Fire Technology, 41, 271–304, 2005

C 2005 Springer Science + Business Media, Inc. Manufactured in The United States
DOI: 10.1007/s10694-005-4050-y

Fire Spread Between Vehicles in Tunnels:

Effects of Tunnel Size, Longitudinal
Ventilation and Vehicle Spacing
R.O. Carvel, Edinburgh Fire Research Centre, School of Engineering
and Electronics, University of Edinburgh, Edinburgh, EH9 3JL, UK
A.N. Beard and P.W. Jowitt, Civil Engineering Section, School of the Built
Environment, Heriot-Watt University, Edinburgh, EH14 4AS, UK

Published online: 25 October 2005

Abstract. In a number of catastrophic tunnel fire incidents, the fire has often spread from
vehicle to vehicle over large distances, occasionally hundreds of metres. Five possible means
of fire spread are briefly discussed. The paper focuses on fire spread by flame impingement
and investigates the conditions under which flame impingement will occur on a 4 m high HGV
downstream of the initial fire. The study uses Bayesian methods to predict the probability of
impingement at distances from 0 to 20 m downstream, based on experimental tunnel fire data.
The influence of tunnel size, ventilation velocity and vehicle separation on the probability of
impingement are discussed. In general it is shown that impingement on a downstream HGV
is more likely in a smaller tunnel with a higher ventilation velocity. It is suggested that flames
from a HGV fire in a tunnel will almost certainly impinge on a downstream HGV at distances
of up to 20 m, and possibly much greater distances.

Key words: tunnel fires, heavy goods vehicles, flame behaviour, longitudinal ventilation

Introduction
The catastrophic fire incidents in European tunnels in recent years have highlighted the
devastating consequences of fires in tunnels, particularly heavy goods vehicle (HGV) fires.
Two thirds of all fatal tunnel fire incidents have involved one or more HGVs at the outset
(see Table 1). In a number of instances, including the Mont Blanc (1999) and St. Gotthard
(2001) tunnel fire incidents, the fire spread from the initial HGV involved to other vehicles
stopped significant distances away. However the mechanisms of spread are, as yet, poorly
understood.
In the Mont Blanc fire incident several vehicles which were considerable distances away
from the initial fire ultimately became involved in the catastrophe [1]. The distances between
the stopped vehicles which had entered the Mont Blanc tunnel from the French portal varied
up to 45 m, yet the fire was able to spread across these gaps. More remarkably, the fire also
managed to spread over a distance of 290 m towards Italy (against the prevailing wind), to
involve a queue of eight HGVs stopped there. The fire also spread over a distance of 450 m
from the last vehicle to enter from the French portal to a fire engine which was stopped in
272 Fire Technology 2005

TABLE 1
Fatal Vehicle Fires in Road Tunnels (1978–2004)

HGV Vehicles
Tunnel Year Fatalities Initial vehicle(s) involved involved destroyed
√
Velsen, Netherlands 1978 5 HGV & cars; collision 6
√
Nihonzaka, Japan 1979 7 4 HGV & 2 cars; collision 189
√
Kajiwara, Japan 1980 1 HGV; gearbox fire 2
√
Sakai, Japan 1980 5 HGV; collision 10
Caldecott, USA 1982 7 Car, tanker & bus; collision 3
Salang, Afghanistan 1982 Many∗ Multiple vehicles; collision Not known
√
Pecorile, Italy 1983 9 HGV; collision Not known
L’Arme, France 1986 3 Car; collision 1
√
Gumefens, Switzerland 1987 2 Car; collision 8
√
Serra a Ripoli, Italy 1993 4 Car; collision 16
Hugenot, South Africa 1994 1 Bus; gearbox fire 1
√
Pfänder, Austria 1995 3 Car, HGV & bus; collision 3
Isola delle Femmine, Italy 1996 5 Bus, cars & Tanker; collision 22
√
Mont Blanc, France/Italy 1999 39 HGV; engine fire 34
√
Tauern, Austria 1999 12 HGV & cars; collision 40
Gleinalm, Austria 2001 5 Two cars; collision 2
√
St. Gotthard, Austria 2001 11 Two HGV; collision 23
Fløyfjell, Norway 2003 1 Car; collision 1
√
Baregg, Switzerland 2004 1 Two HGVs & a car; collision 3

∗
Reports vary. One source suggests that the fatalities were not due to fire.

the tunnel. Another rescue vehicle, positioned 230 m behind the fire engine, suffered great
heat damage, but did not become involved in the fire [1].
In the St. Gotthard fire incident, the nearest HGV to the initial fire, which involved two
crashed HGVs, was stopped 30 m north of the fire, yet the fire quickly spread across this
distance [2]. The fire then spread across a distance of 48 m from this HGV to the next HGV
stopped in the tunnel. From there it spread along the queue of vehicles stopped in the tunnel
until the actions of the fire brigade halted its progress at the seventh HGV [2].
The analysis of the Mont Blanc incident suggests that convective flow leading to spon-
taneous ignition might explain the fire spread to the fire engine, while transfer of burning
fuels could be an alternative explanation for the spread to the other HGVs [1]. The analysis
of the St. Gotthard fire suggests that a combination of convective flow (including convective
heat transfer and radiation from the smoke) and radiation directly from the flames may have
been responsible for the spread of fire across the large distances between vehicles [2].
In the recent fire experiments in the Runehamar tunnel (2003), the flames from a burning
HGV mock-up were recorded as extending as much as 100 m downstream of the fire [3],
so flame impingement may also be a factor in allowing fire spread across such distances.
In their analysis of the Channel Tunnel fire (1996), Liew et al. [4] and Rew and Deaves
[5] considered five different mechanisms of fire spread. A slightly amended version of that
list is:
1. Flame impingement
2. Flame spread
Fire Spread Between Vehicles in Tunnels 273

3. Spontaneous ignition (in some ways analogous to ‘fiashover’ in compartment fires)

4. Fuel transfer (including flow of burning liquid fuels and ‘fire brands’)
5. Explosion
The original list [5] uses the term ‘remote ignition’ instead of ‘spontaneous ignition’
(S.I.). The latter term is preferred here as it specifies the mechanism more precisely.
‘Remote ignition’ may be interpreted as including other mechanisms in addition to S.I.,
although it is assumed that Rew & Deaves intended it to mean only ‘remote ignition by the
S.I. mechanism’.
These mechanisms of spread are, of course, applicable to all fires in tunnels, not merely
fires in rail tunnels. However, very little research has been devoted to these mechanisms;
only spontaneous ignition and flame impingement have received significant attention.

Spontaneous Ignition
The earliest detailed research into fire spread in tunnels was the study by Beard et al. of
fire spread by spontaneous ignition, leading to the FIRE-SPRINT ‘A’ models; which has
been ongoing for over a decade [6–8]. The models consider a single control volume (CV)
which extends downstream from a fire in a longitudinally ventilated tunnel and contains (or
partially contains) a target object; generally modelled as a HGV trailer. The study uses the
principles of non linear dynamics to predict the conditions under which there exist points of
instability in the system which are associated with a sudden increase in temperature. This
instability in the system is considered to be associated with the onset of ‘major fire spread’,
that is, the spontaneous ignition of the target object. This is, in some ways, analogous
to ‘flashover’ in compartment fires; the sudden transition from a localised fire to a fire
involving the whole enclosure. (It is not, strictly speaking, appropriate to talk of ‘flashover’
in tunnel fires as flashover is defined only in the context of compartment fires.)
The earliest model, FIRE-SPRINT Al, assumes that the flames from the initial fire remain
upwind of the control volume and do not approach the target object. Using this model, the
predicted critical heat release rate (HRR) of the initial fire necessary to bring about major
fire spread to a HGV positioned 6.45 m away (a distance typical of the spacing of HGVs on
carrier wagons in the Channel Tunnel) in a tunnel similar in size and shape to the Channel
Tunnel is 55.2 MW for a longitudinal airflow of 2 ms−1 increasing to 72.6 MW at 2.5 ms−1
[6]. At ventilation velocities above 3 ms−1 , there is no instability predicted at any HRR
and so the fire will not spread by spontaneous ignition, given the assumptions made in the
model.
FIRE-SPRINT A2 and A3 are similar to Al except that they assume that the flames
from the initial fire extend along the ceiling of the tunnel above the target object, without
impinging on it. FIRE-SPRINT A2 assumes the flames along the ceiling to be of uniform
thickness, whereas they are considered to be thicker upwind of the target object in A3.
Using these models, the critical HRRs for spread in the same example scenario are 45.3
MW & 38.6 MW at 2 ms−1 (using A2 and A3, respectively) and the velocities above which
no instability is predicted are 5.8 ms−1 & 6.9 ms−1 (once again, for A2 and A3 respectively)
[7, 8].
Rew and Deaves [5] suggest that flame impingement on the target object will occur at
significantly lower HRR values than those predicted by the FIRE-SPRINT ‘A’ models, and
274 Fire Technology 2005

hence that fire spread by spontaneous ignition is unlikely to be the dominant mechanism of
fire spread, at least in the considered case of the Channel Tunnel.

Flame Impingement
A FIRE-SPRINT model (Bl) has recently been developed which includes the assumption
of flame impingement [9]. This model considers two control volumes, one large one similar
to the CV considered in the FIRE-SPRINT ‘A’ models and a very small one which exists,
between target object and flame, at the point of flame impingement on the target object.
Here, a persistent ‘tongue’ of flame is assumed to impinge on a small section of the target
object. Making this assumption, the critical HRR for spread (in the same example case)
reduces to 13.5 MW at 2 ms−1 , 28.5 MW at 4 ms−1 and 44.8 MW at 6 ms−1 [9]. This
model does not, however, predict whether or not flame impingement will actually occur; it
assumes that persistent flame impingement exists.
Thus far, it has not been accurately known under which conditions (tunnel geometry,
ventilation velocity, vehicle spacing) flames from an initial fire will impinge on a nearby
vehicle.
The published analysis of the Channel Tunnel incident suggests that flame impingement
on vehicles downstream of the fire may only occur within restricted ventilation conditions
[4]. At low ventilation velocities the flames from the initial fire will impinge on the tunnel
ceiling and, although they may extend above the vehicles downstream of the fire, will not
impinge on them directly (this is similar to the assumptions of FIRE-SPRINT A2 & A3).
On the other hand, higher ventilation velocities will tend to direct the flames towards the
other vehicles downstream but may also tend to shorten the flame length.
The analysis presented by Rew and Deaves [5], based on an earlier study of the effects of
wind speed on fires in the open air [10], combined with data from the HGV fire test carried
out as part of the EUREKA tunnel fire test series (Norway, 1992) [11], predicted that the
length of the flames (L) from the downstream edge of a HGV fire (of heat release rate Q) in
a tunnel (with longitudinal ventilation velocity u around the initial HGV; hot the bulk flow
upstream) could be estimated according to the following equation:


−0.4
Q u
L = Lc (1)
Qc uc

where Lc , Qc and uc are the flame length, fire heat release rate and ventilation velocity
from an experimental reference test—in this case the EUREKA HGV test, giving values of
22 m, 120 MW and 10 ms−1 , respectively. (The paper observes that this relationship also
appears to hold for the longitudinally ventilated pool fire experiments carried out in the
Memorial Tunnel (USA, 1995) [12].)
Using Equation (1) and assuming that a 2 ms−1 upstream velocity becomes 2.7 ms−1
around the initial HGV on fire (in the same Channel Tunnel example; increased ventilation
velocity calculated by assuming a HGV to create a blockage of 25%), the flames from
the initial fire would not be expected to extend far enough to impinge on the target HGV
(6.45 m away) at HRRs less than 20.8 MW. This is larger than the critical HRR for spread
assuming impingement and smaller than the critical HRR assuming no impingement, so
Fire Spread Between Vehicles in Tunnels 275

we may expect the fire to spread, in this instance, only if the HRR of the initial fire were to
exceed this value.
Similarly at 4 ms−1 (assumed to be 5.4 ms−1 around the HGV) the flames from the initial
fire will not impinge on the target object at HRRs less than 27.5 MW; this is only 1 MW
less than the critical HRR predicted by FIRE-SPRINT Bl for this case, so the fire would be
expected to spread if the HRR of the initial fire were to exceed 28.5 MW.
Aside from the Channel Tunnel analysis [4, 5], which is based on only a single exper-
imental fire test (i.e. the EUREKA HGV fire test), the only other studies to consider the
variation of flame behaviour with longitudinal ventilation velocity are those of Oka and
Kurioka et al. [13, 14] and Apte et al. [15, 16] which considered burner fires and kerosene
pool fires, respectively. These studies looked at the influence of ventilation on the angle
of flame tilt and flame length, rather than considering impingement directly. The analysis
presented by Kurioka et al. [14] presents equations for predicting flame length and angle of
flame tilt, but these equations are reliant on the assumption that stratification in the tunnel
is not destroyed; which may not be a reasonable assumption in most realistic longitudinally
ventilated tunnel fire scenarios. Apte et al. [16] present experimental observations of flame
tilt and height, but do not estimate any relationships amongst ventilation, HRR and flame
tilt with which to predict flame behaviour.
Clearly there is a great need for better understanding of flame behaviour in longitudinally
ventilated tunnels in order to be able to predict whether or not flame impingement on a
downstream object will occur in a given scenario.

Aims of the Current Study

The complexity of fire behaviour means that deterministic fire models are not entirely
representative of fire behaviour, often probabilistic modelling is necessary. Two ostensibly
similar fires may behave in significantly different ways under seemingly identical con-
ditions. For this reason, the authors have used probabilistic methods to estimate various
aspects of fire behaviour in tunnels in the past [e.g. 17, 18].
In the present study, Bayesian methods have again been used to estimate the probability
of an event (in this case, the probability of flames from an initial fire impinging on a target
object) given a specified tunnel geometry, longitudinal ventilation rate, vehicle spacing and
heat release rate.
The study considers a number of different tunnel geometry scenarios. In this paper, only
two cases will be discussed. These tunnels of interest (TOI) are:

1. A single lane tunnel, based on the geometry of the Channel Tunnel

2. A two lane tunnel, based on a typical European tunnel design

The single lane TOI differs from the actual geometry of the Channel Tunnel in one
significant way. The tunnel is circular with a 7.6 m diameter (like the Channel Tunnel) but
the ‘roadway’ of the tunnel has been set higher than the level of the rails in the Channel
Tunnel so that the position of a HGV in the TOI is comparable to the position of a HGV on
a carrier wagon in the Channel Tunnel. The two tunnel shapes are shown in Figure 1.
The results of this study go hand-in-hand with the results of our previous studies predict-
ing the influence of tunnel geometry and longitudinal ventilation on the heat release rate
276 Fire Technology 2005

Figure 1. The single lane and two lane tunnels of interest.

of a vehicle fire [17, 18]. It is ultimately intended to combine all the results from this and
previous studies into a single model.

Methodology
One of the aims of the study is to produce results which are easily understood, memorable
and as realistic as possible. However, the problem considered here is a complex one, so a
number of simplifications and assumptions have been made at the outset:
1. The top of the target HGV is assumed to be 4 m above the roadway (i.e. 2 m below
the ceiling in the single lane tunnel or 2.4 m below the ceiling in the two lane tunnel).
Results will only be presented in terms of the probability of impingement, P(imp), for
this HGV; the actual degree of impingement will not be considered. (The height of the
base of the HGV above the roadway is not defined).
2. The HRR of the initial fire will be assumed to fall into one of nine ranges, namely
1–2, 2–4, 4–8, 8–16, 16–32, 32–64, 64–128, 128–256 and 256+ MW. Similar flame
behaviour is assumed for all fires falling into one of these ranges, that is, for example, the
flames from a 70 MW fire and a 120 MW fire are assumed to extend to approximately
the same distance downstream, at the same height above the roadway.
3. The ventilation in the tunnel is assumed to be longitudinal at one of four velocities; 1,
2, 3 or 4 ms−1 .
It had been hoped to consider a greater range of ventilation velocities, but unfortunately
there is little experimental evidence for higher ventilation velocities, see below.
Bayes’ Theorem
Bayes’ Theorem (see [17] and the references therein) is an established method for updating
any estimate of probability in the light of a single piece of new evidence. This makes it an
ideal choice when considering problems where there are only limited data available, such
as tunnel fire behaviour. In the discrete form used in this study, Bayes’ Theorem may be
expressed as:

L(E | imp)P (imp)

P (imp | E) = (2)
L(E | imp)P (imp) + L(E | imp)P (imp)

where P(imp | E) is the probability of impingement (from a fire of given size, at 4 m

height above the roadway, at a given distance from the initial fire, subject to a given
Fire Spread Between Vehicles in Tunnels 277

longitudinal ventilation velocity), updated in the light of a new piece of evidence, E. P(imp)
is the estimated probability of impingement, prior to the consideration of evidence E.
L(E | imp) is the ‘likelihood’ of the evidence assuming that impingement occurs. P (imp)
is the probability of non-impingement and L(E | imp) is the ‘likelihood’ of the evidence
assuming that impingement does not occur. By definition, P (imp) + P (imp) = 1. Strictly
speaking, L(E | imp) + L(E | imp) need not equal 1, but estimates of likelihood have been
chosen such that they do.
In previous studies, for example the study of the influence of longitudinal ventilation
on heat release rate [17, 18], Bayesian methods have been used to estimate the probability
distribution of a variable of interest (in that instance the heat release rate coefficient, k) at
discrete ventilation velocities. This approach is not applicable in the current project as there
are too many interdependent variables, including ventilation, heat release rate and vehicle
separation distance.
In the current study it was decided to use a matrix of independent probability values.
Separate calculations of probability are made for each of the nine HRR ranges, at 1 m
increments in separation distance from 0 to 20 m, for each of the four ventilation velocities.
Each of the 756 estimates of probability in this three-dimensional (9 × 21 × 4) matrix is
independent of the others.
In order to use Bayes’ Theorem, two things are required:

1. A ‘prior’ (i.e. initial) estimate of the probability of impingement in each case, and
2. An estimate of the ‘likelihood’ of at least one piece of experimental evidence.

In this study it was assumed that the ‘prior’ probability of impingement at all vehicle
spacings, ventilation rates and heat release rates was 0.5 at the outset. That is, all values in
the prior matrix were 0.5.

Evidence Considered
Only a comparatively small number of fires have been carried out in tunnels under experi-
mental conditions. Furthermore, very few of these tests have been carried out in such a way
as to provide significant information regarding the behaviour of the flames in the vicinity
of the fire, or immediately downstream.
In order to estimate flame behaviour in a fire test it is desirable to have detailed video
footage of the experiment (preferably from the side). Alternatively, results from an extensive
array of temperature sensors in the tunnel may be used to infer flame positions. Preferably,
both video footage and extensive temperature measurements should be obtained in a test.
Unfortunately, these data are not available from most tunnel fire tests. Even in highly
instrumented fire tests, for example the recent fire tests in the Runehamar Tunnel (2003)
[19, 20], the video cameras were positioned facing along the length of the tunnel, so
the flame behaviour is not clear from the video footage, and most of the thermocouples
downstream of the fire were positioned near the ceiling, so it is not possible to estimate
temperatures (and hence the likelihood of flames) up to the height of a HGV [20]. There
were two thermocouples positioned 1.8 m above the roadway, but these were positioned
100 and 250 m downstream from the fire.
Of the other large-scale fire test series that have been carried out, only the Memorial
Tunnel Fire Ventilation Test Program (MTFVTP) [12] has provided useful data for this
278 Fire Technology 2005

study. Data from the first two downstream thermocouple trees from the fire have been used
in this study, see below.
It had been intended to use the temperature data from the EUREKA EU499 “Firetun”
experiments as well, but the published final report [11] does not contain sufficient tem-
perature data from the first few downstream thermocouple trees for the purposes of this
project and it appears that the computer tapes used to store the original recorded data have
become corrupted in the past decade, thus the data are lost to posterity [21]. Data from
the thermocouple tree 34 m downstream of the fires in the EUREKA test series are still
available from SP, Sweden [22], but are not of use in the current project.
Given the lack of temperature data from full scale tunnel fire experiments, it is necessary
to consider data from reduced scale and small scale tunnel fire experiments.
In order for reduced or small scale experiments to be in any way useful, there must be a
well defined similarity between the scale model and the full scale case of interest. If there
is a strong similarity then a scale model can be used to investigate specific aspects of the
behaviour of the fire or smoke, if the similarity is not so strong then the behaviour of the
fire can only provide information in general terms. In order to scale between reality and a
scale model with a strong similarity, it is necessary to consider the gas flow at and around
the fire.
The gas flow around the fire can be described using various dimensionless groups,
particularly the Froude and Reynolds numbers [23]. Ideally both of these numbers should
be the same in the scale-model as in the full scale case of concern. In reality however,
this is not possible and models are usually scaled assuming conservation of the Froude
number only. This is a reasonable assumption to make for all but the smallest scale models
as variations in the Reynolds number are not particularly significant for turbulent flows.
Assuming conservation of the Froude number, the HRR is proportional to (length
scale)5/2 , this means that, for example, a model fire which is half the scale of a full
size tunnel will have a heat release rate 0.18 times that of the full scale fire (0.55/2 = 0.18).
In a similar way, the ventilation velocity in the scale model scales as (length scale)1/2 , so
a full scale airflow velocity of 2 ms−1 will be modelled by a half scale airflow velocity of
1.4 ms−1 . The temperature of the fire will be the same at test scale and full scale.
Three reduced and small scale fire test series have provided sufficient temperature data
to be of use in this project, these are:
1. The “Phase II” fire tests carried out in the fire gallery at the Health and Safety Laboratory
(HSL), Buxton, UK [24].
2. The laboratory scale fire tests carried out by Bakar and Wu at the University of Sheffield,
UK [25, 26].
3. The series of laboratory scale experiments carried out by Oka and Kurioka et al. at
Yokohama National University, Japan [13, 14].
These test series are described below. In each of the tests in these series, many temperature
measurements, at various heights above the roadway, were recorded downstream of the fire,
so it is possible to infer the position of the flames under test conditions.
When scaling the data from the reduced scale experiment up to the size of the tunnels of
interest, two scale factors have been considered: tunnel height and mean hydraulic diameter
(defined as four times the cross-sectional area divided by the perimeter). It has previously
been shown that the mean hydraulic diameter is the most appropriate scaling factor to use in
Fire Spread Between Vehicles in Tunnels 279

critical ventilation velocity calculations [27]. In those instances when there is a significant
difference between the length scales, care has been taken to allow for these variations in
the estimates of likelihood, see below.

Summaries of Experimental Test Series

The MTFVTP Fire Tests with Longitudinal Ventilation An extensive series of diesel pool
fire tests were carried out in a disused two lane road tunnel near Charlestown, West
Virginia, USA in 1995 [12]. All fire tests were located centrally in the tunnel. Fifteen of
these fire tests were carried out with longitudinal ventilation: tests 605, 606A and 617A
were approximately 10 MW in size, tests 607, 608, 615A & 623B were approximately
20 MW, tests 610, 611, 612B, 622B and 624B were approximately 50 MW and tests 615B,
621A and 625B were approximately 100 MW in magnitude. In each test, the configuration
of jet fans in the tunnel was changed several times, so it is possible to estimate flame
behaviour under a range of ventilation conditions from these tests.
The Memorial Tunnel is 8.8 m wide and 7.8 m high at the apex of the ceiling, the tunnel
cross-section is shown in Figure 2(a). (A false ceiling was constructed in the tunnel for
several of the fire tests in the series, however it was not present during any of the fire tests
with longitudinal ventilation). The tunnel has a 3.2% gradient. The first two downstream
thermocouple trees were positioned 12 and 29 m from the “fire location”, which corresponds
to 3.3 and 20.7 m from the downstream edge of the 10 MW pool fires, 5.2 and 22.5 m
from the edge of the 20 MW & 100 MW pool fires and 12.5 & 29.8 m from the edge of
the 50 MW pool fires. Temperatures were recorded centrally in the tunnel, at a number of
heights from the roadway up to the ceiling.
In order to infer flame position, it has been assumed in this project that a temperature of
at least 600◦ C is necessary for flames to exist, see below.
The temperature did not exceed 600◦ C at the first thermocouple tree in any of the
10 MW fire tests, thus it is concluded that the flames did not extend as much as 3.3 m
downstream under any of the tested ventilation conditions. Similarly, the temperature at
the first thermocouple tree did not exceed 600◦ C in any of the 20 MW fire tests. In the
50 MW fire tests, the temperature exceeded 600◦ C at the first thermocouple tree in all tests
except test 610, but only near the ceiling. This temperature was not attained at heights less
than 4 m from the roadway, or at any height on the second thermocouple tree, in any of the
tests. In each of the 100 MW fires, temperatures in excess of 600◦ C were recorded at both
thermocouple trees at heights where impingement on a HGV would occur.
This experimental test series provided 15 pieces of evidence to update the matrix of
probabilities of impingement for the two-lane tunnel of interest.

The HSL “Phase II” Fire Tests. A series of nine longitudinally ventilated fire tests were
carried out in the fire gallery at the Health & Safety Laboratory, Buxton, UK in 1993 [24].
The fire gallery is 366 m long with a cross-sectional area of 5.6 m2 as shown in Figure 2(b).
The height is 2.44 m at the apex and the width is 2.75 m at the widest point, it has a gradient
of 0.1%. Longitudinal ventilation was provided by fans at one end of the tunnel, the other
end was open to the air. The “Phase II” tests were carried out on behalf of Eurotunnel and
there were a number of objects in the tunnel which were intended to simulate the scenario
of a typical train carrying HGV in semi-open carriages. All the objects in the gallery
280 Fire Technology 2005

Figure 2. Cross-sections of each of the four experimental tunnels,

(a) The Memorial Tunnel, (b) the HSL fire gallery, (c) the Sheffield
laboratory apparatus and (d) the Yokohama laboratory apparatus.

(representing a locomotive, an amenity coach, a loader wagon and two carrier wagons with
HGV) were constructed at 1/3 of full scale.
Various different fire loads were tested, these were positioned on the trailer of the first
simulated HGV. An extensive array of thermocouples was positioned both upstream and
downstream of the fire; the downstream thermocouples were positioned between the two
simulated HGVs and at various locations around the downstream HGV model (see Figure 3).

This experimental test series provided eight pieces of evidence (instrument failure in one
test meant that the recorded data were unreliable) to update the matrix of probabilities of
impingement for the single-lane tunnel of interest.
Laboratory Scale Fire Tests at the University of Sheffield. Many laboratory scale tunnel
fire tests were carried out by M.Z. Abu Bakar as part of his PhD studies at the University
of Sheffield, UK [25]. Five different tunnel apparatuses were used to study the variation
in critical ventilation velocity with tunnel geometry, although only data from one of the
apparatuses are considered here. The fire source in all experiments was a propane burner
which was flush with the tunnel floor. Longitudinal ventilation was generated in the ap-
paratus by introducing compressed air through an orifice plate which blocked one end of
the tunnel. Temperature data were recorded at many locations inside the tunnel apparatus.
‘Steady-state’ data from the experiments carried out in the square cross-sectional tunnel
apparatus (0.25 m × 0.25 m × 15 m long) have been kindly made available for use in this
project (see Figure 2(c)).

Figure 3. The configuration of fuel, objects and thermocouples in

HSL test 6; the shaded region indicates the maximum extent of the
flames (i.e. the 600◦ C isotherm).
Fire Spread Between Vehicles in Tunnels 281

This experimental test series provided twelve pieces of evidence to update the matrix of
probabilities of impingement for the single-lane tunnel of interest.
Laboratory Scale Fire Tests at Yokohama National University. Many laboratory scale
tunnel fire tests have also been carried out by Oka et al. at Yokohama National University,
Japan [13, 14]. These tests were specifically designed to investigate flame behaviour in
tunnels with different geometries under a range of ventilation conditions. A number of dif-
ferent tunnel apparatuses were used, each containing an extensive array of thermocouples.
Several different shapes of propane burner were used as fire sources, these were generally
positioned flush with the floor. ‘Steady-state’ data from the experiments carried out in one
of the rectangular cross-sectional tunnel apparatuses (0.6 m wide × 0.3 m high × 5.4 m
long) have been kindly made available for use in this project. (see Figure 2(d)).
This experimental test series provided 41 pieces of evidence to update the matrix of
probabilities of impingement for the two-lane tunnel of interest.
Likelihood of the Evidence
The likelihood of a given piece of experimental evidence can be estimated by considering
the temperature data from that test and asking oneself a question of the form:
What is the likelihood that this test would have occurred, in the way it did, if we assume
that flames from a fire between 16 and 32 MW would definitely impinge on a 4 m
high HGV, positioned 5 m away, in the TOI, with a longitudinal ventilation velocity of
2 ms−1 ?
This form of question is conceptually tricky at first, but with practice it becomes easier
to evaluate the likelihood. In some instances it is clear that the experiment would be very
unlikely to have happened as it did if that assumption were valid, i.e. if impingement in the
case of interest were to occur (hence the likelihood would be very low). In other instances it
is clear that it could have happened quite easily with that assumption (and so the likelihood
would be high). In order to make the process of estimating likelihood systematic, five
discrete values of likelihood have been used in this study: 0.1, 0.3, 0.5, 0.7 and 0.9; ranging
from very low likelihood to very high likelihood. In those instances where the likelihood of
a piece of evidence is not clear, values of 0.5 have been used. This ‘update’ does not alter
the prior probability at all. In that case, from Equation (2):

L(E | imp)P (imp)

P (imp | E) =
L(E | imp)P (imp) + L(E | imp)P (imp)
0.5 × P (imp) P (imp)
= =
0.5 × [P (imp) + P (imp)] 1

Assumptions
In the cases discussed in this paper, estimates of likelihood are being made for the case of
a fire in each of:
(1) the HSL fire gallery (arched shape, 2.44 m high by 2.75 m wide)
(2) the Sheffield laboratory apparatus (square shape, 0.25 m wide by 0.25 m high)
(3) the Yokohama laboratory apparatus (rectangular shape, 0.6 m wide by 0.3 m high) or
282 Fire Technology 2005

(4) the Memorial Tunnel (arched shape, 8.8 m wide by 7.9 m high),

based on assumptions made for one of the tunnels of interest. Clearly, some scaling of
dimensions, velocity and HRR is required in order to compare tests. To do this, the variables
have been scaled by assuming conservation of the Froude number [24–26]. This means that
the ventilation velocity scales as D1/2 and HRR scales as D5/2 , where D is a characteristic
dimension of the tunnel scenario; in this case assumed to be either the tunnel height, h, or
the mean hydraulic diameter of the tunnel, h̄ [25].
In the HSL tunnel, the difference between h and h̄ is small (h = 2.44 m and h̄ = 2.54 m)
so D has been taken to be 2.5 m in all calculations. In the Sheffield tunnel apparatus D =
h = h̄= 0.25 m. In the Memorial Tunnel and the Yokohama tunnel apparatus, the differences
between the values of h and h̄ are larger (h = 7.9 m & h̄ = 8.23 m for the Memorial Tunnel
and h = 0.3 m & h̄ = 0.4 m for the Yokohama apparatus) so calculations have been made
using scale factors of D = h and D = h̄. In these calculations, the experimental tests would
only be allocated ‘very high’ or ‘very low’ likelihood values at a given location if the
presence (or absence) of flames was estimated using both scale factors.
Also, estimates of the ‘likelihood’ of each experiment will be used to update the estimates
of probability for hypotheses relating to each of the different HRR ranges. Few tests are
able to be directly comparable to a fire falling into each of the nine different HRR ranges,
so some assumptions regarding flame behaviour need to be made in order to be able to
make estimates. In general, it is assumed that if the flames from a fire of HRR Q extend to
a point N m downstream of the downstream edge of the fire object then:

• The flames would extend at least this far at higher HRRs (given the same ventilation
velocity)
• The flames would not extend this far at lower HRRs (given the same ventilation velocity)

When estimating the position of the flames from temperature data it has been assumed
that temperatures of 600◦ C or more are due to flame. In the absence of experiments with
suitable video footage, this is the kind of assumption which needs to be made. (It should be
noted that in their analysis of flame behaviour, Oka and Kurioka et al. assume a temperature
rise of only 250◦ C above ambient to represent a flame [13, 14].)

Examples
The methodology used in this study is best explained by means of an example. Here, a
matrix of estimates of the probability of impingement at separation distance, s = 0, 1, 2,
. . . , 19 and 20 m and HRR, Q = 1–2, 2–4, . . . , 128–256 and 256+ MW, at a ventilation
velocity of 2 ms−1 is considered. In the ‘prior’ matrix of probabilities, all values are 0.5.
Example 1. Consider the evidence of test 6 from the HSL test series. The approximate
positions of the fire source (two square fuel trays) and the thermocouples are shown in
Figure 3, four of the thermocouple trees are indicated. As well as the fire source, there were
objects in the tunnel to represent HGVs on Channel Tunnel carrier wagons. Figure 4 shows
the evolution of HRR with time. The airflow velocity was constant at about 1.1 ms−1 for
the first twenty minutes of the test, after which it was increased to about 1.6 ms−1 .
This test is given as an example as the relatively slow growth of HRR enables good
estimates of flame behaviour to be made across a span of several of the HRR ranges. If
scaled up to the single-lane TOI dimensions, the ventilation velocity in the tunnel would be
Fire Spread Between Vehicles in Tunnels 283

Figure 4. Approximate representation of the HRR history for HSL

test 6. An additional y-axis showing the equivalent HRR in the
single lane TOI has been added for clarity. The points at which the
first flame impingement would have occurred at thermocouple trees
C and D are indicated.

about 1.8 ms−1 and the HRR of the fire should be multiplied by a factor of about 9.9. Hence
it is clear from the above that for the flames to reach the TOI equivalent of location C, the
HRR of the initial fire needs to be at least 22 MW, for an airflow velocity of approximately
2 ms−1 . Similarly, in order to reach the equivalent of location D, the HRR needs to be at
least 48 MW. When scaled up to TOI dimensions, locations C and D are approximately 7
and 20 m away from the back of the fire, respectively.
The (scaled up) peak HRR of the fire is about 50 MW, thus the ‘likelihood’ of this test,
given impingement in the single-lane TOI, for the 32–64 MW range has been taken to be
‘very high’ (i.e. 0.9) at all separation distances up to 7 m and ‘high’ (i.e. 0.7) at all distances
from 8 to 20 m. The likelihood of the evidence for all higher HRR ranges is taken to be
very high at all separation distances up to 20 m. Based on the flame behaviour in the growth
phase of the fire, the likelihood values for a 16–32 MW fire are taken to be very high at s =
0 to 6 m, high at 7 m, medium at 8 m and low at all other separation distances. At 8–16 MW,
the likelihood of the evidence is taken to be medium at l–2 m, low at 3–7 m and very low
at larger separation distances. The matrix of likelihood values, given impingement in the
TOI, estimated for HSL test 6 at 2 ms−1 , is shown in Table 2.
Bayesian calculations were carried out for each estimate of probability in the prior matrix
to give the ‘posterior’ matrix. These updated estimates of probability were then used as the
‘prior’ matrix when the likelihood of the next piece of evidence was considered, and so on.
This process was repeated for each of the eight tests carried out in the HSL fire gallery and
for each of the fire tests carried out in the laboratory apparatus at Sheffield (note that the
Yokohama laboratory data and the data from the Memorial Tunnel were used for the case
of the two-lane tunnel, not the single-lane one). Similar calculations were also carried out
for ventilation velocities of 1, 3 and 4 ms−1 .
Example 2. As another example of how the calculations work, consider the probability
of impingement on a 4 m high HGV, positioned 4 m downstream of an 8–16 MW fire in the
two-lane tunnel of interest, subject to a 1 ms−1 ventilation velocity. The prior probability
is 0.5.
 284

TABLE 2.
Estimates of Likelihood, L(E | i mp), (at 2 ms−1 ) Based on Test 6 of the HSL Test Series; S is the Separation
Distance (m). L(E | i mp) = 1 − L(E | i mp) in all Instances

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1–2MW .9 .5 .3 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1
2–4 .9 .5 .3 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1
4–8 .9 .5 .3 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1
8–16 .9 .5 .5 .3 .3 .3 .3 .3 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1 .1
16–32 .9 .9 .9 .9 .9 .9 .9 .7 .5 .3 .3 .3 .3 .3 .3 .3 .3 .3 .3 .3 .3
32–64 .9 .9 .9 .9 .9 .9 .9 .9 .7 .7 .7 .7 .7 .7 .7 .7 .7 .7 .7 .7 .7
Higher .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9 .9
Fire Technology 2005
Fire Spread Between Vehicles in Tunnels 285

Seventeen of the tests carried out by Oka and Kurioka et al. provide evidence relevant to
this case in the matrix.
The first test to be considered had a HRR equivalent to about 18 MW (when scaled
up), a ventilation equivalent to about 1.3 ms−1 and the 600◦ C zone extended almost 4 m
downstream of the fire at the height of a HGV and about 6 m downstream of the fire near
the ceiling. The test HRR is only slightly higher than the 8–16 MW range considered, so it
appears quite likely (i.e. a ‘high’ likelihood) that this experiment could have happened as
it did if we were to assume that there definitely would be flames from an 8–16 MW fire at
4 m downstream of the fire. Thus:

L(E1 | imp)P (imp)

P (imp | E1 ) =
L(E1 | imp)P (imp) + L(E1 | imp)P (imp)
0.7 × 0.5
= = 0.7
0.7 × 0.5 + 0.3 × 0.5

The next three tests to be considered had much higher HRR equivalent to 38, 56 and
56 MW, respectively (two of the tests were ostensibly identical). In these instances the
flames extended over 6 m downstream of the fire, but due to the fire size being much greater
than the 8–16 MW range considered, it is only possible to give likelihood values of 0.5 in
these instances, so the posterior estimate of probability remains unchanged.
The fifth test to be considered had a HRR equivalent to about 18 MW, a ventilation
equivalent to about 0.9 ms−1 and the 600◦ C zone extended almost 4 m downstream of
the fire at the height of a HGV. Once again, the test HRR is only slightly higher than the
8–16 MW range considered, so it appears quite likely (i.e. a ‘high’ likelihood) that this
experiment could have happened as it did if we were to assume that there definitely would
be flames from an 8–16 MW fire at 4 m downstream of the fire.
The sixth test to be considered also had a HRR equivalent to about 18 MW, a ventilation
equivalent to about 1.4 ms−1 and the 600◦ C zone extended almost 6 m downstream of the
fire at the height of a HGV. In this case it appears very likely (i.e. a ‘very high’ likelihood)
that this experiment could have happened as it did if we were to assume that there definitely
would be flames from an 8–16 MW fire at 4 m downstream of the fire. Thus the likelihood,
L(E6 | imp) = 0.9 and thus P(imp | E6 ) = 0.98, using Equation (2).
The seventh test to be considered yielded no information relevant to an 8–16 MW fire, so
the posterior probability remained unchanged. The eighth test to be considered was quite
similar to the sixth, so the estimated likelihood was L(E8 | imp) = 0.9 and thus P(imp | E8 )
= 0.9977, using Equation (2).
This process was repeated, with the next three tests giving likelihood values of 0.7, 0.5
and 0.9 respectively. Thus the posterior after eleven iterations was P(imp | E11 )= 0.9999.
However, the likelihoods of the next six tests to be considered were 0.3, 0.1, 0.5, 0.3, 0.3
and 0.1, which brought the posterior after seventeen iterations back down to P(imp | E17 )
= 0.90.
286 Fire Technology 2005

Results
The prior matrix for the single-lane tunnel was updated using each of the experimental tests
from the HSL fire gallery at Buxton and the burner tests in the University of Sheffield lab
apparatus as these two test series were carried out in ‘single-lane’ tunnels. The prior matrix
for the two-lane tunnel was updated using each of the experimental tests from the Memorial
Tunnel and the burner tests in the Yokohama National University lab apparatus as these
two test series were carried out in ‘two-lane’ tunnels. The posterior probabilities after all
iterations are given in Tables 3–6, for the single-lane tunnel at 1, 2, 3 and 4 ms−1 and in
Tables 7–10 for the two-lane tunnel at 1, 2, 3 and 4 ms−1 . It should be stressed that the
probability values given in Tables 3–10 are rounded to two decimal places. The calculated
probabilities given as ‘0.00’ are actually very small, but non-zero, numbers. Similarly,
probabilities given as ‘1.00’ are slightly less than unity. It should, however, be emphasised
that the calculated probabilities are only approximate. Although several of the calculated
posterior probabilities have values like 0.9995 or 0.00001, there is little justification in
relating this level of accuracy to the real world. It is the trends in the data that are of
importance rather than the individual numbers calculated.
Probabilities given as ‘0.5’ generally correspond to scenarios where none of the available
evidence has provided any clear estimates of likelihood. For example, it is deemed unlikely
that there is only a 0.5 probability of impingement at distances between 10 and 20 m
downstream of a 128–256 MW fire in the two-lane tunnel of interest, but none of the
evidence considered can confirm or refute this assumption, so the prior estimates remain
unchanged.
Some of the main features of the results are highlighted below.
Figure 5 shows the variation of the probability of impingement with increasing heat
release rate at 5,10, 15 and 20 m vehicle spacings, in both tunnels of interest, at 2 and
4 ms−1 . It is clear that all fires larger than 128 MW would be expected to produce flames
that would impinge on an adjacent HGV up to 20 m away, irrespective of tunnel size or
ventilation velocity. A similar observation can be made for fires in the 64 to 128 MW range,
although in this case there is a small reduction in probability in the two-lane tunnel, at
spacings larger than 10 m. As many typical HGV fires exhibit heat release rates of 64 MW
or above, it is reasonable to assume from these results that flames from a typical HGV in a
tunnel will almost certainly impinge on an adjacent HGV, up to 20 m away, irrespective of
ventilation velocity or tunnel size.
For fires in the 32–64 MW range (which may be taken to be a small HGV fire or a
large van fire) there is considerable difference between the two tunnels of interest. In the
two-lane tunnel, impingement is not likely if the separation is 10 m or greater, whereas
in the single-lane tunnel the likelihood of impingement is fairly high, especially at higher
ventilation velocities.
For fires in the 16 to 32 MW range (which may be taken to be a van fire) there is a very
low likelihood of impingement except with small vehicle spacings or, in the case of the
single-lane tunnel, at high ventilation velocities. Finally, for fires in the 8 to 16 MW range
(which may be a very large car fire or a small van fire), there is only a significant likelihood
of spread at small separation distances and high ventilation velocities.
Figure 6 shows the variation of the probability of impingement with increasing separation
distance, in both tunnels of interest, at 2 and 4 ms−1 . These graphs clearly show the effect
TABLE 3
Posterior Probabilities of Impingement for a Single-lane Tunnel at 1 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Fire Spread Between Vehicles in Tunnels

1–2MW 1.00 0.99 0.30 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 0.97 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4–8 1.00 1.00 0.97 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 0.94 0.16 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 0.90 0.16 0.03 0.03 0.03 0.03 0.03 0.03 0.01 0.01 0.00 0.00 0.00 0.00 0.00
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.93 0.84 0.70 0.70 0.70 0.70 0.70 0.70 0.30 0.30 0.30 0.30 0.30
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.99 0.99 0.99 0.99
128–256 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
287
 288

TABLE 4
Posterior Probabilities of Impingement for a Single-lane Tunnel at 2 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1–2MW 1.00 0.99 0.62 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 1.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4–8 1.00 1.00 1.00 0.30 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 1.00 0.79 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.50 0.07 0.03 0.03 0.03 0.03 0.03 0.01 0.01 0.00 0.00 0.00 0.00 0.00
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.97 0.93 0.84 0.84 0.84 0.84 0.84 0.84 0.50 0.50 0.50 0.50 0.50
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.99 0.99 0.99 0.99
128–256 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Fire Technology 2005
TABLE 5
Posterior Probabilities of Impingement for a Single-lane Tunnel at 3 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Fire Spread Between Vehicles in Tunnels

1–2MW 1.00 0.99 0.62 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 1.00 0.10 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4–8 1.00 1.00 1.00 0.79 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 1.00 0.99 0.30 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.50 0.16 0.03 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.93 0.84 0.84 0.84 0.84 0.84 0.84 0.50 0.50 0.50 0.50 0.50
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.99 0.99 0.99 0.99
128–256 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
289
 290

TABLE 6
Posterior Probabilities of Impingement for a Single-lane Tunnel at 4 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1–2MW 1.00 1.00 0.79 0.16 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 1.00 1.00 0.38 0.03 0.01 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4–8 1.00 1.00 1.00 1.00 0.90 0.70 0.10 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.90 0.10 0.10 0.10 0.02 0.02 0.02 0.02 0.02 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.90 0.79 0.50 0.50 0.50 0.50 0.50 0.16 0.16 0.05 0.05 0.05
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.97 0.97 0.97 0.97 0.93 0.93 0.93 0.84 0.84 0.84 0.84 0.84
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.99 0.99 0.99 0.99 0.99 0.99
128–256 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Fire Technology 2005
TABLE 7
Posterior probabilities of impingement for a two-lane tunnel at 1 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1–2MW 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Fire Spread Between Vehicles in Tunnels

4–8 1.00 1.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 1.00 0.41 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 0.95 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.70 0.30 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.50 0.30 0.16 0.16 0.16 0.16 0.16 0.16 0.16 0.16 0.16
128–256 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50

Note that probabilities of 0.5 generally correspond to combinations of spacing, ventilation and HRR for which there are no relevant evidence and these figures should not therefore be
interpreted to mean that, under these conditions, flames will not impinge in 50% of cases
291
 292

TABLE 8
Posterior Probabilities of Impingement for a Two-lane Tunnel at 2 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1–2MW 1.00 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4–8 1.00 1.00 1.00 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 1.00 1.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.96 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.50 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.98 0.90 0.90 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79
128–256 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99
Fire Technology 2005
TABLE 9
Posterior Probabilities of Impingement for a Two-lane Tunnel at 3 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Fire Spread Between Vehicles in Tunnels

1–2MW 1.00 0.50 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 0.50 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4–8 1.00 1.00 1.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 1.00 1.00 0.91 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.54 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.70 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.98 0.95 0.95 0.95 0.95 0.95 0.95 0.95 0.95 0.95
128–256 1.00 .1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
293
 294

TABLE 10
Posterior Probabilities of Impingement for a Two-lane Tunnel at 4 ms−1

S= 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1–2MW 1.00 0.50 0.05 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2–4 1.00 1.00 0.50 0.10 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4–8 1.00 1.00 1.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
8–16 1.00 1.00 1.00 1.00 1.00 0.73 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
16–32 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.54 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
32–64 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.84 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
64–128 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.99 0.98 0.95 0.95 0.95 0.95 0.95 0.95 0.95 0.95
128–256 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
256+ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Fire Technology 2005
 Fire Spread Between Vehicles in Tunnels

Figure 5. Variation of the probability of flame impingement with heat release rate: (a) in the
single-lane TOI with a 2 ms−1 longitudinal airflow, (b) in the single-lane TOI with a 4 ms−1 longitudinal
airflow, (c) in the two-lane TOI with a 2 ms−1 longitudinal airflow and (d) in the two lane TOI with a
4 ms−1 longitudinal airflow.
295
296 Fire Technology 2005

of tunnel size on the likelihood of impingement. For a 32 to 64 MW fire the likelihood

of impingement in the single-lane tunnel is greater than 50% at all separation distances,
and is significantly higher at higher ventilation velocities. Whereas in the two-lane tunnel
there is a very low likelihood of impingement at separation distances of 12 m or greater,
irrespective of ventilation velocity.
Figure 7 shows the influence of ventilation velocity on probability of impingement, for
both tunnels of interest, at a fixed separation distance of 10 m. This clearly shows that
higher ventilation rates will generally lead to higher likelihoods of impingement, especially
in the two-lane tunnel.

Discussion
As emphasised at the beginning of this article, there are several different mechanisms of
fire spread. While this study has focussed on fire spread by flame impingement, it should
not be forgotten that these other mechanisms may also be active in any given tunnel fire
scenario even if the probability of flame impingement is low.
It has previously been shown how longitudinal ventilation and tunnel geometry can
influence the heat release rate of car and HGV fires in tunnels [17]. From that study it
was clear that car fires behave very differently from HGV fires and it has also been shown
that different sizes of pool fire behave in different ways to vehicle fires [18]. However, the
influence of these factors on flame behaviour was not considered at the time.
In order that the results of the current study be easily combined with the results of the
earlier studies, the results will be discussed below in terms of each of the different fire
scenarios considered in the first study; i.e. car fires, small pool fires, medium pool fires,
large pool fires and HGV fires.

Car and Small Pool Fires

From the experimental data considered in the earliest study [17] it would appear that the
HRR of a car fire in a tunnel may range from about 2 MW to about 8 MW, thus the 2–4 MW
and 4–8 MW ranges considered in the current study may be taken to be approximately
equivalent to small and large car fires. Small pool fires are also likely to exhibit heat release
ranges of this magnitude.
From the results presented above it is clear that the flames from a car or small pool fire in
a tunnel would not be expected to extend as much as 5 m downstream of the fire, in either
tunnel of interest, for any of the considered ventilation velocities. Indeed, it is only in the
single-lane tunnel and at a ventilation velocity of 4 ms−1 that the flames appear likely to
extend more than 3 m downstream of the fire.
Based on these results it is recommended that at least a 5 m separation distance be
maintained downstream of a car fire in a tunnel to avoid flame impingement. However,
given the approximate nature of the results from this study, erring on the more cautious
side might imply a 10 m minimum separation distance be maintained to avoid flame
impingement. To avoid fire spread by mechanisms other than flame impingement, e.g.
spontaneous ignition, a greater separation may be required.
 Fire Spread Between Vehicles in Tunnels

Figure 6. Variation of the probability of flame impingement with separation distance: (a) in the
single-lane TOI with a 2 ms−1 longitudinal airflow, (b) in the single-lane TOI with a 4 ms−1 longitudinal
airflow, (c) in the two-lane TOI with a 2 ms−1 longitudinal airflow and (d) in the two lane TOI with a
4 ms−1 longitudinal airflow.
297
 298

Figure 7. Variation of the probability of flame impingement with ventilation velocity for initial fires
with HRRs in the 8–16, 16–32, 32–64 and 64–128 MW ranges and a vehicle separation distance of
10 m: (a) in the single-lane TOI and (b) in the two-lane TOI.
Fire Technology 2005
Fire Spread Between Vehicles in Tunnels 299

Van and Medium Pool Fires

Van fires were not considered in the original study, but it is assumed that the 8–16 MW and
16–32 MW ranges approximately correspond to small and large van fires. Medium pool
fires are also likely to be in the 16–32 MW range.
From the results presented above it is clear that the outcome of fires of these magnitudes
can be influenced greatly by variations in ventilation and vehicle spacing. For example,
Figure 5 clearly demonstrates that the flames are unlikely to impinge on a HGV 5 m
downstream from an 8–16 MW fire if the ventilation velocity is 2 ms−1 (in either tunnel
of interest). However, if the ventilation velocity is increased to 4 ms−1 , the likelihood of
impingement is very high (almost 100% in a single lane tunnel and over 70% in a two-
lane tunnel). Figure 6 shows that in a single-lane tunnel the probability of impingement at
separation distances more than 8 m is very low at 2 ms−1 , but if the ventilation is increased
to 4 ms−1 the flames are reasonably likely to extend as much as 14 m. This effect is not
apparent in the larger tunnel.
There appears to be a large ‘jump’ in the probability of impingement at 10 m downstream
of a 16–32 MW fire in a single lane tunnel between ventilation velocities of 3 and 4 ms−1 .
This apparent jump may simply be due to the nature of the data and the location of the
thermocouples. However, it may be that only ventilation velocities of the order of 4 ms−1
are able to force the flames from a fire of this magnitude to impinge on a vehicle 10 m
downstream.
Based on these results, it is recommended that a separation distance of at least 10 m be
maintained downstream of a van fire in a two-lane tunnel and at least 15 m in a single-lane
tunnel. Also, for van fires in single-lane tunnels, ventilation rates of the order of 4 ms−1
should not be used if the smoke can be controlled using smaller ventilation velocities. As
with car fires, erring on the side of caution might imply that, in practice, minimum distances
of, say, 20 m in the two-lane tunnel and 25 m in the single-lane tunnel should be maintained
to avoid flame impingement.
HGV and Large Pool Fires. While some HGV fires might produce heat release rates in
the 16–32 MW, it is much more likely that HGV fires will fall into one of the 32–64 MW,
64–128 MW or 128–256 MW ranges. According to the model CERBERUS A 1.2 (which
predicts the probability of HRR for HGV fires in longitudinally ventilated tunnels) [28, 29],
there is only a 5% chance that a HGV with a cargo of furniture, subject to a 2 ms−1 airflow
velocity, in the single-lane TOI, will have a peak HRR of less than 32 MW. Similarly there
is almost a 50% chance that such a HGV fire will have a peak HRR greater than 64 MW. For
a 4 ms−1 airflow, CERBERUS predicts that the chance of such a fire being below 32 MW is
only 2% while the chance of it being above 64 MW is 66%. For this reason we consider that
the 32–64 MW range represents a small HGV fire, while the 64–128 MW range represents
a large HGV fire. Large pool fires will also have HRRs in these ranges.
From Figures 5(a), (b) and 6(a), (b) it can be seen that for small HGV fires (i.e. 32–
64 MW) there is a high (i.e. greater than 0.7) probability of impingement at almost all
separation distances up to 20 m downstream of the fire in the single-lane tunnel of interest.
(Flame impingement may occur at greater distances, but the upper limit considered in this
study was 20 m). However, in the two-lane TOI, the probability of impingement is only
high at distances up to 10 m downstream. In both cases, the probability of impingement is
slightly increased by increasing the longitudinal ventilation velocity.
300 Fire Technology 2005

For large HGV fires (i.e. over 64 MW) there is a very high likelihood of impingement
at all considered downstream positions, in both tunnels of interest and at all considered
ventilation velocities. While the study did not consider separation distances greater than
20 m, it appears likely from the evidence considered that the flames from a large HGV
fire would extend significantly more than 20 m downstream of the fire. Based on this
observation, it is recommended that as large a separation distance as possible be maintained
downstream of a HGV fire in a tunnel.

Tanker Fires and Very Large Pool Fires. Tanker fires and very large pool fires (as well as
large HGV fires) may be taken to fall into the 128–256 MW range and may be even greater.
From the results presented above it is certain that flames from such a fire would impinge on
a HGV positioned up to 20 m downstream, in either tunnel of interest and at all ventilation
velocities considered. It is very likely that the flames from such fires would extend much
further than the 20 m distance considered in this study.

Trends in the Data

Given the nature of the study, it is stressed that the trends observed in the data are of
greater importance than individual data points. For example, Table 4 shows that there is a
0.5 probability of flame impingement on a HGV 7 m downstream of a 16–32 MW fire in a
single lane tunnel, whereas Table 8 shows a probability of nearly 1 for impingement at the
same location in a two-lane tunnel. This observation is contrary to the generally observed
trend that impingement is less likely at a given distance downstream of a fire in a two-lane
tunnel than in a single-lane tunnel. The discrepancy in this instance is possibly due to the
location of the thermocouples in the experiments that the results are based on, rather than
as a result of a real phenomenon.
In a similar way, the probability of impingement at a location 4 m downstream of an 8–
16 MW fire, in a tunnel with 2 ms−1 longitudinal airflow appears to be less in a single-lane
tunnel than in a two-lane tunnel (see Figure 6(a) and (c)). Again, this is assumed to be due
to the nature of the data considered and not a real phenomenon.
While this study has concentrated on trends in the data, by assuming similar behaviour of
all fires within a given range of HRR values, it should be stressed that in reality there will be
significant differences of behaviour within some of the ranges considered, especially those
at higher HRRs. It has been assumed here that broad trends in behaviour will be shown by
considering the ranges defined.

Unresolved Questions
There are a number of issues that this study has not resolved. For example, it is inherently
assumed that all fires of a given heat release rate will behave in approximately the same
manner, but this may not be the case; the plume of a 10 MW fire produced by a 1 m2
fire source has a much greater upward thrust (and, hence, is less likely to impinge on
downstream objects) than the plume of a 10 MW fire produced by a 10 m2 fire source—the
HRR, Q, may be the same in both instances, but the non-dimensional heat release rate, Q∗
(a quantity used to classify fire type, see [23] p. l10), is very different. Q∗ is defined as:

Q
Q∗ = (3)
ρo To Cp g 1/2 h5/2
Fire Spread Between Vehicles in Tunnels 301

where ρ 0 is the gas density, T0 the ambient temperature, Cp the specific heat at constant
pressure, g the acceleration due to gravity and h the tunnel height (or other scale factor). At
present, it has been assumed that variations amongst the Q∗ values of the considered tunnel
fire experiments are insignificant; especially as this study considers fairly broad ranges of
HRR. Variations in Q∗ may be of significance when considering the results from burner
fire experiments, so these have been used with caution.
Another factor which has not been considered in detail is the influence of the height of
the fire source above the roadway. If the flames from a HGV fire emerged from the top of
the trailer, 4 m above the roadway, the likelihood of impingement on a downstream vehicle
would be expected to be less than if the flames from the fire emerged from the bottom of the
trailer, perhaps only 1 m above the roadway. In this instance it has been assumed that the
flames from the initial fire emerge near the roadway as this has been assumed to represent
the greater risk of spread; in other words we are erring on the side of caution. Also, in each
of the experiments considered in this study so far, the fire source is either on the roadway
or in a location similar to the base of a HGV trailer. As far as the authors are aware, no
experiments have yet been carried out where the fire source was located in the upper half
of a tunnel.

Sensitivity Issues
This study is based on a number of assumptions and estimates. If the results are to have any
value it is important to consider how much the final posterior probabilities are influenced
by variations in these assumptions and estimates.
One of the primary assumptions in this study is that a recorded temperature of 600◦ C
or greater in a tunnel fire experiment is equivalent to a flame. Several of the calculations
have been repeated assuming flame temperatures of 500 and 700◦ C instead of 600◦ C.
As would be expected, the probability of impingement at a given location was generally
slightly higher assuming a flame temperature of 500◦ C and slightly lower assuming a flame
temperature of 700◦ C, but the conclusions drawn from the posterior results were not altered
in any way by these variations.
Another of the primary assumptions is that the top of the target HGV is 4 m above the
roadway, yet ‘real’ HGVs may be shorter or taller than this. Some of the calculations were
redone assuming that the target HGV was 4.5 m high. The results for the two-lane tunnel
were not significantly altered by this change, but the likelihood of impingement values
for the single-lane tunnel were significantly increased, particularly at lower ventilation
velocities. The reason for this is that at higher ventilation velocities the flames from the
fire tend to extend reasonably horizontally from the fire location (and hence impinge on the
lower parts of the target HGV), whereas the flames from fires subject to lower ventilation
rates are more likely to rise up to the tunnel ceiling and be deflected by it.
It is also of interest to consider the sensitivity to the likelihood values chosen, i.e. 0.9
for ‘very high’ likelihood, 0.7 for ‘high’, 0.5 for ‘medium’, 0.3 for ‘low’ and 0.1 for ‘very
low’. The calculations for the case of a fire in a two-lane tunnel at 2 ms−1 were redone
using values of 0.7, 0.6, 0.5, 0.4 and 0.3 for ‘very high’ to ‘very low’ respectively. All of
the 20 experiments from the Japanese test series which are applicable to this scenario were
considered as evidence. With the exception of the posterior probability of impingement
from a 8–16 MW fire at 5 m spacing (which was 0.1 with the original likelihood values and
0.3 with the alternative values), none of the posterior values was altered by more than 0.1,
302 Fire Technology 2005

indeed the majority of values were not changed by as much as 0.001 by the change in the
likelihood values. Thus it is concluded that the posterior results are not greatly influenced
by the choice of likelihood values, given a sufficient number of pieces of evidence.

Conclusions
Given the assumptions stated above, some conclusions from this project are:
• The probability of impingement tends to increase with increasing ventilation velocity.
• Small vehicle fires (e.g. car fires, which are highly unlikely to exceed 8 MW) are not
likely to produce flames which will impinge on an object more than five metres away.
• HGV fires larger than 64 MW will, almost certainly, produce flames which will impinge
on another HGV up to 20 m away, at all ventilation velocities considered here (except at
1 ms−1 in the two lane-tunnel).
• In a single lane tunnel, fires between 32 and 64 MW will, almost certainly, produce
flames which will impinge on another HGV up to 10 m away and there is generally a
greater than 50% chance that flames from such a HGV will impinge on a HGV between
10 and 20 m away; for all ventilation velocities from 1–4 ms−1 . In a two-lane tunnel the
flames from such a fire will only impinge on vehicles less than about 8 m away.
• For fires less than 32 MW (i.e. van fires), subject to airflow velocities of 3 ms−1 or less,
a vehicle spacing of at least 10 m should be maintained to avoid flame impingement.
It should, once again, be stressed that the results of this study are approximate and that
the trends in the data are of greater importance than the numerical values presented. More
experimental data are required before the behaviour of flames from fires in tunnels can be
more accurately understood. In particular, those involved in experimental testing of fires
in tunnels are urged to position video cameras in their experiments at locations where the
behaviour of the flames can be accurately observed.

Future work
It is intended that the results from this study will be combined with the CERBERUS model
[29] to enable it to make more accurate predictions of fire spread in tunnels.

Acknowledgments
The authors would like to thank Dr. Richard Bettis, HSL, UK, Dr. Yajue Wu, University of
Sheffield, UK, Dr. Yasushi Oka, Yokohama National University, Japan and the KAJIMA
Corporation, Japan, for providing the data used in this study. This project was funded by the
Engineering and Physical Sciences Research Council (EPSRC) under grant GR/S02143/01.

References
[1] D. Lacroix, “The Mont Blanc Tunnel Fire: What Happened and What has been Learned,” in
Proc. 4th Int. Conf. on Safety in Road and Rail Tunnels, Madrid, Spain, 2–6 April, 2001, pp.
3–16.
Fire Spread Between Vehicles in Tunnels 303

[2] J.-C. Martin, O. Delémont, and C. Calisti, “Tunnel Fire Investigation II: The St. Gotthard Tunnel
fire, 24th October 2001,” in The Handbook of Tunnel Fire Safety, A. Beard and R. Carvel (eds.,)
Thomas Telford, London, pp. 53–76, 2005. ISBN 0727731688.
[3] Private communication with H. Ingason, November 2004.
[4] S.K. Liew, D.M. Deaves, and A.G. Blyth, “Eurotunnel HGV Fire on 18th November 1996—Fire
Development and Effects,” in Proc. 3rd Int. Conf. on Safety in Road and Rail Tunnels, Nice,
France, March 9–11, 1998, pp. 29–40.
[5] C. Rew and D. Deaves, “Fire Spread and Flame Length in Ventilated Tunnels, a Model Used
in Channel Tunnel Assessments,” in 1st Int. Conf. on Tunnel Fires and Escape From Tunnels,
Lyon, France, May 5–7, 1999, pp. 397–406.
[6] A.N. Beard, D.D. Drysdale, and S.R. Bishop, “A non-linear model of major fire spread in a
tunnel,” Fire Safety Journal, vol. 24, 1995, pp. 333–357.
[7] A.N. Beard, “A Model for Predicting Fire Spread in Tunnels,” Journal of Fire Sciences, vol.
15, 277–307, pp. 1997.
[8] A.N. Beard, “Major Fire Spread in a Tunnel: A Non-Linear Model,” in Proc. 4th Int.
Conf. on Safety in Road and Rail Tunnels, Madrid, Spain, 2–6 April, 2001, pp. 467–
476.
[9] A.N. Beard, “Major Fire Spread in a Tunnel: A Non-linear Model with Flame Impingement,”
in Proc. 5th Int. Conf. on Safety in Road and Rail Tunnels, Marseille, France, 6–10 October
2003, pp. 511–521.
[10] P.H. Thomas, “The Size of Flames from Natural Fires,” in Proc. 9th Int. Symp. on Combustion,
The Combustion Institute, Pittsburgh, USA, pp. 844–859.
[11] “Fires in Transport Tunnels: Report on Full-scale Tests” EUREKA-Project EU499:FIRETUN
Studiengesellschaft Stahlanwendung elV. D-40213 Dusseldorf. 1995.
[12] “Memorial Tunnel Fire Ventilation Test Programme,” Interactive CD-ROM & Comprehensive
Test Report. Parsons Brinckerhoff 4D Imaging, 1996
[13] Y. Oka, T. Yamada, H. Kurioka, H. Sato, H. Kuwana, and O. Sugawa, “Flame Behaviour in a
Tunnel,” in 3rd Int. Conf. on Safety in Road and Rail Tunnels, Nice, France, March 9–11, 1998,
pp. 159–168.
[14] H. Kurioka, Y. Oka, H. Satoh, and O. Sugawa, “Fire Properties in Near Field of Square Fire
Source with longitudinal Ventilation in Tunnels,” Fire Safety Journal, vol. 38, 2003, pp. 319–
340.
[15] V.B. Apte, A.R. Green, and J.H. Kent, “Pool Fire Plume Flow in a Large-scale Wind Tunnel,”
Fire Safety Science—Proc. 3rd Int. Symposium. University of Edinburgh, Scotland, 8–12 July,
1991, pp. 425–434.
[16] V.B. Apte, A.R. Green, and J.H. Kent, “Fire Plume Flow in Mines,” End of grant report.
NERD&D project no.1078. Londonderry Occupational Safety Centre (NSW, Australia), June
1991.
[17] R.O. Carvel, A.N. Beard, P.W. Jowitt, and D.D. Drysdale, “Variation of Heat Release Rate with
Forced Longitudinal Ventilation for Vehicle Fires in Tunnels,” Fire Safety Journal, vol. 36,
2001, pp. 569–596.
[18] R.O. Carvel, A.N. Beard, and P.W. Jowitt, “The Influence of Longitudinal Ventilation on Fire
Size in Tunnels: Update,” in Proc. 5th Int. Conf. on Safety in Road and Rail Tunnels, Marseille,
France, 6–10, October 2003, pp. 431–441.
[19] H. Ingason and A. Lönnermark, “Large-Scale Fire Tests in the Runehamar Tunnel—Heat
Release Rate (HRR),” in Proc. Int. Symp. on Catastrophic Tunnel Fires, Boras, Sweden, 20–
21st November 2003. Published by SP Fire Technology, Sweden. Report 2004:05, pp. 81–92,
ISBN 91-7848-978-4.
[20] A. Lönnermark and H. Ingason, “Large-Scale Fire Tests in the Runehamar Tunnel—Gas Tem-
perature and Radiation,” in Proc. Int. Symp. on Catastrophic Tunnel Fires, Borås, Sweden,
304 Fire Technology 2005

20–21st November 2003. Published by SP Fire Technology, Sweden. Report 2004:05, pp.
93–104, ISBN 91-7848-978-4.
[21] Private communication with E. Rikkhard, May 2004.
[22] Private communication with H. Ingason, May 2004.
[23] D.D. Drysdale, Introduction to Fire Dynamics (1st edition, 1985, 2nd edition 1999), John Wiley
and Sons.
[24] R.J. Bettis, S.F. Jagger, and Y. Wu, “Interim Validation of Tunnel Fire Consequence Models:
Summary of Phase 2 Tests,” Section paper IR/L/93/11. Health & Safety Executive, Buxton,
1993.
[25] M.Z.A. Bakar, “A Study of the Effect of Tunnel Aspect Ratio on Control of Smoke Flow in
Tunnel Fires,” PhD Thesis, University of Sheffield, November 1999.
[26] Y. Wu and M.Z.A. Bakar, “Control of Smoke Flow in Tunnel Fires Using Longitudinal Ven-
tilation Systems—A Study of the Critical Velocity,” Fire Safety Journal, vol. 35, 2000, pp.
363–390.
[27] Y. Wu, M.Z.A. Bakar, G.T. Atkinson, and S. Jagger, “A Study of the Effect of Tunnel Aspect
Ratio on Control of Smoke Flow in Tunnel Fires,” in 9th Int. Conf. on Aerodynamics and
Ventilation of Vehicle Tunnels, Aosta Valley, Italy, October 6–8 1997, pp. 573–587.
[28] R.O. Carvel, A.N. Beard, and P.W. Jowitt, “How Longitudinal Ventilation Influences Fire Size
in Tunnels,” Tunnel Management International, vol. 7, no. 1, 2004, pp. 46–54.
[29] R.O. Carvel, A.N. Beard, and P.W. Jowitt, “CERBERUS: A New Model to Estimate Size and
Spread for Fires in Tunnels with Longitudinal Ventilation,” in Proc. Int. Conf. on Tunnel Safety
and Ventilation—New Developments in Tunnel Safety, Graz, Austria, 8–10th April, 2002, pp.
69–76.