, there has been in the past decade a considerable
Hence volume (or flow rate), headways, and speeds are the only increase in the amount of research investigating the underlying
direct measurements at a point. relationships among traffic stream characteristics, which is
reflected in Section 2.3.

2.1.2 Measurements Over a Short Section

2.1.3 Measurement Along a Length of Road
Early studies used a second pneumatic tube, placed very close to
the first, to obtain speeds. More recent systems have used paired Measurements along a length of road come either from aerial
presence detectors, such as inductive loops spaced perhaps five photography, or from cameras mounted on tall buildings or
to six meters apart. With video camera technology, two detector poles. It is suggested that at least 0.5 kilometers (km) of road
'lines' placed close together provide the same capability for be observed. On the basis of a single frame from such sources,
measuring speeds. Even with such short distances, one is no only density can be measured. The single frame gives no sense
longer dealing strictly with a point measurement, but with of time, so neither volumes nor speed can be measured.
measurement along a section of road, albeit a short section. All
of these presence detectors continue to provide direct Once several frames are available, as from a video-camera or
measurement of volume and of time headways, as well as of from time-lapse photography over short time intervals, speeds
speed when pairs of them are used. can also be measured, often over a distance approximating the
entire section length over which densities have been calculated.
Most point detectors currently used, such as inductive loops or Note however the shift in the basis of measurement. Even
microwave beams, take up space on the road, and are therefore though both density and speed can be taken over the full length
a short section measurement. These detectors produce a new of the section, density must be measured at a single point in time,
variable, which was not available from earlier technology, whereas measurement of speed requires variation over time, as
namely occupancy. Occupancy is defined as the percentage of well as distance. In general, flow refers to vehicles crossing a
time that the detection zone of the instrument is occupied by a point or line on the roadway, for example one end of the section
vehicle. This variable is available because the loop gives a in question. Hence, flow and density refer to different
continuous reading (at 50 or 60 Hz usually), which pneumatic measurement frameworks: flow over time at a point in space;
tubes or manual counts could not do. Because occupancy density over space at a point in time.
depends on the size of the detection zone of the instrument, the
measured occupancy may differ from site to site for identical Despite considerable improvements in technology, and the
traffic, depending on the nature and construction of the detector. presence of closed circuit television on many freeways, there is
It would be possible mathematically to standardize the very little use of measurements taken over a long section at the
measurement of occupancy to a zero-length detection zone, but present time. The one advantage such measurements might
this has not yet happened. For practical purposes, many freeway provide would be to yield true journey times over a lengthy
management systems rely solely on flow and occupancy section of road, but that would require better computer vision
information. (See for example, Payne and Tignor 1978; algorithms (to match vehicles at both ends of the section) than
Collins 1983.) are currently possible. There have been some efforts toward the
objective of collecting journey time data on the basis of the
As with point measurements, short-section data acquisition does details of the 'signature' of particular vehicles or platoons of
not permit direct measurement of density. Where studies based vehicles across a series of loops over an extended distance
on short-section measurements have used density, it has been (Kühne and Immes 1993), but few practical implementations as
calculated, in one of two ways to be discussed in Section 2.3. yet. Persaud and Hurdle (1988b) describe another way to make
use of measurements along a length, following the method
The large quantity of data provided by modern freeway traffic proposed by Makagami et al. (1971). By constructing
management systems is noteworthy, especially when compared cumulative arrival curves at several locations, they were able to
with the quantity of data used in the early development of traffic derive both the average flow rate and the average density within
flow theory. As a consequence of this large amount of relatively a section, and consequently the average speeds through it.


  7 5$)),& 675($0 &+$5$&7(5,67,&6

2.1.4 Moving Observer Method where,

q is the estimated flow on the road in the direction of
There are two approaches to the moving observer method. The interest,
first is a simple floating car procedure in which speeds and travel x is the number of vehicles traveling in the direction of
times are recorded as a function of time and location along the interest, which are met by the survey vehicle while
road. While the intention in this method is that the floating car traveling in the opposite direction,
behaves as an average vehicle within the traffic stream, the y is the net number of vehicles that overtake the survey
method cannot give precise average speed data. It is, however, vehicle while traveling in the direction of interest
effective for obtaining qualitative information about freeway (i.e. those passing minus those overtaken),
operations without the need for elaborate equipment or ta is the travel time taken for the trip against the
procedures. One form of this approach uses a second person in stream,
the car to record speeds and travel times. A second form uses a tw is the travel time for the trip with the stream, and
modified recording speedometer of the type regularly used in t is the estimate of mean travel time in the direction
long-distance trucks or buses. One drawback of this approach of interest.
is that it means there are usually significantly fewer speed
observations than volume observations. An example of this kind Wright (1973) revisited the theory behind this method. His
of problem appears in Morton and Jackson (1992). paper also serves as a review of the papers dealing with the
method in the two decades between the original work and his
The other approach was developed by Wardrop and own. He finds that, in general, the method gives biased results,
Charlesworth (1954) for urban traffic measurements and is although the degree of bias is not significant in practice, and can
meant to obtain both speed and volume measurements be overcome. Wright's proposal is that the driver should fix the
simultaneously. Although the method is not practical for major journey time in advance, and keep to it. Stops along the way
urban freeways, it is included here because it may be of some would not matter, so long as the total time taken is as determined
value for rural expressway data collection, where there are no prior to travel. Wright's other point is that turning traffic (exiting
automatic systems. While it is not appropriate as the primary or entering) can upset the calculations done using this method.
mode of data collection on a busy freeway, there are some useful This fact means that the route to be used for this method needs
points that come out of the literature that should be noted by to avoid major exits or entrances. It should be noted also that a
those seeking to obtain average speeds through the moving car large number of observations are required for reliable estimation
method. The technique has been used in the past especially on of speeds and flow rates; without that, the method has very
urban arterials, for example in connection with identifying limited precision.
progression speeds for coordinated signals.

The method developed by Wardrop and Charlesworth is based 2.1.5 ITS Wide-Area Measurements
on a survey vehicle that travels in both directions on the road.
The formulae allow one to estimate both speeds and flows for Some forms of Intelligent Transportation Systems involve the
one direction of travel. The two formulae are use of communications from specially-equipped vehicles to a
central system. Although the technology of the various
(x  y) communications systems differ, all of them provide for
q
transmission of information on the vehicles' speeds. In some
(ta  tw ) (2.1)
cases, this would simply be the instantaneous speed while
passing a particular reporting point. In others, the information
would be simply a vehicle identifier, which would allow the
y
t
tw (2.2) system to calculate journey times between one receiving location
q
and the next. A third type of system would not be based on fixed
interrogation points, but would poll vehicles regardless of
location, and would receive speeds and location information
back from the vehicles.


  7 5$)),& 675($0 &+$5$&7(5,67,&6

The first system would provide comparable data to that obtained The major difficulty with implementing this approach is that of
by paired loop detectors. While it would have the drawback of establishing locations precisely. Global positioning systems
sampling only a small part of the vehicle fleet, it would also have have almost achieved the capability for doing this well, but they
several advantages. The first of these is that system maintenance would add considerably to the expense of this approach.
and repair would not be so expensive or disruptive as is fixing
broken loops. The second is that the polling stations could be set The limitation to all three systems is that they can realistically be
up more widely than loops currently are, providing better expected to provide information only on speeds. It is
coverage especially away from freeways. notgenerally possible for one moving vehicle to be able to
identify flow rates or densities in any meaningful way. Of
The second system would provide speeds over a length of road, course, with appropriate sensors, each instrumented vehicle
information that cannot presently be obtained without great effort could report on its time and space headway, but it would take a
and expense. Since journey times are one of the key variables of larger sample than is likely to occur in order to have any
interest for ITS route guidance, better information would be an confidence in the calculation of flow as the reciprocal of
advantage for that operational purpose. Such information will reported time headways, or density as the reciprocal of reported
also be of use for theoretical work, especially in light of the spacings. Thus there remains the problem of finding comparable
discussion of speeds that appears in Section 2.2.2. flow or density information to go along with the potentially
improved speed information.
The third system offers the potential for true wide-area speed
information, not simply information at selected reporting points.

2.2 Variables of Interest

In general, traffic streams are not uniform, but vary over both The total elapsed study time is made up of the sum of the
space and time. Because of that, measurement of the variables headways recorded for each vehicle:
of interest for traffic flow theory is in fact the sampling of a
random variable. In some instances in the discussion below, that
M
N
is explicit, but in must cases it is only implicit. In reality, the T
hi (2.4)
traffic characteristics that are labeled as flow, speed, and i
1
concentration are parameters of statistical distributions, not
absolute numbers.
If the sum of the headways is substituted in Equation 2.3 for total
time, T, then it can be seen that the flow rate and the average
2.2.1 Flow Rates headway have a reciprocal relationship with each other:
1
q

N

N


1
Flow rates are collected directly through point measurements,
and by definition require measurement over time. They cannot
T Mh
i
i
1
N
Mh i
h (2.5)
i
be estimated from a single snapshot of a length of road. Flow
rates and time headways are related to each other as follows.
Flow rate, q, is the number of vehicles counted, divided by the Flow rates are usually expressed in terms of vehicles per hour,
elapsed time, T: although the actual measurement interval can be much less.
Concern has been expressed, however, about the sustainability
of high volumes measured over very short intervals (such as 30
N
q
(2.3) seconds or one minute) when investigating high rates of flow.
T The 1985 Highway Capacity Manual (HCM 1985) suggests
using at least 15-minute intervals, although there are also


  7 5$)),& 675($0 &+$5$&7(5,67,&6

situations in which the detail provided by five minute or one is termed the time mean speed, because it is an average of
minute data is valuable. The effect of different measurement observations taken over time.
intervals on the nature of resulting data was shown by Rothrock
and Keefer (1957). The second term that is used in the literature is space mean
speed, but unfortunately there are a variety of definitions for it,
not all of which are equivalent. There appear to be two main
2.2.2 Speeds types of definition. One definition is found in Lighthill and
Whitham (1955), which they attribute to Wardrop (1952), and
Measurement of the speed of an individual vehicle requires is the speed based on the average time taken to cross a given
observation over both time and space. The instantaneous speed distance, or space, D:
of an individual vehicle is defined as

dx x x
ui

limit(t2 t1)  0 2 1 (2.6) us

D
t2 t1
M
dt 1 (2.8)
ti
N i

Radar or microwave would appear to be able to provide speed

measurements conforming most closely to this definition, but
even these rely on the motion of the vehicle, which means they where ti is the time for vehicle I to cross distance D
take place over a finite distance and time, however small those D
may be. Vehicle speeds are also measured over short sections, ti
(2.9)
ui
such as the distance between two closely-spaced (6 m) inductive
loops, in which case one no longer has the instantaneous speed
of the vehicle, but a close approximation to it (except during A similar definition appears in handbooks published by the
rapid acceleration or deceleration). Institute of Transportation Engineers (ITE 1976; ITE 1992), and
in May (1990), and is repeated in words in the 1985 HCM. One
In the literature, the distinction has frequently been made question regarding Equation 2.8 is what the summation is taken
between different ways of calculating the average speed of a set over. Implicitly, it is over all of the vehicles that crossed the full
of vehicles. The kind of difference that can arise from different section, D. But for other than very light traffic conditions, there
methods can be illustrated by the following example, which is a will always be some vehicles within the section that have not
kind that often shows up on high school mathematics aptitude completed the crossing. Hence the set of vehicles to include
tests. If a traveler goes from A to B, a distance of 20 km, at an must necessarily be somewhat arbitrary.
average speed of 80 kilometers per hour (km/h), and returns at
an average speed of 40 km/h, what is the average speed for the The 1976 ITE publication also contains a related definition,
round trip? The answer is of course not 60 km/h; that is the where space mean speed is defined as the total travel divided by
speed that would be found by someone standing at the roadside the total travel time. This definition calls for specifying an
with a radar gun, catching this car on both directions of the explicit rectangle on the space-time plane (Figure 2.1), and
journey, and averaging the two observations. The trip, however, taking into account all travel that occurs within it. This
took 1/4 of an hour one way, and 1/2 an hour for the return, for definition is similar to Equation 2.8 in calling for measurement
a total of 3/4 of an hour to go 40 km, resulting in an average of speeds over a distance, but dissimilar in including vehicles
speed of 53.3 km/h. that did not cover the full distance.

The first way of calculating speeds, namely taking the arithmetic Some authors, starting as far back as Wardrop (1952),
mean of the observation, demonstrate that Equation 2.8 is equivalent to using the
harmonic mean of the individual vehicle speeds, as follows.
ut

1 N
M u
N i
1 i
(2.7)


  7 5$)),& 675($0 &+$5$&7(5,67,&6

vehicle. This is equivalent to assuming that the speed

D D
us

distribution does not change with location (Hurdle 1994). A
1
N
Mt
i
i
1
N
M uD
i i
similar definition of space mean speed, without the isoveloxic
assumption, appears in HRB SR 79 (Gerlough and Chappelle
1 (2.10)

1964, viii): "the arithmetic mean of the speeds of vehicles
1
N
M u1
i i
occupying a given length of lane at a given instant."

Wohl and Martin (1967, 323) are among the few authors who
recognize the difference in definition. They quote the HRB
SR79 definition in a footnote, but use as their definition "mean
The difficulty with allowing a definition of space mean speed as of the speeds of the vehicles traveling over a given length of road
the harmonic mean of vehicle speeds is that the measurement and weighted according to the time spent traveling that length".
over a length of road, D, is no longer explicit. Consequently, the Regardless of the particular definition put forward for space
last right-hand side of Equation 2.10 makes it look as if space mean speed, all authors agree that for computations involving
mean speed could be calculated by taking the harmonic mean of mean speeds to be theoretically correct, it is necessary to ensure
speeds measured at a point over time. Wardrop (1952), Lighthill that one has measured space mean speed, rather than time mean
and Whitham (1955), and Edie (1974) among other authors speed. The reasons for this are discussed in Section 2.2.3.
accepted this use of speeds at a point to calculate space mean Under conditions of stop-and-go traffic, as along a signalized
speed. For the case where speeds do not change with location, street or a badly congested freeway, it is important to distinguish
the use of measurements at a point will not matter, but if speeds between these two mean speeds. For freely flowing freeway
vary over the length of road there will be a difference between traffic, however, there will not be any significant difference
the harmonic mean of speeds at a point in space, and the speed between the two, at least if Equation 2.10 can be taken to refer
based on the average travel time over the length of road. As to speeds taken at a point in space, as discussed above. Wardrop
well, Haight (1963) and Kennedy et al. (1973) note that (1952, 330) showed that the two mean speeds differ by the ratio
measurements at a point will over-represent the number of fast of the variance to the mean of the space mean speed:
vehicles and under-represent the slow ones, and hence give a
)2s
higher average speed than the true average. ut
us  (2.11)
us
The second principal type of definition of space mean speed
involves taking the average of the speeds of all of the vehicles on
a section of road at one instant of time. It is most easily here )2s is defined as (ki(ui us)2/K , ki is the density of sub-
visualized with the example given by Haight (1963, 114): "an stream I, and K is the density of the total stream. When there is
aerial photograph, assuming each car to have a speedometer on great variability of speeds, as for example at the time of
its top." Leutzbach (1972; 1987) uses a similar example. In breakdown from uncongested to stop and go conditions, there
Figure 2.1, this method is represented by the vertical line labeled will be considerable difference between the two. Wardrop
"along a length". Kennedy et al. (1973) use a slightly more (1952) provided an example of this kind (albeit along what must
realistic illustration, of two aerial photographs taken in close certainly have been a signalized roadway -- Western Avenue,
succession to obtain the speeds of all of the vehicles in the first Greenford, Middlesex, England), in which speeds ranged from
photo. Ardekani and Herman (1987) used this method as part of a low of 8 km/h to a high of 100 km/h. The space mean speed
a study of the relationships among speed, flow, and density. was 48.6 km/h; the time mean speed 54.0 km/h. On the basis of
Haight goes on to show mathematically that a distribution of such calculations, Wardrop (1952, 331) concluded that time
speeds collected in this fashion will be identical to the true mean speed is "6 to 12 percent greater than the space mean"
distribution of speeds, whereas speeds collected over time at one speed. Note however, from the high school mathematics
point on the road will not match the true distribution. In deriving example given earlier, that even with a factor of 2 difference in
this however, he assumes an "isoveloxic" model, one in which speeds, there is only about a 13 percent difference between ut
each car follows a linear trajectory in the space time diagram, and us .
and is not forced to change speed when overtaking another


  7 5$)),& 675($0 &+$5$&7(5,67,&6

In uncongested freeway traffic, the difference between the two is more expensive than using single loops. Those systems that
speeds will be quite small. Most vehicles are traveling at very do not measure speeds, because they have only single-loop
similar speeds, with the result that )s2 will be small, while detector stations, sometimes calculate speeds from flow and
us will be relatively large. Gerlough and Huber (1975) provide occupancy data, using a method first identified by Athol (1965).
an example based on observation of 184 vehicles on Interstate In order to describe and comment on Athol’s derivation of that
94 in Minnesota. Speeds ranged from a low of 35.397 km/h to procedure, it is first necessary to define the measurement of
a high of 45.33 km/h. The arithmetic mean of the speeds was occupancy, a topic which might otherwise be deferred to Section
39.862 km/h; the harmonic mean was 39.769 km/h. As 2.3.3.
expected, the two mean speeds are not identical. However, the
original measurements were accurate only to the nearest whole Occupancy is the fraction of time that vehicles are over the
mile per hour. To the accuracy of the original measurements, the detector. For a specific time interval, T, it is the sum of the time
two means are equal. In other words, for relatively uniform flow that vehicles cover the detector, divided by T. For each
and speeds, the two mean speeds are likely to be equivalent for individual vehicle, the time spent over the detector is determined
practical purposes. Nevertheless, it is still appropriate to specify by the vehicle's speed, ui , and its length, Li , plus the length of
which type of averaging has been done, and perhaps to specify the detector itself, d. That is, the detector is affected by the
the amount of variability in the speeds (which can provide an vehicle from the time the front bumper crosses the start of the
indication of how similar the two are likely to be). detection zone until the time the rear bumper clears the end of
the detection zone.
Even during congestion on freeways, the difference is not very
great, as shown by the analyses of Drake et al. (1967). They
occupancy

M (L d)/u
i
i i
calculated both space mean speed and time mean speed for the
T
same set of data from a Chicago freeway, and then regressed one (2.13)

M M
against the other. The resulting equation was Li

1  d 1
u s
1.026 u t 1.890 (2.12) T i ui T i ui

with speeds in miles per hour. The maximum speeds observed Athol then multiplied the second term of this latter equation by
approached 96.6 km/h, at which value of time mean speed, space N (1/N ), and substituted Equations 2.3 and 2.10:
mean speed would be 96.069 km/h. The lowest observed speeds
were slightly below 32.2 km/h. At a time mean speed of 32.2
km/h, the equation would yield a space mean speed of 29.99
occupancy

1
T
M Lu
i
i

i
 d# N # 1
T N
M u1
i i
km/h. However, it needs to be noted that the underlying (2.14)
M
relationship is in fact non-linear, even though the linear model in Li
Equation 2.12 resulted in a high R2. The equation may
1  d# q
T i ui us
misrepresent the amount of the discrepancy, especially at low
speeds. Nevertheless, the data they present, and the equation,
when applied at high speeds, support the result found by Assuming that the "fundamental equation" holds (which will be
Gerlough and Huber (1975): at least for freeways, the practical dealt with in detail in the next section), namely
significance of the difference between space mean speed and
time mean speed is minimal. However, it is important to note q
ku s (2.15)
that for traffic flow theory purists, the only ‘correct’ way to
measure average travel velocity is to calculate space-mean speed this becomes
directly.

Only a few freeway traffic management systems acquire speed

occupancy

1
T
M Lu i
i

i
 d#k (2.16)
information directly, since to do so requires pairs of presence
detectors at each of the detector stations on the roadway, and that


  7 5$)),& 675($0 &+$5$&7(5,67,&6

Noting that T is simply the sum of the individual vehicle mean of the vehicle lengths. On the other hand, if both lengths
headways, Athol made the substitution, and then multiplied top and speeds vary, then the transition from Equation 2.17 to 2.18
and bottom of the resulting equation by 1/N: and 2.19 cannot be made in this simple fashion, and the
relationship between speeds, flows, and occupancies will not be
so clear-cut (Hall and Persaud 1989). Banks (1994) has recently
demonstrated this result in a more elegant and convincing
M Lu i fashion.
occupancy
 d#k
i i
T Another method has recently been proposed for calculating

Mu
1 Li speeds from flow and occupancy data (Pushkar et al. 1994).
This method is based on the catastrophe theory model for traffic
N

 d#k
i i
(2.17) flow, presented in Section 2.3.6. Since explanation of the
1
N
Mh i
procedure for calculating speeds requires an explanation of that
model, discussion will be deferred until that section.
1
N
M L
u
i

 d#k
i i
2.2.3 Concentration
h
Concentration has in the past been used as a synonym for
In order to proceed further, Athol assumed a uniform vehicle density. For example, Gerlough and Huber (1975, 10) wrote,
length, L, which allows the following simplification of the "Although concentration (the number of vehicles per unit length)
equation: implies measurement along a distance...." In this chapter, it
seems more useful to use 'concentration' as a broader term

M uL
encompassing both density and occupancy. The first is a
1
measure of concentration over space; the second measures
N
occupancy
 d#k
i i concentration over time of the same vehicle stream.
h

1 #L# 1
h N
M u1  d # k
i i
Density can be measured only along a length. If only point
measurements are available, density needs to be calculated,
(2.18) either from occupancy or from speed and flow. Gerlough and

L# q  d#k Huber wrote (in the continuation of the quote in the previous
us
paragraph), that "...traffic engineers have traditionally estimated

(Ld )k
ck k concentration from point measurements, using the relationship

Since at a single detector location, d is constant, this equation

means that occupancy and density are constant multiples of each k
q/us ”. (2.20)
other (under the assumption of constant vehicle lengths).
Consequently, speeds can be calculated as
This is the same equation that was used above in Athol's
derivation of a way to calculate speeds from single-loop detector
q # ck data. The difficulty with using this equation to estimate density
us
(2.19) is that the equation is strictly correct only under some very
occupancy
restricted conditions, or in the limit as both the space and time
measurement intervals approach zero. If neither of those
It is useful to note that the results in Equation 2.18, and hence situations holds, then use of the equation to calculate density can
Equation 2.19, are still valid if vehicle lengths vary and speeds give misleading results, which would not agree with empirical
are constant, except that L would have to be interpreted as the measurements. These issues are important, because this


  7 5$)),& 675($0 &+$5$&7(5,67,&6

equation has often been uncritically applied to situations that constant value. The density as measured over one portion of the
exceed its validity. substream may well be different from the density as measured
over a different portion. The ki calculated in Equation 2.21 is not
The equation was originally developed by Wardrop (1952). His the true density of the stream, but only an estimate. If that is the
derivation began with the assumption that the traffic stream case, however, the derivation of Equation 2.20 is in jeopardy,
could be considered to be a number of substreams, "in each of because it calls upon Equation 2.21 subsequently in the
which all the vehicles are traveling at the same speed and form derivation. In short, there appears to be an implicit assumption
a random series" (Wardrop 1952, 327). Note that the of constant spacing as well as constant speed in the derivation.
randomness must refer to the spacing between vehicles, and that
since all vehicles in the substream have constant speed, the Gerlough and Huber (1975) reproduce some of Wardrop's
spacing within the substream will not change (but is clearly not derivation, but justify the key equation,
uniform). Wardrop's derivation then proceeded as follows
(Wardrop 1952, 327-328), where his symbol for speed, vi, has
ki
qi /ui i
1, 2, ...c (2.23)
been replaced by the one used herein, ui):

Consider the subsidiary stream with flow, qi and speed ui. on the basis of "analysis of units" (Gerlough and Huber 1975,
The average time-interval between its vehicles is evidently 10). That is, the units of flow, in vehicles/hour, can be obtained
1/q i , and the distance travelled in this time is u /qi . i It by multiplying the units for density, in vehicles/km, by the units
follows that the density of this stream in space, that is to say, for speed, in km/hour. The fact that space mean speed is needed
the number of vehicles per unit length of road at any instant for the calculation, however, relies on the assumption that the
(the concentration), is given by key equation for substreams holds true. They have not avoided
that dependence.
ki
qi / ui i
1, 2, ...c (2.21) Although Equation 2.20 has been called the fundamental
identity, or fundamental equation of traffic flow, its use has often
The next step involved calculating the overall average speed on exceeded the underlying assumptions. Wardrop's explicit
the basis of the fractional shares of total density, and using the assumption of substreams with constant speed is approximately
above equation to deduce the results: true for uncongested traffic (that is at flows of between 300 and
perhaps 2200 pcphpl), when all vehicles are moving together

us

M ku
i
i i

Mq
i
i

q (2.22)
quite well. The implicit assumption of constant spacing is not
true over most of this range, although it becomes more nearly
accurate as volumes increase. During congested conditions,
k k k
even the assumption of constant speed substreams is not met.
Congested conditions are usually described as stop-and-go
The equation for the substreams is therefore critical to the (although slow-and-go might be more accurate). In other words,
derivation to show that Equation 2.20 above holds when space calculation of density from speed and flow is likely to be
mean speed is used. accurate only over part of the range of operating conditions.
(See also the discussion in Hall and Persaud 1989.)
There are two problems with this derivation, both arising from
the distinction between a random series and its average. Equation 2.20, and its rearranged form in Equation 2.15,
Wardrop is correct to say that the "average time-interval" explicitly deal in averages, as shown by us in both of them. The
between vehicles (in a substream) is 1/qi, but he neglects to same underlying idea is also used in theoretical work, but there
include the word 'average' in the next sentence, about density: the relationship is defined at a point on the time space plane. In
the average density of this stream in space. The issue is not that context, the equation is simply presented as
simply one of wording, but of mathematics. Because the
q
uk (2.24)
substream is random, which must mean random spacing since it
has a uniform speed, the density of the substream cannot have a

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

This equation "may best be regarded as an idealization, which is occupancy, using the relationship identified by Athol, as derived
true at a point if all measures concerned are regarded as in the previous section (Equation 2.18). That equation is also
continuous variables" (Banks 1994, 12). Both Banks and Newell valid only under certain conditions. Hall and Persaud (1989)
(1982, 60) demonstrated this by using the three-dimensional identified those as being either constant speeds or constant
surface proposed by Makagami et al. (1971), on which the vehicle lengths. Banks has identified the conditions more
dimensions are time (t); distance (x); and cumulative number of precisely as requiring both the covariance of vehicle length with
vehicles (N). If one assumes that the discrete steps in N can be the inverse of vehicle speed and the covariance of vehicle
smoothed out to allow treatment of the surface A(x,t), spacing with the inverse of vehicle speed to be zero. Speeds
representing the cumulative vehicle arrival function, as a within a lane are relatively constant during uncongested flow.
continuous function, then Hence the estimation of density from occupancy measurements
is probably reasonable during those traffic conditions, but not
q

0A during congested conditions. Athol's (1965) comparison of

(2.25)
0t occupancy with aerial measurements of density tends to confirm
this generalization. In short, once congestion sets in, there is
k

0A probably no good way to estimate density; it would have to be

(2.26)
0x measured.

Temporal concentration (occupancy) can be measured only over

a short section (shorter than the minimum vehicle length), with
presence detectors, and does not make sense over a long section.
and Perhaps because the concept of density has been a part of traffic
measurement since at least the 1930s, there has been a

0x consensus that density was to be preferred over occupancy as the

(2.27) measure of vehicular concentration. For example, Gerlough and
0t
Huber (1975, 10) called occupancy an "estimate of density", and
as recently as 1990, May referred to occupancy as a surrogate for
Since density (p. 186). Almost all of the theoretical work done prior
to 1985 either ignores occupancy, or else uses it only to convert
0A
0x # 0A to, or as a surrogate for, density. Athol's work (1965) is a
(2.28)
0t 0t 0x notable exception to this. On the other hand, much of the
freeway traffic management work during the same period (i.e.
practical as opposed to theoretical work) relied on occupancy,
It follows that q=uk for the continuous surface, at a point. Real and some recent theoretical work has used it as well. (See in
traffic flows, however, are not only made up of finite vehicles particular Sections 2.3.5 and 2.3.6 below.)
surrounded by real spaces, but are inherently stochastic (Newell
1982). Measured values are averages taken from samples, and It would be fair to say that the majority opinion at present
are therefore themselves random variables. Measured flows are remains in favor of density, but that a minority view is
taken over an interval of time, at a particular place. Measured thatoccupancy should begin to enter theoretical work instead of
densities are taken over space at a particular time. Only for density. There are two principal reasons put forward by the
stationary processes (in the statistical sense) will the time and minority for making more use of occupancy. The first is that
space intervals be able to represent conditions at the same point there should be improved correspondence between theoretical
in the time-space plane. Hence it is likely that any measurements and practical work on freeways. If freeway traffic management
that are taken of flow and density (and space mean speed) will makes extensive use of a variable that freeway theory ignores,
not be very good estimates of the expected values that would be the profession is the poorer. The second reason is that density,
defined at the point of interest in the time space plane -- and as vehicles per length of road, ignores the effects of vehiclength
therefore that Equation 2.22 (or 2.15) will not be consistent with and traffic composition. Occupancy, on the other hand, is
the measured data. Density is also sometimes estimated from directly affected by both of these variables, and therefore gives

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

a more reliable indicator of the amount of a road being used by impossible to measure directly in Newton's time, and difficult to
vehicles. There are also good reasons put forward by the measure even indirectly, yet he built a theory of mechanics in
majority for the continued use of density in theoretical work. Not which it is one of the most fundamental parameters. In all
least is that it is theoretically useful in their work in a way that likelihood there will continue to be analysts who use each of
occupancy is not. Hurdle (1994) has drawn an analogy with the occupancy and density: this is a debate that will not be resolved.
concept of acceleration in Newtonian physics. Acceleration was

2.3 Traffic Stream Models

This section provides an overview of work to establish the traffic at location A. These vehicles can be considered to be in
relationship among the variables described in the previous a queue, waiting their turn to be served by the bottleneck section
section. Some of these efforts begin with mathematical models; immediately downstream of the entrance ramp. The data
others are primarily empirical, with little or no attempt to superimposed on graph A reflect the situation whereby traffic at
generalize. Both are important components for understanding A had not reached capacity before the added ramp volume
the relationships. Two aspects of these efforts are emphasized caused the backup. There is a good range of uncongested data
here: the measurement methods used to obtain the data; and the (on the top part of the curve), and congested data concentrated
location at which the measurements were obtained. Section in one area of the lower part of the curve. The volumes for that
2.3.1 discusses the importance of location to the nature of the portion reflect the capacity flow at B less the entering ramp
data obtained; subsequent sections then deal with the models, flows.
first for two variables at a time, then for all three variables
simultaneously. At location B, the full range of uncongested flows is observed,
right out to capacity, but the location never becomes congested,
in the sense of experiencing stop-and-go traffic. It does,
2.3.1 Importance of Location to however, experience congestion in the sense that speeds are
below those observed in the absence of the upstream congestion.
the Nature of the Data
Drivers arrive at the front end of the queue moving very slowly,
and accelerate away from that point, increasing speed as they
Almost all of the models to be discussed represent efforts to
move through the bottleneck section. This segment of the speed-
explain the behavior of traffic variables over the full range of
flow curve has been referred to as queue discharge flow (Hall et
operation. In turning the models from abstract representations
al. 1992). The particular speed observed at B will depend on
into numerical models with specific parameter values, an
how far it is from the front end of the queue (Persaud and Hurdle
important practical question arises: can one expect that the data
1988a). Consequently, the only data that will be observed at B
collected will cover the full range that the model is intended to
are on the top portion of the curve, and at some particular speed
cover? If the answer is no, then the difficult question follows of
in the queue discharge segment.
how to do curve fitting (or parameter estimation) when there
may be essential data missing.
If the exit ramp between B and C removes a significant portion
of the traffic that was observed at B, flows at C will not reach
This issue can be explained more easily with an example. At the
the levels they did at B. If there is no downstream situation
risk of oversimplifying a relationship prior to a more detailed
similar to that between A and B, then C will not experience
discussion of it, consider the simple representation of the speed-
congested operations, and the data observable there will be as
flow curve as shown in Figure 2.2, for three distinct sections of
shown in Figure 2.2. None of these locations taken alone can
roadway. The underlying curve is assumed to be the same at all
provide the data to identify the full speed-flow curve. Location
three locations. Between locations A and B, a major entrance
C can help to identify the uncongested portion, but cannot deal
ramp adds considerable traffic to the road. If location B reaches
with capacity, or with congestion. Location B can provide
capacity due to this entrance ramp volume, there will be a
information on the uncongested portion and on capacity.
backup of traffic on the mainstream, resulting in stop-and-go

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.2
Effect of Measurement Location on Nature of Data
(Similar to figures in May 1990 and Hall et al. 1992).

This would all seem obvious enough. A similar discussion 2.3.2 Speed-Flow Models
appears in Drake et al. (1967). It is also explained by May
(1990). Other aspects of the effect of location on data patterns The speed-flow relationship is the bivariate relationship on
are discussed by Hsu and Banks (1993). Yet a number of which there has been the greatest amount of work within the past
important efforts to fit data to theory have ignored this key point half-dozen years, with over a dozen new papers, so it is the first
(ie. Ceder and May 1976; Easa and May 1980). one to be discussed here. This sub-section is structured
retrospectively, working from the present backwards in time.
They have recognized that location A data are needed to fit the The reason for this structure is that the current understanding
congested portion of the curve, but have not recognized that at provides some useful insights for interpreting earlier work.
such a location data are missing that are needed to identify
capacity. Consequently, discussion in the remaining subsections Prior to the writing of this chapter, the Highway Capacity and
will look at the nature of the data used in each study, and at Quality of Service Committee of the Transportation Research
where the data were collected (with respect to bottlenecks) in Board approved a revised version of Chapter 3 of the Highway
order to evaluate the theoretical ideas. As will be discussed with Capacity Manual (HCM 1994). This version contains the speed-
reference to specific models, it is possible that the apparent need flow curve shown in Figure 2.3. This curve has speeds
for several different models, or for different parameters for the remaining flat as flows increase, out to somewhere between half
same model at different locations, or even for discontinuous and two-thirds of capacity values, and a very small decrease in
models instead of continuous ones, arose because of the nature speeds at capacity from those values. The curves in Figure 2.3
(location) of the data each was using.

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.3
Speed-Flow Curves Accepted for 1994 HCM.

do not represent any theoretical equation, but instead represent HCM (Figure 2.6) was for lower speeds and a lower capacity.
a generalization of empirical results. In that fundamental respect, Since it seemed unlikely that speeds and capacities of a freeway
the most recent research differs considerably from the earlier could be improved by replacing grade-separated overpasses or
work, which tended to start from hypotheses about first interchanges by at-grade intersections (thereby turning a rural
principles and to consult data only late in the process. freeway into a multi-lane rural highway), there was good reason
to reconsider the situation for a freeway.
The bulk of the recent empirical work on the relationship
between speed and flow (as well as the other relationships) was Additional empirical work dealing with the speed-flow
summarized in a paper by Hall, Hurdle, and Banks (1992). In it, relationship was conducted by Banks (1989, 1990), Hall and
they proposed the model for traffic flow shown in Figure 2.4. Hall (1990), Chin and May (1991), Wemple, Morris and May
This figure is the basis for the background speed-flow curve in (1991), Agyemang-Duah and Hall (1991) and Ringert and
Figure 2.2, and the discussion of that figure in Section 2.3.1 is Urbanik (1993). All of these studies supported the idea that
consistent with this relationship. speeds remain nearly constant even at quite high flow rates.
Another of the important issues they dealt with is one that had
It is perhaps useful to summarize some of the antecedents of the been around for over thirty years (Wattleworth 1963): is there a
understanding depicted in Figure 2.4. The initial impetus came reduction in flow rates within the bottleneck at the time that the
from a paper by Persaud and Hurdle (1988a), in which they queue forms upstream? Figure 2.4 shows such a drop on the
demonstrated (Figure 2.5) that the vertical line for queue basis of two studies. Banks (1991a, 1991b) reports roughly a
discharge flow in Figure 2.4 was a reasonable result of three percent drop from pre-queue flows, on the basis of nine
measurements taken at various distances downstream from a days of data at one site in California. Agyemang-Duah and Hall
queue. (This study was an outgrowth of an earlier one by Hurdle (1991) found about a 5 percent decrease, on the basis of 52 days
and Datta (1983) in which they raised a number of questions of data at one site in Ontario. This decrease in flow is often not
about the shape of the speed-flow curve near capacity.) Further observable, however, as in many locations high flow rates do not
impetus for change came from work done on multi-lane rural last long enough prior to the onset of congestion to yield the
highways that led in 1992 to a revised Chapter 7 of the HCM stable flow values that would show the drop.
(1992). That research, and the new Chapter 7, suggested a
shape for those roads very like that in Figure 2.4, whereas the The 1994 revision of the figure for the HCM (Figure 2.3)
conventional wisdom for freeways, as represented in the 1985 elaborates on the top part of Figure 2.4, by specifying the curve
for different free-flow speeds. Two elements of these curves

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.4
Generalized Shape of Speed-Flow Curve
Proposed by Hall, Hurdle, & Banks
(Hall et al. 1992).

Figure 2.5
Speed-Flow Data for Queue Discharge Flow at Varied
Distances Downstream from the Head of the Queue
(Modified from Persaud and Hurdle 1988).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.6
1985 HCM Speed-Flow Curve (HCM 1985).

were assumed to depend on free-flow speed: the breakpoints at

which speeds started to decrease from free-flow, and the speeds u
(A e BQ) e C K e DQ (2.29)
at capacity. Although these aspects of the curve were only
assumed at the time that the curves were proposed and adopted, where u is speed, Q is traffic volume, C and D are constant
they have since received some confirmation in a paper by Hall "curvature factors" taking values between 0.2 and 0.003, and A,
and Brilon (1994), which makes use of German Autobahn B, and K are parameters of the model. This function tended to
information, and another paper by Hall and Montgomery (1993) give the kind of result shown by the upper curve in Figure 2.8,
drawing on British experience. despite the fact that the curve does not accord well with the data
near capacity. The lower curve is a polygon representation,
Two empirical studies conducted recently in Germany support based on an assumed shape as well. In Figure 2.8, each point
the general picture in Figure 2.3 quite well. Heidemann and represents a full hour of data, and the graph represents five
Hotop (1990) found a piecewise-linear 'polygon' for the upper months of hourly data. Note that flows in excess of 2200
part of the curve (Figure 2.7). Unfortunately, they did not have veh/hr/lane were sustained on several occasions, over the full
data beyond 1700 veh/hr/lane, and had that only for two lanes hour.
per direction, so could not address what happens at capacity.
Stappert and Theis (1990) conducted a major empirical study of The relevant British data and studies are those that serve as the
speed-flow relationships on various kinds of roads. However, basis for the manual for cost-benefit analyses (COBA9 1981).
they were interested only in estimating parameters for a specific The speed-flow curve in that manual is shown in Figure 2.9.
functional form, While it shows a decline in speed (of 6 km/h) per each additional
1000 vehicles per hour per lane) from the first vehicle on the
road, detailed inspection of the data behind that conclusion (in

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.7
Results from Fitting Polygon Speed-Flow Curves to German Data
(Modified and translated from Heidemann and Hotop 1990).

Martin and Voorhees 1978, and in Duncan 1974) shows that the 1935, in which he derived the following parabolic equation for
data are ambiguous, and could as easily support a slope of zero the speed-flow curve on the basis of a linear speed- density
out to about the breakpoint of 1200 vphpl (Hall and relationship together with the equation, flow = speed  density:
Montgomery 1993). A more recent British study (Hounsell et
u 2)
al. 1992) also supports the notion that speeds remain high even q
kj (u (2.30)
out to capacity flows. Hence there is good international support uf
for the type of speed-flow curve shown in Figure 2.3, and
nothing to contradict the picture put forward there and in Figure
2.4. where uf is the free-flow speed, and kj is the jam density. Figure
2.10 contains that curve and the data it is based on, redrawn.
The problem for traffic flow theory is that these curves are The numbers adjacent to the data points represent the "number
empirically derived. There is not really any theory that would of 100-vehicle groups observed," on Labor Day 1934, in one
explain these particular shapes, except perhaps for Edie et al. direction on a two-lane two-way road (p. 464). In counting the
(1980), who propose qualitative flow regimes that relate well to vehicles on the road, every 10th vehicle started a new group (of
these curves. The task that lies ahead for traffic flow theorists is 100), so there is a 90 percent overlap between two adjacent
to develop a consistent set of equations that can replicate this groups (p. 451). The groups are not independent observations.
reality. The models that have been proposed to date, and will be Equally important, the data have been grouped in flow ranges of
discussed in subsequent sections, do not necessarily lead to the 200 veh/h and the averages of these groups taken prior to
kinds of speed-flow curves that data suggest are needed. plotting. The one congested point, representing 51
(overlapping) groups of 100 observations, came from a different
It is instructive to review the history of depictions of speed-flow roadway entirely, with different cross-section and pavement,
curves in light of this current understanding. Probably the which were collected on a different day.
seminal work on this topic was the paper by Greenshields in

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.8
Data for Four-Lane German Autobahns (Two-Lanes per Direction),
as reported by Stappert and Theis (1990).

Figure 2.9
UK Speed-Flow Curve (Source: COBA9 1981).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

These details are mentioned here because of the importance to analysis of the data, with overlapping groups and averaging prior
traffic flow theory of Greenshields' work. The parabolic shape to curve-fitting, would not be acceptable. The third problem is
he derived was accepted as the proper shape of the curve for that despite the fact that most people have used a model that was
decades. In the 1965 Highway Capacity Manual, for example, based on holiday traffic, current work focuses on regular
the shape shown in Figure 2.10 appears exactly, despite the fact commuters who are familiar with the road, to better ascertain
that data displayed elsewhere in the 1965 HCM showed that what a road is capable of carrying.
contemporary empirical results did not match the figure. In the
1985 HCM, the same parabolic shape was retained (Figure There is a fourth criticism that can be addressed to Greenshields'
2.6), although broadened considerably. It is only with the 1994 work as well, although it is one of which a number of current
revision to the HCM that a different empirical reality has been researchers seem unaware. Duncan (1976; 1979) has shown
accepted. that calculating density from speed and flow, fitting a line to the
speed-density data, and then converting that line into a speed-
In short, Greenshields' model dominated the field for over 50 flow function, gives a biased result relative to direct estimation
years, despite at least three problems. The most fundamental is of the speed-flow function. This is a consequence of three things
that Greenshields did not work with freeway data. Yet his result discussed earlier: the non-linear transformations involved in
for a single lane of traffic was adopted directly for freeway both directions, the stochastic nature of the observations, and the
conditions. (This of course was not his doing.) The second inability to match the time and space measurement frames
problem is that by current standards of research the method of exactly.

Figure 2.10
Greenshields' Speed-Flow Curve and Data
(Greenshields 1935).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

It is interesting to contrast the emphasis on speed-flow models .3.3 Speed-Density Models

in recent years, especially for freeways, with that 20 years ago,
for example as represented in TRB SR 165 (Gerlough and This subsection deals with mathematical models for the
Huber 1975), where speed-flow models took up less than a page
of text, and none of the five accompanying diagrams dealt with u
uf (1 k/ kj) (2.31)
freeways. (Three dealt with the artificial situation of a test
track.) In contrast, five pages and eleven figures were devoted speed-density relationship, going back to as early as 1935.
to the speed-density relationship. Speed-flow models are now Greenshields' (1935) linear model of speed and density was
recognized to be important for freeway management strategies, mentioned in the previous section. It can be written as:The
and will be of fundamental importance for ITS implementation measured data were speeds and flows; density was calculated
of alternate routing; hence there is currently considerably more using Equation 2.20. The most interesting aspect of this
work on this topic than on the remaining two bivariate topics. particular model is that its empirical basis consisted of half a
Twenty years ago, the other topics were of more interest. As dozen points in one cluster near free-flow speed, and a single
Gerlough and Huber stated (p. 61), "once a speed-concentration observation under congested conditions (Figure 2.11). The linear
model has been determined, a speed-flow model can be relationship comes from connecting the cluster with the single
determined from it." That is in fact the way most earlier speed- point. As Greenshields stated (p. 468), "since the curve is a
flow work was treated (including that of Greenshields). Hence, straight line it is only necessary to determine accurately two
it is sensible to turn to discussion of speed-concentration models, points to fix its direction." What is surprising is not that such
and to deal with any other speed-flow models as a consequence simple analytical methods were used in 1935, but that their
of speed-concentration work, which is the way they were results (the linear speed-density model) have continuedto be so
developed. widely accepted for so long. While there have been studies that

Figure 2.11
Greenshields' Speed-Density Graph and Data (Greenshields 1935).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

claimed to have confirmed this model, such as that in Figure a number of studies that found contradictory evidence, most
2.12a (Huber 1957), they tended to have similarly sparse importantly that by Drake et al. (1967), which will be discussed
portions of the full range of data, usually omitting both the lowest in more detail subsequently.
flows and flow in the range near capacity. There have also been

Figure 2.12
Speed-Density Data from Merritt Parkway and Fitted Curves.

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

A second early model was that put forward by Greenberg in this comparison, and the careful way the work was done, the
(1959), showing a logarithmic relationship: statistical analyses proved inconclusive: "almost all conclusions
were based on intuition alone since the statistical tests provided
u
cln(k/kj) (2.32) little decision power after all" (Drake et al., p. 76). To assert
that intuition alone was the basis is no doubt a bit of an
His paper showed the fit of the model to two data sets, both of exaggeration. Twenty-one graphs help considerably in
which visually looked very reasonable. However, the first data differentiating among the seven hypotheses and their
set was derived from speed and headway data on individual consequences for both speed-volume and volume-density
vehicles, which "was then separated into speed classes and the graphs.
average headway was calculated for each speed class" (p. 83).
In other words, the vehicles that appear in one data point (speed Figure 2.13 provides an example of the three types of graphs
class) may not even have been traveling together! While a used, in this case the ones based on the Edie model. Their
density can always be calculated as the reciprocal of average comments about this model (p. 75) were: "The Edie formulation
headway, when that average is taken over vehicles that may well gave the best estimates of the fundamental parameters. While its
not have been traveling together, it is not clear what that density R2 was the second lowest, its standard error was the lowest of all
is meant to represent. It is also the case that lane changing was hypotheses." One interesting point with respect to Figure 2.13
not permitted in the Lincoln Tunnel (where the data were is that the Edie model was the only one of the seven to replicate
obtained), so this is really single-lane data rather than freeway capacity operations closely on the volume-density and speed-
data. The second data set used by Greenberg was Huber's. This volume plots. The other models tended to underestimate the
is the same data that appears in Figure 2.12a; Greenberg's graph maximum flows, often by a considerable margin, as is illustrated
is shown in Figure 2.12b. Visually, the fit is quite good, but in Figure 2.14, which shows the speed-volume curve resulting
Huber reported an R of 0.97, which does not leave much room from Greenshields' hypothesis of a linear speed-density
for improvement. relationship. (It is interesting to note that the data in these two
figures are quite consistent with the currently accepted speed-
These two forms of the speed-density curve, plus five others, flow shape identified earlier in Figures 2.3 and 2.4.) The overall
were investigated in an important empirical test by Drake et al. conclusion one might draw from the Drake et al. study is that
in 1967. The test used data from the middle lane of the none of the seven models they tested provide a particularly good
Eisenhower Expressway in Chicago, 800 m (one-half mile) fit to or explanation of the data, although it should be noted that
(upstream from a bottleneck whose capacity was only slightly they did not state their conclusion this way, but rather dealt with
less than the capacity of the study site. This location was chosen each model separately.
specifically in order to obtain data over as much of the range of
operations as possible. A series of 1224 1-minute observations There are two additional issues that arise from the Drake et al.
were initially collected. The measured data consisted of volume, study that are worth noting here. The first is the methodological
time mean speed, and occupancy. Density was calculated from one identified by Duncan (1976; 1979), and discussed earlier
volume and time mean speed. A sample was then taken from with regard to Greenshields' work. Duncan showed that the three
among the 1224 data points in order to create a data set that was step procedure of (1) calculating density from speed and flow
uniformly distributed along the density axis, as is assumed by data, (2) fitting a speed-density function to that data, and then (3)
regression analysis of speed on density. The intention in transforming the speed-density function into a speed-flow
conducting the study was to compare the seven speed-density function results in a curve that does not fit the original speed-
hypotheses statistically, and thereby to select the best one. In flow data particularly well. This is the method used by Drake et
addition to Greenshields' linear form and Greenberg's al., and certainly most of their resulting speed-flow functions did
exponential curve, the other five investigated were a two-part not fit the original speed-flow data very well. Duncan's 1979
and a three-part piecewise linear model, Underwood's (1961) paper expanded on the difficulties to show that minor changes in
transposed exponential curve, Edie's (1961) discontinuous the speed-density function led to major changes in the speed-
exponential form (which combines the Greenberg and flow function. This result suggests the need for further caution
Underwood curves), and a bell-shaped curve. Despite the in using this method of double transformations to calibrate a
intention to use "a rigorous structure of falsifiable tests" (p. 75) speed-flow curve.

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.13
Three Parts of Edie's Hypothesis for the Speed-Density Function,
Fitted to Chicago Data (Drake et al. 1967).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.14
Greenshields' Speed-Flow Function Fitted to Chicago Data (Drake et al. 1967).

The second issue is the relationship between car-following Early studies of highway capacity followed two principal
models (see Chapter 4) and the models tested by Drake et al. approaches. Some investigators examined speed-flow
They explicitly mention that four of the models they tested "have relationships at low concentrations; others discussed
been shown to be directly related to specific car-following rules," headway phenomena at high concentrations. Lighthill and
and cite articles by Gazis and co-authors (1959; 1961). The Whitham (1955) have proposed use of the flow-
interesting question to raise in the context of the overall concentration curve as a means of unifying these two
appraisal of the Drake et al. results is whether the results raise approaches. Because of this unifying feature, and
some questions about the validity of the car-following models for because of the great usefulness of the flow-concentration
freeways. The car-following models gave rise to four of the curve in traffic control situations (such as metering a
speed-density models tested by Drake et al. The results of their freeway), Haight (1960; 1963) has termed the flow-
testing suggest that the speed-density models are not particularly concentration curve "the basic diagram of traffic".
good. Logic says that if the consequences of a set of premises
are shown to be false, then one (at least) of the premises is not Nevertheless, most flow-concentration models have been
valid. It is possible, then, that the car-following models are not derived from assumptions about the shape of the speed-
valid for freeways. This is not surprising, as they were not concentration curve. This section deals primarily with work that
developed for this context. has focused on the flow-concentration relationship directly.
Under that heading is included work that uses either density or
occupancy as the measure of concentration.
2.3.4 Flow-Concentration Models
Edie was perhaps the first to point out that empirical flow-
Although Gerlough and Huber did not give the topic of flow- concentration data frequently have discontinuities in the vicinity
concentration models such extensive treatment as they gave the of what would be maximum flow, and to suggest that therefore
speed-concentration models, they nonetheless thought this topic discontinuous curves might be needed for this relationship. (An
to be very important, as evidenced by their introductory example of his type of curve appears in Figure 2.13.) This
paragraph for the section dealing with these models (p. 55): suggestion led to a series of investigations by May and his
students (Ceder 1975; 1976; Ceder and May 1976; Easa and

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

May 1980) to specify more tightly the nature and parameters of clearly below maximum flows. Although Parts A and B may be
these "two-regime" models (and to link those parameters to the taken to confirm the implicit assumption many traffic engineers
parameters of car-following models). The difficulty with their have that operations pass through capacity prior to breakdown,
resulting models is that the models often do not fit the data well Part C gives a clear indication that this does not always happen.
at capacity (with results similar to those shown in Figure 2.14 for Even more important, all four parts of Figure 2.15 show that
Greenshields' single-regime model). In addition, there seems operations do not go through capacity in returning from
little consistency in parameters from one location to another. congested to uncongested conditions. Operations can 'jump'
Even more troubling, when multiple days from the same site from one branch of the curve to the other, without staying on the
were calibrated, the different days required quite different curve. This same result, not surprisingly, was found for speed-
parameters. flow data (Gunter and Hall 1986).

Koshi et al. (1983) gave an empirically-based discussion of the Each of the four parts of Figure 2.15 show at least one data point
flow-density relationship, in which they suggested that a reverse between the two 'branches' of the usual curve during
lambda shape was the best description of the data (p.406): "the the return to uncongested conditions. Because these were
two regions of flow form not a single downward concave curve... 5-minute data, the authors recognized that these points might be
but a shape like a mirror image of the Greek letter lamda [sic] the result of averaging of data from the two separate branches.
( )". These authors also investigated the implications of this Subsequently, however, additional work utilizing 30-second
phenomenon for car-following models, as well as for wave intervals confirmed the presence of these same types of data
propagation. (Persaud and Hall 1989). Hence there appears to be strong
evidence that traffic operations on a freeway can move from one
Although most of the flow-concentration work that relies on branch of the curve to the other without going all the way around
occupancy rather than density dates from the past decade, Athol the capacity point. This is an aspect of traffic behavior that none
suggested its use nearly 30 years earlier (in 1965). His work of the mathematical models discussed above either explain or
presages a number of the points that have come out subsequently lead one to expect. Nonetheless, the phenomenon has been at
and are discussed in more detail below: the use of volume and least implicitly recognized since Lighthill and Whitham's (1955)
occupancy together to identify the onset of congestion; the discussion of shock waves in traffic, which assumes
transitions between uncongested and congested operations at instantaneous jumps from one branch to the other on a speed-
volumes lower than capacity; and the use of time-traced plots flow or flow-occupancy curve. As well, queuing models (e.g.
(i.e. those in which lines connected the data points that occurred Newell 1982) imply that immediately upstream from the back
consecutively over time) to better understand the operations. end of a queue there must be points where the speed is changing
rapidly from the uncongested branch of the speed-flow curve to
After Athol's early efforts, there seems to have been a dearth of that of the congested branch. It would be beneficial if flow-
efforts to utilize the occupancy data that was available, until the concentration (and speed-flow) models explicitly took this
mid-1980s. One paper from that time (Hall et al. 1986) that possibility into account.
utilized occupancy drew on the same approach Athol had used,
namely the presentation of time-traced plots. Figure 2.15 shows One of the conclusions of the paper by Hall et al. (1986), from
results for four different days from the same location, 4 km which Figure 2.15 is drawn, is that an inverted 'V' shape is a
upstream of a primary bottleneck. The data are for the left-most plausible representation of the flow-occupancy relationship.
lane only (the high-speed, or passing lane), and are for 5-minute Although that conclusion was based on limited data from near
intervals. The first point in the time-connected traces is the one Toronto, Hall and Gunter (1986) supported it with data from a
that occurred in the 5-minute period after the data-recording larger number of stations. Banks (1989) tested their proposition
system was turned on in the morning. In Part D of the figure, it using data from the San Diego area, and confirmed the
is clear that operations had already broken down prior to data suggestion of the inverted 'V'. He also offered a mathematical
being recorded. Part C is perhaps the most intriguing: statement of this proposition and a behavioral interpretation of
operations move into higher occupancies (congestion) at flows it (p. 58):

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.15
Four Days of Flow-Occupancy Data from Near Toronto (Hall et al. 1986).

The inverted-V model implies that drivers maintain a them, provided their speed is less than some critical value.
roughly constant average time gap between their front Once their speed reaches this critical value (which is as fast as
bumper and the back bumper of the vehicle in front of they want to go), they cease to be sensitive to vehicle
spacing....

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

2.3.5 Three-Dimensional Models The original computer work and the photography for the
original report (Gilchrist 1988) were in color, with five
There has not been a lot of work that attempts to treat all three different colors representing different speed ranges. For
traffic flow variables simultaneously. Gerlough and Huber this paper, black and white were alternated for the five
presented one figure (reproduced as Figure 2.16) that speed ranges, which allows each of them to stand out clearly
represented all three variables, but said little about this, other in many figures. The figure (2.19) is a good example....
than (1) "The model must be on the three-dimensional surface u Area A contains the data with speeds above 80 km/hr. Area
= q/k," and (2) "It is usually more convenient to show the model B (light lines) covers the range 70 to 80 km/hr; area C (dark
of (Figure 2.16) as one or more of the three separate lines) the range 60 to 70 km/hr; area D (light lines) the
relationships in two dimensions..." (p. 49). As was noted earlier, range 50 to 60 km/hr; and area E the range below 50 km/hr.
however, empirical observations rarely accord exactly with the
relationship q=u k, especially when the observations are taken One of the conclusions drawn by Gilchrist and Hall was that
during congested conditions. Hence focusing on the two- "conventional theory is insufficient to explain the data", and that
dimensional relationships will not often provide even implicitly the data were more nearly consistent with an alternative model
a valid three-dimensional relationship. based on catastrophe theory (p. 99). A different approach to
three-dimensional modeling was presented by Makagami et al.
these figures, and the letters on Figure 2.19, a quote from the (1971), as discussed in Section 2.2.3 above (and Equations 2.25
original paper is helpful (p. 101). through 2.27). In that model, the dimensions were time,

Figure 2.16
The Three-Dimensional Surface for Traffic Operations, as in Transportation Research Board
Special Report 165 (Gerlough and Huber 1975).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.17
Two-Dimensional Projection of Data Used in Three-Dimensional Study
(Gilchrist and Hall 1989).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.18
One Perspective on the Three-Dimensional
Speed-Flow-Concentration Relationship
(Gilchrist and Hall 1989).

Figure 2.19
Second Perspective on the Three-Dimensional Relationship
(Gilchrist and Hall 1989).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

distance, and cumulative vehicle count. The derivatives of the catastrophe model could represent the speed-flow curve. A
surface representing the cumulative count are speed, flow, and more fruitful model was proposed by Navin (1986), who
density. This three-dimensional model has been applied by suggested that the three-dimensional 'cusp' catastrophe model
Newell (1993) in work on kinematic waves in traffic. In was appropriate for the three traffic variables.
addition, Part I of his paper contains some historical notes on the
use of this approach to modeling. The feature of the cusp catastrophe surface that makes it of
interest in the traffic flow context is that while two of the
One recent approach to modeling the three traffic operations variables (the control variables) exhibit smooth continuous
variables directly has been based on the mathematics of change, the third one (the state variable) can undergo a sudden
catastrophe theory. (The name comes from the fact that while 'catastrophic' jump in its value. Navin suggested that speed
most of the variables being modeled change in a continuous was the variable that underwent this catastrophic change, while
fashion, at least one of the variables can make sudden flow and occupancy were the control variables. While Navin's
discontinuous changes, referred to as catastrophes by Thom presentation was primarily an intuitive one, without recourse to
(1975), who originally developed the mathematics for seven data, Hall and co-authors picked up on the idea and attempted to
such models, ranging from two dimensions to eight.) The first flesh it out both mathematically and empirically. Figure 2.20
effort to apply these models to traffic data was that by Dendrinos shows the current visualization of the model.
(1978), in which he suggested that the two-dimensional

Figure 2.20
Conceptualization of Traffic Operations on a Catastrophe Theory Surface
Using the Maxwell Convention (Persaud and Hall 1989).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Acha-Daza and Hall (1993) compared the effectiveness of the idea to develop a method that could be used for estimating
catastrophe theory model for estimating speeds with four of the speeds from single-loop detector data. The method involves
models discussed above: Greenshields'; Greenberg's; Edie's; estimating parameters for the model from at least one location at
and the double linear regime model. The comparison was done which speeds have been measured, and transferring those
using a data set in which all three variables had been measured, parameters to other nearby locations.
so that the speeds calculated using each model could be
compared with actual measured values. Typical results for the The catastrophe theory model has received some confirmation
catastrophe theory model (Figure 2.21, which yielded an R2 of from its ability to replicate speed measurements. It has two
0.92) can be compared with those for Edie's model (Figure 2.22; added advantages as well on the intuitive level. First, it
R2 of 0.80), which had been found to be best by Drake et al. illustrates graphically that freeway operations do not have to stay
(1967). Although the Greenshields' and double-linear model on the (e.g. speed-flow) curve; jumps are possible from one
resulted in higher R2 values (0.87 and 0.89 respectively) than did branch to the other, and when they occur, there will be sudden
Edie's, both models gave very clustered speed estimates, with changes in speeds. Second, it also illustrates graphically the fact
few predictions in the 60-80 km/h range, and a similar set of that different locations will yield different types of data (see
points below the diagonal in the observed range of 60 to 80 Figure 2.2) in that at some locations the data will go around the
km/h. It is worth noting that real data show very few discontinuity in the surface, while at others the data will cross
observations in the range of 60 - 80 km/h also, so in that respect, directly over the discontinuity. The catastrophe theory model
both models are effective. Pushkar et al. (1994) extended this provides a consistent way to explain

Figure 2.21
Comparison of Observed Speed with Speeds Estimated using
Catastrophe Theory Model (Acha-Daza and Hall 1994).

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Figure 2.22
Comparison of Observed Speeds with Speeds Estimated
by Edie's Model (Acha-Daza and Hall 1994).

2.3.6 Conclusions About from the mid-60s (and earlier) do not measure up to those data
that are available; it is not clear whether the newer models such
Traffic Stream Models
as catastrophe theory will ultimately be any more successful.
The current status of mathematical models for speed-flow-
Despite those words of caution, it is important to note that there
concentration relationships is in a state of flux. The models that
have been significant advances in understanding traffic stream
dominated the discourse for nearly 30 years are incompatible
behavior since the publication in 1975 of the last TRB Special
with the data currently being obtained, and with currently
Report on Traffic Flow Theory (Gerlough and Huber). For
accepted depictions of speed-flow curves, but no replacement
example, the speed-flow relationship shown in Figure 2.3 is
models have yet been developed. Part of the reason is probably
considerably different from the one in the 1965 Highway
that many theoreticians continue to work with density, whereas
Capacity Manual, which was still accepted in 1975. The
the empirical data are in terms of occupancy. The relation
recognition that there are three distinct types of operation, as
between those two measures of concentration is sufficiently weak
shown in Figure 2.4, will affect future analysis of traffic stream
that efforts to transform one into the other only muddy the picture
behavior.
further. The other problem was noted by Duncan (1976; 1979):
transforming variables, fitting equations, and then transforming
Since the appearance of the 1985 Highway Capacity Manual,
the equations back to the original variables can lead to biased
there has been a sizeable amount of research on traffic stream
results, and is very sensitive to small changes in the initial curve-
models, which has led to a different understanding of how traffic
fitting.
operates, especially on freeways. Efforts to implement ITS, with
regard to both traffic management and traffic information
Recognition of three-dimensional relationships is also important
provision, will provide challenges for applying this improvement
for improved understanding. Consequently, it is important to
in understanding. Equally important, ITS will likely provide the
make more use of those sets of freeway data in which all three
opportunity for acquiring more and better data to further advance
variables have been measured and no estimation is needed, and
understanding of these fundamental issues.
to work with practitioners to ensure that there are more data sets
for which all three variables have been measured. The models

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

References
Acha-Daza , J. A. and F. L. Hall (1993). A Graphical Transportation Engineering Studies, edited by D. Robertson,
Comparison of the Predictions for Speed Given by to be published in 1994 by Prentice-Hall for the Institute.
Catastrophe Theory and Some Classic Models. Ceder, A. (1975). Investigation of Two-Regime Traffic
Transportation Research Record 1398, Transportation Flow Models at the Micro- and Macro-Scopic Levels. Ph.D.
Research Board (TRB), National Research Council (NRC), dissertation, University of California, Berkeley, California.
Washington, DC, pp. 119-124. Ceder, A. (1976). A Deterministic Traffic Flow Model for
Acha-Daza, J. A. and F. L. Hall. (1994). The Application of the Two-Regime Approach. Transportation Research Record
Catastrophe Theory to Traffic Flow Variables. 567, TRB, NRC, Washington, DC, pp. 16-32.
Transportation Research Record 28B, NRC, Washington, Ceder, A. and A. D. May (1976). Further Evaluation of
DC, pp. 235-250. Single- and Two-Regime Traffic Flow Models.
Agyemang-Duah, K. and F. L. Hall. (1991). Some Issues Transportation Research Record 567, TRB, NRC,
Regarding the Numerical Value of Freeway Capacity, in Washington, DC, pp. 1-15.
Highway Capacity and Level of Service. Proceedings of the Chin, H. C. and A. D. May (1991). Examination of the
International Symposium on Highway Capacity, Karlsruhe. Speed-Flow Relationship at the Caldecott Tunnel.
(U. Brannolte, ed.) Rotterdam, Balkema, Transportation Research Record 1320, TRB, NRC,
pp. 1-15. Washington, DC.
Ardekani, S. and R. Herman (1987). Urban Network-wide COBA9 (1981). COBA9 (Cost-Benefit Analysis) Manual.
Traffic Variables and their Relations. Transportation U. K. Department of Transport, Assessment Policy and
Science 21, pp. 1-16. Methods Division, London.
Athol, P. (1965). Interdependence of Certain Operational Collins, J. F. (1983). Automatic Incident-Detection
Characteristics Within a Moving Traffic Stream. Highway Experience with Trl Algorithm HIOCC. Transport and Road
Research Record, 72, pp. 58-87. Research Laboratory, TRRL Supplementary Report 775,
Banks, J. H. (1989). Freeway Speed-Flow Concentration Crowthorne, England.
Relationships: More Evidence and Interpretations. Dendrinos, D. S. (1978). Operating Speeds and Volume to
Transportation Research Record, 1225, TRB, NRC, Capacity Ratios: The Observed Relationship and the Fold
Washington, DC, pp. 53-60. Catastrophe. Transportation Research Record 12, TRB,
Banks, J. H. (1990). Flow Processes at a Freeway Bottleneck. NRC, Washington, DC, pp. 191-194.
Transportation Research Record 1287, TRB, NRC, Drake, J. S. , J. L. Schofer, and A. D. May (1967).
Washington, DC, pp. 20-28. A Statistical Analysis of Speed Density Hypotheses.
Banks, J. H. (1991a). Two-Capacity Phenomenon at Freeway Highway Research Record, 154, pp. 53-87.
Bottlenecks: A Basis for Ramp Metering? Transportation Drew, Donald R. (1968). Traffic Flow Theory and Control.
Research Record 1320, TRB, NRC, Washington, DC, New York: McGraw-Hill.
pp. 83-90. Duncan, N. C. (1974). Rural Speed/Flow Relations. Transport
Banks, J. H. (1991b). The Two-Capacity Phenomenon: and Road Research Laboratory, TRRL LR 651, Crowthorne,
Some Theoretical Issues. Transportation Research Record Berkshire, England.
1320, TRB, NRC, Washington, DC, pp. 234-241. Duncan, N. C. (1976). A Note on Speed/Flow/Concentration
Banks, J. H. (1994). The Relationship of Measured to Relations. Traffic Engineering and Control, pp. 34-35.
Calculated Speeds and the "Fundamental Relationship": Duncan, N. C. (1979). A Further Look at Speed/ Flow/
Some Comments. Paper submitted to the Transportation Concentration. Traffic Engineering and Control, pp.
Research Board for its January 1995 meeting. 482-483.
Box, P. C. and J. C. Oppenheimer (1976). Manual of Traffic Easa, S. M. and A. D. May (1980). Generalized Procedures
Engineering Studies, Fourth Edition. Arlington, VA: for Estimating Single- and Two-Regime Traffic Flow
Institute of Transportation Engineers. Note that this edition Models. Transportation Research Record 772, TRB, NRC,
will soon be superseded by the 5th Edition: Manual of Washington, DC, pp. 24-37.

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Edie, L. C. (1961). Car Following and Steady-State Theory Hall, F. L. and B. N. Persaud (1989). An Evaluation of Speed
for Non-Congested Traffic. Operations Research, 9, Estimates Made with Single-Detector Data from Freeway
pp. 66-76. Traffic Management Systems. Transportation Research
Edie, L. C. (1974). Flow theories, in Traffic Science, Record 1232, TRB, NRC, Washington, DC, pp. 9-16.
D. C. Gazis, Editor, Wiley, New York. Hall, F. L. and L. M. Hall (1990). Capacity and Speed Flow
Edie, L. C., R. Herman, and T. Lam (1980). Observed Analysis of the Queue in Ontario. Transportation Research
Multilane Speed Distributions and the Kinetic Theory of Record 1287, TRB, NRC, Washington, DC, pp. 108-118.
Vehicular Traffic. Transportation Science, 14, pp. 55-76. Hall, F. L, V. F. Hurdle, and J. H. Banks (1992). Synthesis of
Gazis, D. C., R. Herman, and R. Potts (1959). Car-Following Recent Work on the Nature of Speed-Flow and Flow-
Theory of Steady-State Traffic Flow. Operations Research, Occupancy (Or Density) Relationships on Freeways.
7, pp. 499-505. Transportation Research Record 1365, TRB, National
Gazis, D. C., R. Herman, and R.W. Rothery (1961). Research Council, Washington, DC, pp. 12-18.
Nonlinear Follow-The-Leader Models of Traffic Flow. Hall, F. L. and F. O. Montgomery (1993). Investigation of
Operations Research, 9, pp. 545-567. an Alternative Interpretation of the Speed-Flow
Gerlough, D. L. and M. J. Huber (1975). Traffic Flow Relationship for UK Motorways. Traffic Engineering and
Theory: a Monograph. Special Report 165, Transportation Control, Volume 34, No. 9, September, pp. 420-425.
Research Board (Washington DC: National Research Hall, F. L. and W. Brilon (1994). Comparison of Uncongested
Council). Speed-Flow Relationships in German Autobahn and North
Gilchrist, R. (1988). Three-Dimensional Relationships in American Freeway Data. Transportation Research Record
Traffic Flow Theory Variables. Unpublished Master's 1457 , TRB, NRC, Washington, DC, pp. 35-42.
degree report, Department of Civil Engineering, McMaster Heidemann, D. and R. Hotop (1990). Verteilung der Pkw-
University, Hamilton, Ontario, Canada. Geschwindigkeiten im Netz der Bundesautobahnen-
Gilchrist, R. S. and F. L. Hall (1989). Three-Dimensional Modellmodifikation und Aktualisierung (Distribution of
Relationships among Traffic Flow Theory Variables. passenger-car speeds on the Autobahn network -- model
Transportation Research Record 1225,TRB, NRC, modifications and actualization). Straße und Autobahn -
Washington, DC, pp. 99-108. Heft 2, pp. 106-113.
Greenberg, H. (1959). An Analysis of Traffic Flow. HCM (1965). Special Report 87: Highway Capacity Manual,
Operations Research, Vol 7, pp. 78-85. Transportation Research Board.
Greenshields, B. D. (1935). A Study of Traffic Capacity. HCM (1985). Special Report 209: Highway Capacity
Highway Research Board Proceedings 14, pp. 448-477. Manual. Transportation Research Board.
Gunter, M. A. and F. L. Hall (1986). Transitions in the HCM (1992). Chapter 7, Multilane Rural and Suburban
Speed-Flow Relationship. Transportation Research Record Highways. A Replacement for the Chapter That Appeared
1091, TRB, NRC, Washington, DC, pp. 18-21. in the Third (1985) Edition of the HCM. Transportation
Haight, F. A. (1960). The Volume, Density Relation in the Research Board. Washington, DC: National Academy of
Theory of Road Traffic. Operations Research 8, pp. 572- Sciences.
573. HCM (1994). Committee A3A10, Highway Capacity and
Haight, F. A. (1963). Mathematical Theories of Traffic Quality of Service. Highway Capacity Manual, Chapter 3,
Flow. (New York: Academic Press) "Basic Freeway Segments". TRB, National Research
Hall, F. L., B. L. Allen, and M. A. Gunter (1986). Empirical Council, Washington, DC, 1993.
Analysis of Freeway Flow-Density Relationships. Hounsell, N. B., S. R. Barnard, M. McDonald, and D. Owens
Transportation Research 20A, TRB, NRC, Washington, DC, (1992). An Investigation of Flow Breakdown and Merge
pp. 197-210. Capacity on Motorways. TRL Contractor Report 338.
Hall, F. L. and M. A. Gunter (1986). Further Analysis of the Transport and Road Research Laboratory, Crowthorne,
Flow-Concentration Relationship. Transportation Research Berkshire.
Record 1091, TRB, NRC, Washington, DC,
pp. 1-9.

  
  7 5$)),& 675($0 &+$5$&7(5,67,&6

Hsu, P. and J. H. Banks (1993). Effects of Location on May, A. D. (1990). Traffic Flow Fundamentals.
Congested-Regime Flow-Concentration Relationships for Prentice-Hall.
Freeways. Transportation Research Record 1398, TRB, May, A. D., P. Athol, W. Parker, and J. B. Rudden (1963).
NRC, Washington, DC, pp. 17-23. Development and Evaluation of Congress Street
Huber, M. J. (1957). Effect of Temporary Bridge on Parkway Expressway Pilot Detection System. Highway Research
Performance. Highway Research Board Bulletin 167, Record, 21, pp. 48-70.
pp. 63-74. Morton, T. W. and C. P. Jackson (1992). Speed/Flow
Hurdle, V. F. (1994). Personal communication. Geometry Relationships for Rural Dual-Carriageways and
Hurdle, V. F. and P. K. Datta (1983). Speeds and Flows on Motorways. TRRL Contractor Report, 279. Transport and
an Urban Freeway: Some Measurements and a Hypothesis. Road Research Laboratory, Crowthorne, Berkshire, England.
Transportation Research Record 905, TRB, NRC, Navin, F. (1986). Traffic Congestion Catastrophes.
Washington, DC, pp. 127-137. Transportation Planning Technology, 11, pp. 19-25.
ITE (1976). Institute of Transportation Engineers. Newell, G. F. (1982). Applications of Queueing Theory,
Transportation and Traffic Engineering Handbook, Second Edition. London: Chapman and Hall.
J.E. Baerwald, Editor. Prentice-Hall. Newell, G. F. (1993). A Simplified Theory of Kinematic
ITE (1992). Institute of Transportation Engineers. Traffic Waves in Highway Traffic. Transportation Research B,
Engineering Handbook, 4th edition, J.L. Pline, Editor. 27B: Part I, General theory, pp. 281-287; Part II, Queueing
Prentice-Hall. at freeway bottlenecks, pp. 289-303; Part III, Multi-
Kennedy, N., J. H. Kell, and W. Homburger (1973). destination flows, pp. 305-313.
Fundamentals of Traffic Engineering, 8th edition. Institute Payne, H. J. and S. C. Tignor (1978). Incident-Detection
of Transportation an Traffic Engineering, Berkeley, CA. Algorithms Based on Decision Trees with States.
Koshi, M., M. Iwasaki, and I. Okhura (1983). Some Findings Transportation Research Record 682, TRB, NRC,
and an Overview on Vehicular Flow Characteristics. Washington, DC, pp. 30-37.
Proceedings, 8th International Symposium on Transportation Persaud, B. N. and F. L. Hall (1989). Catastrophe Theory
and Traffic Flow Theory (Edited by Hurdle, V.F., Hauer, E. and Patterns in 30-Second Freeway Data -- Implications
and Steuart, G.F.) University of Toronto Press, Toronto, for Incident Detection. Transportation Research 23A, TRB,
Canada, pp. 403-426. NRC, Washington, DC, pp. 103-113.
Kühne, R. and S. Immes (1993). Freeway Control Systems Persaud, B. N. and V. F. Hurdle (1988a). Some New Data
for Using Section-Related Traffic Variable Detection. That Challenge Some Old Ideas about Speed-Flow
Proceedings of the Pacific Rim TransTech Conference, Relationships. Transportation Research Record 1194, TRB,
Seattle. New York: American Society of Civil Engineering, NRC, Washington, DC, pp. 191-198.
pp. 56-62. Persaud, B. N. and V. F. Hurdle (1988b). Freeway Data
Leutzbach, W. (1972). Einffuhrung in die Theorie des from Time-Lapse Aerial Photography -- an Alternative
Verkehrsflusses, Springer, Berlin. Approach. Transportation Science, 22, pp. 251-258.
Leutzbach, W. (1987). Introduction to the Theory of Traffic Pushkar, A., F. L. Hall, and J. A. Acha-Daza (1994).
Flow. Springer-Verlag, New York. Estimation of Speeds from Single-Loop Freeway Flow and
Lighthill, M. J. and G. B. Whitham (1955). On Kinematic Occupancy Data Using the Cusp Catastrophe Theory
Waves: Ii. A Theory of Traffic Flow on Long Crowded Model. Transportation Research Record 1457, TRB, NRC,
Roads. Proceedings of the Royal Society A229 No. 1178, Washington, DC, pp. 149-157.
pp. 317-145. Ringert, J. and T. Urbanik, II (1993). Study of Freeway
Makagami, Y., G. F. Newell, and R. Rothery (1971). Three- Bottlenecks in Texas. Transportation Research Record
Dimensional Representation of Traffic Flow. 1398, TRB, NRC, Washington, DC, pp. 31-41.
Transportation Science, Volume 5, No. 3, pp. 302-313. Rothrock, C. A. and L. A. Keefer (1957). Measurement of
Martin & Voorhees Associates. (1978). A Study of Urban Traffic Congestion. Highway Research Board
Speed / Flow Relationships on Rural Motorways and All- Bulletin 156, pp. 1-13.
Purpose Dual Carriageways. Prepared for Department of
Transport: TRRL and Traffic Engineering Division.

  
Back to the main Publication page.

 7 5$)),& 675($0 &+$5$&7(5,67,&6

Stappert, K. H. and T. J. Theis (1990). Aktualisierung der Wardrop, J. G. and G. Charlesworth (1954). A Method of
Verkehrsstärke- /Verkehrsgeschwindigkeitsbeziehungen des Estimating Speed and Flow of Traffic from a Moving
BVWP-Netzmodells (Actualization of the speed-flow Vehicle. Proceedings of the Institution of Civil Engineers,
relations for the federal Transportation Masterplan Part II, Vol. 3, pp. 158-171.
Network). Research Report VU18009V89 Wattleworth, J. A. (1963). Some Aspects of Macroscopic
Heusch/Boesefeldt, Aachen. Freeway Traffic Flow Theory. Traffic Engineering, 34: 2,
Thom, R. (1975). Structural Stability and Morphogenesis. pp. 15-20.
English translation by D.H. Fowler of Stabilite Structurelle Wemple, E. A., A. M. Morris, and A. D. May (1991). Freeway
et Morphogenese. Benjamin: Reading, MA. Capacity and Level of Service Concepts. Highway Capacity
Underwood, R. T. (1961). Speed, Volume, and Density and Level of Service, Proceedings of the International
Relationships: Quality and Theory of Traffic Flow. Yale Symposium on Highway Capacity, Karlsruhe, (U. Brannolte,
Bureau of Highway Traffic, pp. 141-188, as cited in Drake Ed.) Rotterdam: Balkema, pp. 439-455.
et al. 1967. Wright, C. (1973). A Theoretical Analysis of the Moving
Wardrop, J.G. (1952). Some Theoretical Aspects of Road Observer Method. Transportation Research 7, TRB, NRC,
Traffic Research. Proceedings of the Institution of Civil Washington, DC, pp. 293-311.
Engineers, Part II, Volume I, pp. 325-362.