Content uploaded by Peter Lawrence
Author content
All content in this area was uploaded by Peter Lawrence on Dec 20, 2013
Content may be subject to copyright.
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
A PRELIMINARY INVESTIGATION OF THE
EVACUATION OF THE WTC NORTH TOWER USING
COMPUTER SIMULATION
Galea, E.R., Lawrence P.J., Blake, S.J., Gwynne S. and Westeng H.
Fire Safety Engineering Group
The University of Greenwich
London SE10 9LS, UK
http://fseg.gre.ac.uk/
ABSTRACT
This paper concerns a preliminary numerical simulation study of the evacuation of the World Trade
Centre North Tower on 11 September 2001 using the buildingEXODUS evacuation simulation software. The
analysis makes use of response time data derived from a study of survivor accounts appearing in the public
domain. While exact geometric details of the building were not available for this study, the building geometry
was approximated from descriptions available in the public domain. The study attempts to reproduce the
events of 11 September 2001 and pursue several ‘what if’ questions concerning the evacuation. In particular,
the study explores the likely outcome had a single staircase survived in tact from top to bottom.
INTRODUCTION
The evacuation of the World Trade Centre (WTC) towers is of fundamental importance to the future
design of high-rise buildings. The attack on the WTC towers brought home to the world the importance of
providing adequate and robust means of evacuation in high-rise buildings. For many – fire safety professionals
and lay people alike - the evacuation performance of the WTC towers on 11 September was considered a
success as it has been estimated that somewhere between 10,000-14,000 people were within the towers at the
time of the attack [1] and ‘only’ an estimated 2,033 building occupants failed to escape (15% - 20% of the
estimated occupants), of which 531 (4% - 5%) perished on the impact floors [1]. However, for the estimated
1,500 (11% - 15%) people who survived the impact trauma, but were unable to evacuate, the evacuation can
hardly be claimed a success. And how would the evacuation have progressed if the normal working
occupancy of 25,000 people occupied each building? Furthermore, regardless of whether the evacuation was
considered a success or not, could the building design or evacuation procedures be improved so as to increase
evacuation efficiency?
These important questions are beginning to be addressed using evacuation simulation software. Five
years ago, it would have been considered a challenge to perform an evacuation design analysis for a 110 storey
building with 25,000 people. With today’s sophisticated modelling tools and high-end personal computers this
is now possible. However, in attempting to utilise computer simulation to forensically reproduce and explore
an event as complex as the WTC evacuation requires a higher level of sophistication. Design applications,
while still demanding, rely on generally agreed engineering parameters (e.g. response time distributions) and are
usually comparative in nature, contrasting one design with an alternative. To perform a forensic analysis it is
essential to have reliable data pertaining to the actual incident. Furthermore, the evacuation simulation tool
used in the forensic analysis must be sufficiently sophisticated to represent the relevant behaviours exhibited in
the particular incident. In this paper we explore the use of the buildingEXODUS evacuation modelling tool
[2,3] in examining the evacuation of the North Tower of the WTC (WTC1). The results of this study are
considered preliminary, as much detailed information necessary for a robust forensic investigation, while
available, has not been provided to the investigators.
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
THE EVENT
While the events of 11 September 2001 are well known, it is worth recounting the main facts. WTC1
was hit by American Airline Flight 11 at 08:46 a.m. The impact was nearly centred on the north face of the
building which was hit between the 94th and 98th floors. The South Tower (WTC2) was hit by United
Airlines Flight 175 at 09:03 a.m. The impact was at a skewed angle toward the southeast corner of the south
face of the building which was hit between the 78th and 84th floors. WTC2 collapsed at 09:59, 56 minutes 10
seconds after being hit and WTC1 collapsed at 10:37, 1 hour 42 minutes 5 seconds after being hit. There are
various estimates for the number of people in the building and the number of fatalities. Denis Couchon of US
newspaper USA Today estimates that there were between 5,000 and 7,000 people in the buildings at the time
of the impact and estimates that 2,784 people perished [4]. He estimates that 1,432 building occupants
perished in WTC1 and 599 in WTC2 [1]. From a study that the authors have made of the occupant response
times [5,6] it is clear that many of the occupants in WTC2 started their evacuation when WTC1 was hit. If
we assume that the buildings contained 7,000 people each, then WTC1 achieved an average exit flow rate of
54 survivors/minute and WTC2 achieved an average flow rate of 88 survivors/minute.
THE EVACUATION SIMULATION SOFTWARE
The modelling of the evacuation of WTC1 was performed using the buildingEXODUS simulation
software, developed by the Fire Safety Engineering Group at the University of Greenwich. The basis of the
model has frequently been described in other publications and so will only be briefly described here [2,3].
EXODUS is a suite of software tools designed to simulate circulation and evacuation of large numbers of
people from complex enclosures. The EXODUS software takes into consideration people-people, people-
structure and people-fire interactions. The model tracks the trajectory of each individual as they move around
the geometry. In evacuation applications involving fire, the model can also predict when occupants will be
affected by fire hazards such as heat, smoke and toxic gases. The software has been written in C++ using
Object Orientated techniques utilising rule base technology to control the simulation. Thus, the behaviour and
movement of each individual is determined by a set of heuristics or rules. For additional flexibility these rules
have been categorised into five interacting submodels, the OCCUPANT, MOVEMENT, BEHAVIOUR,
TOXICITY and HAZARD submodels. These submodels operate on a region of space defined by the
GEOMETRY of the enclosure. The version of EXODUS used to simulate evacuation from the built
environment is known as buildingEXODUS.
Within EXODUS, the building layout can be specified using a DXF file produced by a CAD package.
The Occupant submodel allows the nature of the occupant population to be specified. The population can
consist of a range of people with different movement abilities, reflecting age, gender and physical disabilities as
well as different levels of knowledge of the enclosure’s layout, response times etc. On the basis of an
individual's personal attributes, the Behaviour Submodel determines the occupant's response to the current
situation, and passes its decision on to the Movement Submodel. The behaviour model considers such
behaviours as; determining the occupants initial response, conflict resolution, overtaking, etc. In addition a
number of localised decision-making processes are available to each individual according to the conditions in
which they find themselves and the information available to them. This includes the ability to customise their
travel path according to the levels of congestion around them, the environmental conditions, the social
relationships within the population and interaction with signage. As certain behaviour rules, such as conflict
resolution, are probabilistic in nature, the model will not produce identical results if a simulation is repeated.
Individuals can also be tasked with a range of different itineraries enabling them to undertake specific functions
or visit specific sites within the geometry. The model produces interactive two-dimensional graphics allowing
the user to observe and interrogate every aspect of the simulation as it takes place. To aid in the interpretation
of results, a post-processor virtual-reality graphics environment known as vrEXODUS has been developed,
providing an animated three-dimensional representation of the simulation.
THE MODEL
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
Within buildingEXODUS it is necessary to specify the geometry of the building, the number, nature
and distribution of the population and finally the nature of the evacuation scenario that will be investigated.
Here we define each of these items in turn.
The Geometry
In attempting to simulate the events of 11 September 2001, the geometry of WTC1 was approximated
within the buildingEXODUS software. The model assumes that there is no significant damage to the building
below the impact floors and that the elevators are not available to assist in the evacuation. The geometry is an
approximation to the actual building as many details concerning its layout, while available were not provided to
the authors despite a number of requests for such information. What information was gathered came from
several different public sources including FEMA [7] and web sites for the New York Times [8] and USA
Today [9]. The broad structure of the building geometry represented within the software included the number
and width of staircases, number of floors, number of unoccupied floors, layout of staircase geometry, widths
of main doors, etc. However, none of these sources provided an exact representation of the layout of each of
the floors and so the model was kept as simple as possible.
WTC1 consisted of 110 floors above ground with a roof height of 417 m and a ceiling height of 3.65
m for each occupied floor [10] and approximately 4.27 m for the machine floors located on the 7-8, 41-42,
75-76 and 108-110 floors [11]. The building had a square base measuring 63 m x 63 m and a rectangular
service core, with dimensions of approximately 26.5 m by 41.75 m (see figure 1a). The core area
accommodated 3 exit stairs, 99 elevators and 16 escalators [7]. Tenant improvements over the years also
resulted in removal of portions of floors and placement of new private stairs between floors, in a somewhat
random manner.
(a) Generalised floor plan of WTC1 [7] (b) representation within
buildingEXODUS
Figure 1: The WTC1 floor plan layout
There were three independent emergency fire exit stairs located in the core of WTC1 (see figure 2a).
The stairs, as well as the rest of the “shaft-wall” interior cores, were all made of gypsum-based wallboard.
Two of the stairs, Stair A and Stair C, were both 1.1 m (44 inches) wide. They both had exits out onto the
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
Mezzanine Level and ran all the way up to the 110th floor. The third stair, Stair B, had a width of 1.4 m (56
inches). This went all the way from the Lobby and up to the 108th floor. The stairs did not run in continuous
vertical shafts from the top to the bottom of the structure. Instead the plan location of the stairs shifted at
some levels. This meant that people traversing the stairs were required to move from one vertical shaft to
another through a transfer corridor (see figure 2b). Both Stair A and Stair C had transfers on the 42nd, 48th,
76th and 82nd floor. Stair C also had an additional transfer on the 26th floor. Stair B had only one transfer and
this was located on the 76th floor. As a result the layout in the core area varied throughout the building with
elevators, staircases and free space regions interchanging.
The layout of the non-core space on each floor varied from floor to floor and was dependent on
tenant preference. Some floors had large open plan areas while others where more subdivided. Details of
these layouts were not available for this study. As details of these open plan areas were not available and in
order to reduce model memory requirements and execution time, most of the free space area outside the core
area was not included within the model. Instead a 4 m wide area around the perimeter was modelled to
represent the available space on each floor (see figure 1b). This space provided sufficient to distribute the
entire building population around the core in such a manner that allowed them to more freely without artificially
incurred congestion related delays. Limited sensitivity analysis was conducted on the idealised and complete
model representations. This suggested that this approach would not have a significant impact on model
predictions as the majority of the delays experience by occupants occurred as they attempted to enter the
staircases. The time crossing the open floors was insignificant.
(a) Stair positioning on the 73rd floor [9] (b) Transfer corridor on the 76th floor[8]
Figure 2: Stair layout in WTC1
Within the model the available space within the core area was represented by cross shaped corridors
running from east to west and north to south (see figure 1b). These corridors lead people from the floor to the
three staircases in the core area. The entries to the staircases were located along these corridors.
Compartments within the core area i.e. bathrooms, changing rooms, copy rooms, small offices etc, were not
represented as space that could be occupied by individuals.
The width of each stair riser was 1.1m for Stair A and C and 1.4m for Stair B. Survivor accounts
[5,6] and photographs have suggest that handrails were present on each staircase. The actual size of these
handrails was not found in the available literature and so this was estimated based on evidence from
photographs. Edge effects were taken into consideration reducing the effective width of each stair to
approximately 0.9m for stairs A and C and 1.2 m for stair C. The landings on Stairs A and C were taken as
2.64 m wide and for Stair B the landings were assumed to be 3.0 m. Within the model, the total distance
travelled, along the stair slope, per floor was 11.2 m (36.7 feet) for Stair A and C and 11.9 m (39.2 feet) for
Stair B. As already described, there were a number of transfer corridors (see figure 2b) within the building
effectively requiring occupants to leave a staircase, travel along a ‘short’ protected corridor before re-entering
the staircase. While the exact details relating to the length of these transfer corridors was not available they
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
have been represented within the model based on information available from the internet [8,9]. As a result, the
total travel distance represented within the model may not precisely represent the actual travel distance within
the building. However, as the total stair length is more than 900 m, these small discrepancies are not expected
to have a large impact on the final results.
The Population
Within the model the population was distributed only on the rented floors. So the floors known to
have no tenants such as the sky lobbies, machine floors and non-rented floors were left unoccupied. This
meant that there were no people on floor 1 (lobby); floors 2-6 (not rented); floors 7-8, 41-42, 75-76, 107-110
(mechanical floors) and floor 78 (sky lobby). In total three different sized populations consisting of 5,000,
7,000 and 25,000 people were considered. The 5,000 and 7,000 cases are intended to represent the spread in
the number of people thought to have been within WTC1 at the time of the attack. Here we will only consider
the 7,000 population. From press accounts it is thought that 1,432 people in WTC1 died, this included
essentially everyone that was above the 91st floor (i.e. floors 92-110) and a few people on the lower levels
[1,4] resulting in 5568 survivors able to evacuate from WTC1. These people were distributed evenly amongst
the remaining 77 floors producing an average number of 72 people per occupied floor and a total of 5,544
people within the entire simulation (referred to as Population 1). Another population distribution (referred to as
Population 2) was based on the assumption that a total of 304 people were killed on the impact floors and that
a total of 6,696 people were able to evacuate, 5,544 below the impact floors and 1,152 people above the
impact floors. This overestimates the suspected potential survivors above the impact floors by 205 however,
at the time of this analysis this was not known. The survivors above the impact zone were also uniformly
distributed. The number of people per floor was spread randomly across the open floor space in the
geometry. The third population distribution considered in this paper is intended to represent the maximum
building occupancy. Using a load factor of 268 people per floor produces a total building population of 24,924
across all the occupied floors. The population below the impact floors and thus able to evacuate in this case
consists of 20,636 people (referred to as Population 3). The remaining 4,288 are assumed to be either impact
victims or trapped above the impact floors.
The default buildingEXODUS age group settings (17-29, 30-50 and 51-80 years of age) were used as there
was no better information available at the time. It was further decided that 30% of the population would be in
the youngest age group, 50% in the middle age group and 20% in the older age group. The gender distribution
was set at 65% males and 35% females based on information from the media accounts [5,6]. The default
maximum travel speed settings available within buildingEXODUS, which are functions of age and gender were
also adopted for these simulations. The response time distribution used within the simulations (see Table 1)
was based on media account data [5,6]. Individuals were assigned random response times based on the
response time band in which they were assigned. It should be noted that the response time data from the
study was based on the analysis of accounts from 58 people. It is acknowledged that this is a small sample
and may not be representative of the entire building; however it was consider the best data relating to the
incident available at the time of the analysis.
Table 1: Response time distribution used within the WTC simulations
Response time Range % of population
0 – 8 minutes 77
8 – 63 minutes 19
63 – 64 minutes 4
The Scenarios
For each of the 5,000, 7,000 and 25,000 populations, a number of scenarios were investigated. These
included cases in which; each population was assumed to react instantaneously (i.e. zero response time), each
population assumed the response time distribution shown in Table 1, a single staircase remained in tact above
the impact zone and rescue personnel were inserted into the building. In this paper we will briefly consider a
subset of these cases as outlined in Table 2.
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
Accurate information regarding the number of rescue service personnel that may have entered WTC1
during the evacuation was not available at the time of the analysis. For the scenarios involving rescue services
it was decided to investigate the impact of 300 rescue workers being sent into the building on the overall
evacuation times. Two types of scenario were considered. In the first case, the rescue workers were sent
into the building via Staircase B. In the second scenario, the rescue workers entered the building through all
three stairs, 100 per staircase. From media accounts it was found that the fire fighters started their rescue
attempt at approximately 09:00. It was therefore decided to insert the fire fighters into the simulation 14
minutes into the simulation. The procedures used by the fire fighters were not known at the time of this
analysis and so they were represented as 30 teams of 10 people. The fire fighters were generated using the
buildingEXODUS source node capability and for the situation where just Stair B was used, they were generated
in the Lobby and when all three stairs were used, they were also generated in the open floor space just outside
Stairs A and C on the Mezzanine. Each member of the team would enter the staircase in a 30 second burst and
there would be a 1 minute break before the next team was generated. Thus for the cases involving only Stair
B, the first fire fighter was generated after 14 minutes and the last after 58 minutes. For the cases involving all
three stairs, the first fire fighters were generated after 14 minutes and the last after 28 minutes. The fire
fighters were given an itinerary task instructing them to go to the 91st floor. Once they entered the 91st floor
they disappeared from the simulation.
The physical settings for all the fire fighters were identical. They assumed the default
buildingEXODUS settings for 30 year old males, with several important changes. The fire fighter travel speed
up the stairs was arbitrarily reduced by 60% from the default value (effectively 0.38 m/s). This was intended
to represent the fact that the fire fighters would be carrying a considerable amount of equipment and that
fatigue would likely set in during the ascent. While buildingEXODUS does not currently have a fatigue model,
it would have been possible to insert into the itineraries of the fire fighting rest periods of fixed duration after
ascending a certain number of floors. However, as this procedural information was not available at the time of
the analysis it was not included and is left for further work. In addition, the buildingEXODUS parameter
“DRIVE” was set arbitrarily high for each fire fighter. Within buildingEXODUS this means that whenever they
were involved in a conflict with other people (of lower drive) they would always win the conflict. In effect,
this would mean that building occupants would stand aside and let the fire fighter pass each time they vied for
space on the stairs.
Table 2: Summary of Scenarios considered in this paper.
Scenario Population Response Time State of Stairs Rescue Services
1a 1 survey based severed absent
1b 1 survey based severed present
2a 2 survey based single surviving absent
3a 3 survey based severed absent
Other buildingEXODUS specific behavioural settings that were set included enabling the staircase
packing parameter. This means that the stairs can be occupied to their full capacity if necessary. Finally,
occupants on each floor were attracted to their closest entrance into the core region. From there they would
select their nearest staircase entrance.
RESULTS AND DISCUSSION
The first simulations involved testing the model. This was to ensure that the model was correctly
connected and involved sending a 40 year old male and a 40 year old female down the whole length of each
stair. The statistics for these simulations are presented within Table 3. Clearly these times and speeds are
very fast. However, it must be remembered that the buildingEXODUS model does not include a fatigue sub-
model and so people within the simulation do not tire and as a result they are able to maintain a constant
descent speed. The results shown in Table 3 could be argued to be between 50 – 100% faster than what
would be expected for a lone individual descending some 100 floors. It does not however follow that the
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
results for the full simulation will necessarily also be twice as fast as what may be expected. This is because
not everyone in the simulation will be a candidate for possible fatigue effects due to their starting locations.
Furthermore, the interaction between individuals and the resulting congestion that develops on the stairs and
the stair entry/exits points will effectively slow people down and not offer them the opportunity to attain their
maximum unimpeded travel speeds.
Table 3: Results for test simulations involving single occupant
Stair Time
Male Time
Female Number of floors Travel distance Travel Speed m/s
Male:female
A 16 min 51s 20 min 30s 108 951 m 0.94 : 0.77
B 15 min 50s 19 min 28s 107 879 m 0.92 : 0.75
C 16 min 59s 20 min 18s 108 951 m 0.93 : 0.78
As buildingEXODUS is a stochastic based simulation tool, it is necessary to repeat simulations a
number of times in order to generate a distribution of results. For the main scenarios investigated in this paper,
each case was repeated five times and average times are presented here. In addition, the total evacuation times
quoted in this paper are for the occupants to exit the staircase on which they are traveling, not to exit the
building.
Scenario 1a: 5,544 survivors, survey based response times, staircase severed above 91st floor, no fire
fighters.
This scenario is an attempt to reproduce the primary events of the actual incident. The population size
and distribution matches the best estimates available at the time and the response time distribution is derived
from the media accounts. This simulation involved 5,544 people who were able to evacuate the structure from
the 91st floor and below. This scenario, as with all the cases described in this paper, was executed using a
2.66 GHz Pentium PC with 4 GB RAM. The run time for a single simulation was approximately 8 minutes 30
seconds.
Table 4: Summary of results for Scenario 1a sensitivity analysis (average across 5 repeat simulations)
Scenario Average Total
Evacuation Time Last Stair to
finish most often Average Spread in
stair finish times
B: Default stair speeds, building
based response time distribution 1 hour 21 minutes C 1.3 minutes
A: Default stair speeds, floor based
response time distribution 1 hour 32 minutes C 1.5 minutes
D: 80% stair speeds, building
based response times 1 hour 30 minutes A 1.3 minutes
C: 80% stair speeds, floor based
response time distribution 1 hour 34.6 minutes C 3.3 minutes
A sensitivity analysis involving the application of the response time distribution and stair speeds was
undertaken in order to determine model response to these parameters. The identified response time
distribution as described above was either enforced on each floor or applied over the entire building. In both
cases while the number of people on each floor and their precise starting locations do not change, the response
times of people at particular locations will change. When enforced on a floor by floor basis, this reduces the
spread in the variability for repeat simulations as each repeat simulation will have the same distribution of
response times on each floor. In contrast, when enforced building wide, this can result in very different
distributions of response time throughout the building for each repeat simulation. In an attempt to capture the
potential influence of fatigue on the descent rate of individuals, the maximum descent stair speeds of
individuals was systematically reduced from its default value. Here we present the results for both the default
stair descent speed and a 20% reduction in stair descent speeds.
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
The results for these simulations are summarised in Table 4 and Figure 3. The arrival curves for a
representative simulation for each of the cases is depicted in Figure 3. All four cases produce broadly similar
curves suggesting that the evacuation dynamics for these cases are also broadly similar. The shape of the
curves (see in Figure 3) suggests that the evacuation takes place in three phases: an initial start up period
which is relatively short (0 – 300 seconds) and is a function of the response time distribution; a central phase
(300 – 2,500 seconds) in which there is a relatively high egress rate during which approximately 80% of the
occupants have exited; and a final phase (2,500 – 5,500 seconds) in which the egress rate diminishes. It is
worth noting that the simulations with reduced maximum travel speeds produce slower evacuations in the early
phases of the evacuation.
As can be seen in Table 4 the maximum variation between the various scenarios is some 16%, with
the manner in which the response times are distributed creating a 13% variation and the 20% reduction in stair
descent speeds creating a 10% variation. In all the cases examined, Stair B carries the largest number of
people (at least 42% of the building population) but Stair A or C is always the last to finish. As the results
produced by each of these cases are broadly similar and there is little justification in arbitrarily reducing the
maximum stair descent speed for all the occupants, the Case B model setup is preferred and was selected for
further analysis.
Figure 3: Comparison of arrival time curves for Scenario 1a sensitivity analysis
cases
An informative personal parameter that is determined by buildingEXODUS for each individual is the
Cumulative Wait Time (CWT). This is a measure of the total amount of time that a person wastes in
congestion. For this application, this includes the time queuing to get into the staircase and the time queuing
on the stairs. Another useful parameter is the Personal Evacuation Time (PET). This is a measure of the time
each individual requires to evacuate. The ratio of the two is a measure of the evacuation inefficiency incurred
by the individual. A large value suggests that the individuals’ personal evacuation was highly inefficient,
spending most of the evacuation caught in congestion. An average value for this ratio can be determined for
each floor of the building providing a view of the relative evacuation efficiency for each floor. Furthermore,
the average floor evacuation efficiency can be determined for each staircase used by the occupants from that
floor. This then gives a view of the relative evacuation efficiency for each floor as a function of the exit route
(i.e. staircase) used.
This is shown in Figure 4 for Case B. From these graphs we note that the higher up the building a
person starts their evacuation, the more inefficient the evacuation becomes relative to their personal evacuation
time; i.e. the more of their personal evacuation time will be wasted in congestion. Were this trend to continue
without abating, it would suggest that an eventual height would be attained above which the time spent in
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
congestion by occupants would overwhelm the travel time (plus response time) making evacuation by stairs
extremely undesirable, if not untenable. This is not the case in these scenarios, and we find a maximum of
13% of an occupants PET is lost in congestion across all the floors on Stairs A and C and a maximum of 35%
is lost to congestion on the upper floors for Stair B.
These results also suggest that Stair B, which is the largest of the three stairs, was the most heavily
used stair. This is indeed the situation with 44% of the occupants making use of Stair B. For this particular
simulation, it is thought that the majority of this congestion is incurred at the stair entrance on each floor. This
is supported by the observation that Stair B was heavily used, yet the evacuation time for Stair B was always
quicker than the other stairs. As the average maximum inefficiency level is small (less than 15%) for Stairs A
and C, this suggests that, for the number of people represented in this scenario combined with the imposed
response time distribution, the building provides sufficient staircase and exit capacity. Furthermore, the
primary factors driving this scenario are the response time distribution coupled with the size of the population
and the travel speeds attained by the individuals. In reviewing these results it should be recalled that all of these
factors are somewhat uncertain.
0
20
40
60
80
100
120
0 2 4 6 8 10 12
(average CWT / average PET) %
Floor
(a) Stair A
0
20
40
60
80
100
120
0 10 20 30 40
(average CWT / average PET) %
Floor
(b) Stair B
0
20
40
60
80
100
120
0 2 4 6 8 10 12 14
(average CWT / average PET) %
Floor
(c) Stair C
Figure 4: Evacuation efficiency per floor as a function of Staircase used for Scenario 1a Case B
While Case C produces the longest evacuation time and the closest agreement with the actual incident,
Case B (the preferred model setup) produces an average evacuation time of approximately 1 hour 21 minutes
with an average flow rate of 68 survivors/minute. If we compare this with the actual incident then it is
estimated that 5,544 people managed to evacuate WTC1 some time before the building collapsed which was 1
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
hour 42 minutes 5 seconds after it was attacked. If we assume that the last person managed to evacuate the
building just prior to building collapse, then this is equivalent to an average flow rate of 54 survivors/minute.
The results produced from this simulation appear to be in good agreement with the available information.
However, it should be noted that there were no fire fighters present in the simulation and hence no potential
obstructions on the stairs.
For the population and response time distributions used in this simulation, the model predictions
suggest that the occupants did not experience high levels of congestion on the stairs. If this is the case, it is
suggested that it is unlikely that the presence of fire fighters entering the building to undertake search and
rescue operations would have had a significant detrimental impact on the evacuation times. This case is
examined in the next scenario.
Scenario 1b: 5,544 survivors, survey based response times, staircase severed above 91st floor, fire
fighters inserted.
This scenario is an attempt to gauge the impact that the fire fighters may have had on the overall
evacuation efficiency. In the case considered here, the fire fighters are all inserted at the base of Stair B. As
described above, a total of 300 rescue workers are inserted into the building 14 minutes after the start of the
evacuation. It was noted in the previous case (Scenario 1a) that Stair B was the heaviest utilised staircase and
it was not the last stair to finish. As in Scenario 1a, the population size and distribution matches the best
estimates available at the time and the response time distribution is derived from the media accounts. This
simulation involved 5,544 people who were able to evacuate the structure from the 91st floor and below. The
average evacuation time produced for this simulation was approximately 1 hour 22 minutes. This is only one
minute longer than the result for the equivalent case without fire fighters present (i.e. Scenario 1a Case B).
Stair B
0
20
40
60
80
100
120
010 20 30 40 50
(Average CWT / Average PET) %
Floor
B1
B2
0
1000
2000
3000
4000
5000
6000
7000
8000
01000 2000 3000 4000 5000 6000
Evacuation time (seconds)
Arrival No.
B3
B2
B1
Case B3
Case B1
Case B2
(a) Evacuation efficiency per floor for Stair B,
Scenario 1b (Case B2) and Scenario 1a (Case B1). (b) Arrival curves for Scenarios 1a (Case B1), 1b
(Case B2) and 2a (Case B3)
Figure 5: Evacuation data for Scenarios 1a, 1b and 2a
As expected, introducing the fire fighters into the evacuation simulation did not have a significant
effect on the total evacuation time for the simulation. However, if we consider the average floor evacuation
inefficiency curve for Stair B we note that this has increased from a maximum of 35% in Scenario 1a to a
maximum of approximately 45% in Scenario 1b (see Figure 5a). Thus on average the people using Stair B
were hindered by the passage of the fire fighters as on average, their time spent in congestion has increased as
a fraction of their total evacuation time. Those who experience the greatest effect are the occupants starting
from the highest floors. The impact of the fire fighters can also been seen in the arrivals graphs for this
scenario (see Figure 5b). In the mid portion of the evacuation sequence there is a dip in the arrival rate of
occupants but after the passage of the fire fighters the arrival rate returns to normal. In addition, the entire
evacuation sequence is extended slightly due to the earlier dip in arrival rates. It is also worth noting that in the
previous case (Scenario 1a), Stair B was always the first to finish. However, in Scenario 1b, Stair B is almost
always the last stair to finish. The average finish time for Stair B has increased from an average of 1 hour 20
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
minutes to an average of 1 hour 22 minutes.
The primary factors driving this scenario are similar to those of the previous case, namely the
response time distribution coupled with the size of the population and the travel speeds attained by the
individuals. In addition, the nature of the fire fighter insertion and fire fighter physical capabilities, will also
dictate the overall outcome of this scenario. In reviewing these results it should be recalled that all of these
factors are somewhat uncertain.
Scenario 2a: 6,696 people, survey based response times, single surviving staircase above 91st floor, no
fire fighters.
In this scenario we attempt to assess whether it was possible for the surviving building occupants
trapped above the impact floors to have successfully evacuated prior to building collapse had a single staircase
remained in tact throughout the impact zone. This model is based on Scenario 1a and so all the assumptions
and caveats which applied to that simulation equally apply here.
In the work presented here, Stair B is selected as the surviving staircase. As in Scenario 1a, the
population size and distribution matches the best estimates available at the time and the response time
distribution is derived from media accounts. This simulation involved 5,544 people who were able to evacuate
the structure from the 91st floor and below, but in addition, a further 1152 people are included above the 91st
floor. These people are given the same response time distribution as the people below the impact zone. It is
accepted that this assumption has little justification as the media accounts do not include response time data for
those trapped above the impact floors. It is plausible that these people would have reacted with greater haste
than those below the impact floors for which we have response time data. Equally, some may have
undertaken other actions such as attempt to evacuate via the roof prior to attempting to descend via the stairs.
As no other reliable response time data was available it was decided to use the media account data.
Stair B
0
20
40
60
80
100
120
0 10 20 30 40 50
(Average CWT / Average PET) %
Floor
B1
B3
Figure 6: Evacuation efficiency per floor for those using Stair B for Scenario 1a and Scenario 1b
The average evacuation time produced for this simulation was approximately 1 hour 24 minutes.
This is approximately 3 minutes longer than the time for Scenario 1a. This modest increase in evacuation time
is somewhat unexpected but can be explained by the evacuation dynamics. As noted in Scenario 1a, due to the
numbers of people involved, their physical distribution and the applied response time distribution, all three stairs
are not utilised to their full capacity at any one time, and certainly not in the early and later stages of the
evacuation. In the early stages of the evacuation this allows the early responders above the 91st floor to make
good progress down the stairs. In the later stages of the evacuation this enables the long responders above the
91st floor to utilize the spare stair capacity.
This can be seen more clearly by studying the evacuation efficiency plot for Stair B (see Figure 6) and
the arrivals graph for this scenario (see Figure 5b). The presence of the additional people in this scenario is
felt all the way down to approximately the 60th floor as above the 60th floor we note the average floor
evacuation inefficiency increases slightly (see Figure 6). The occupants below the 60th floor are effectively
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
able to evacuate as if these additional occupants were not present. Comparing the arrivals curve for this
scenario with that of Scenario 1a (see Figure 5b) we note that this is indeed the case as the two curves follow
each other precisely up to approximately 2,300 seconds. After this time, the arrivals curve for Scenario 1a
begins to flatten, reducing its slope (and hence exit flow rate) as there are insufficient people remaining within
the structure to maintain the exiting rate. However, in Scenario 2a there are additional people available to
maintain the flow rate and so the curve continues with its original trajectory.
This analysis suggests that had a single stair survived the impact above the 91st floor, survivors above
the impact zone would have been able to escape the building prior to its collapse. It should also be noted that
this analysis assumed a greater number of potential survivors above the impact floors than is likely to have
been the case. However, this reinforces the conjecture that the survivors would have been able to escape prior
to building collapse had an exit route been available.
Scenario 3a:20,636 survivors, survey based response times, staircase severed above 91st floor, no fire
fighters.
This scenario is an attempt to explore the possible outcomes assuming that the building was occupied
with a normal working load of people. The simulation thus involved some 24,924 people, of which 20,636
people were considered able to evacuate the structure from the 91st floor and below. In total, 4,288 people
were considered unable to evacuate the building from above the 91st floor and are thus considered fatalities.
The run time for a single simulation using this population was 1.5-2 hours.
The arrival curve for a typical simulation of this scenario is depicted in Figure 7. This shows a steady
increase in the numbers of people evacuating the building up to approximately 1 hour 25 minutes into the
evacuation. After this point the evacuation rate drops of to a lower constant value. As in the other scenarios,
Stair B takes the majority of the building population (45%) but, unlike the other scenarios, it now produces the
longest evacuation time. The change in slope noted in the evacuation curve is due to Stairs A and C finishing
after 1 hour 22 minutes and 1 hour 25 minutes respectively. The average evacuation time produced for this
simulation was approximately 2 hours 14 minutes. This produces an average flow rate of 154
survivors/minute. This is some 2.3 times faster than the flow rate achieved in Scenario 1a. These results
imply that at the time of WTC1 collapse, some 2,246 people would have still been on the stairs trying to exit
the building, in addition to the 4,288 people caught above the 91st floor. Thus, it is likely that there could have
been 6,534 fatalities for WTC1 under these conditions. In addition, unlike in Scenario 1b, had fire fighters
been inserted into Stair B during the evacuation, this would have had a more significant effect, delaying the
evacuation process even longer and resulting in a higher number of fatalities.
0
5000
10000
15000
20000
25000
0 2000 4000 6000 8000 10000
Evacuation time (seconds)
Arrival No
Figure 7: Evacuation curve for Scenario 3a
The impact of the larger population is shown in Figure 8 where the evacuation inefficiency for Stair B
has increased dramatically. The evacuation inefficiency reaches a maximum of approximately 50% for people
using Stair A while for Stairs C and B maxima of 60% and 80% respectively are attained. Clearly congestion
on all the stairs has reached critical levels with Stair B being the most congested due to the disproportionate
use of the stair. Clearly for such a large scale evacuation it is desirable to achieve a balanced distribution of
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
people between the staircases which is biased according to stair capacity. Evacuation procedures should be
developed that attempt to take this into consideration.
0
20
40
60
80
100
120
0 20 40 60 80 100
(average CWT / average PET) %
Floor
Figure 8: Evacuation efficiency per floor for Staircase B in Scenario 3a
CONCLUSIONS
This work has been an attempt to explore the use of the buildingEXODUS evacuation modelling tool
for forensic investigation. The model has been used in an attempt to reproduce the evacuation of the North
Tower of the WTC resulting from the terrorist attack of 11 September 2001. While the work is preliminary in
nature, early results suggest that the model is capable of reproducing the broad trends in this disaster. The
model predicts the total evacuation time of the building for 5,544 survivors to be between 1 hour 21 minutes
and 1 hour 35 minutes, depending on the precise nature of the model assumptions. These times compare
favourably with the observation that the building collapsed after some 1 hour 42 minutes and supports the
view that everyone that was able to escape from WTC1 on the day of the incident did get out. The model
results also suggest that the insertion of the fire fighters into the building had minimal detrimental impact on the
overall evacuation of the building.
The model was also used to explore several additional scenarios, in particular whether those people
trapped above the impact floors could have managed to escape prior to building collapse if at least one stair had
remained intact through the impact zone and what would have been the likely outcome had the building been
fully occupied with 25,000 people. The model suggests that had Stair B remained in tact throughout the
building an additional 1,152 survivors trapped above the impact floor could have escaped prolonging the
evacuation by approximately a further 3 minutes. These results suggest that had at least one staircase
survived from top to bottom, it is possible that everyone that survived the initial trauma of the impact could
have managed to safely escape. This underlines the importance of staircase dispersal within buildings. It is
essential to make it less likely to lose all means of escape in the event of plausible catastrophic incidents, and
that staircases are sufficiently hardened to withstand plausible threats. Had WTC1 been fully occupied, and
using the idealised assumptions of this study, the predicted time required to evacuate the building is estimated
to be 2 hours 14 minutes. This implies that at the time of WTC1 collapse, the expected death toll would be
6,534, with some 2,246 people caught on the stairs and 4,288 people either killed in the impact or caught
above the 91st floor. The results suggest that a mass evacuation of the building in a 9/11 scenario would lead
to extremely heavy congestion on the stairs leading to an inefficient evacuation. For such a large scale
evacuation it is essential that a balanced distribution (with respect to stair and exit capacity) of people between
the staircases is achieved. Evacuation procedures should be developed that attempt to take this into
consideration.
This work and the results generated must be viewed as preliminary in nature as further research
remains to be done. In particular it is desirable that the analysis is repeated with; a more accurate
representation of the building layout, a more precise specification of the number of people in the building at the
time of the incident, their initial location and response time distribution, the inclusion of disabled occupants, the
representation of groups and group behaviour and a more precise representation of the fire fighter deployment.
Presented at the 3rd Int Symp on Human Behaviour in Fire,1-3 Sept 2004, Belfast UK.
The existing buildingEXODUS model is capable of including a representation of all of these features, however
it is essential that the necessary information is available and provided to the model developers. It is anticipated
that this information will be provided as part of a UK project investigating the evacuation performance of the
WTC twin towers. The project called HEED – High-rise Evacuation Evaluation Database – funded by the UK
EPSRC (project GR/S74201/01) and involving the Universities of Greenwich, Ulster and Liverpool, aims to
interview 2000 survivors of the WTC twin towers evacuation. The project will also continue the modelling
described here and extend it to the analysis of WTC2, which stood for a much shorter period of time, but in
which the occupants started the evacuation process prior to the assault on their building.
There are also some factors that at present cannot be easily and reliably represented within evacuation
models such as fatigue. On descending 110 floors the average person will tire and slow down; they may even
stop for a rest. While it is possible to artificially include a representation of fatigue within the simulation, it is
desirable that a fatigue model is developed capable of predicting the onset and implications of fatigue on
evacuation performance. Calculations of the type presented in this paper are providing us with insight into
building performance under extreme conditions. It is hoped that calculations of this type will assist in building
safer buildings, and develop procedures for existing buildings that will assist in maximising chances of survival
in extreme events.
REFERENCES
1. Cauchon D., For Many on Sept 11, Survival was no accident, USA Today 19/12/01.
2. Gwynne, S., Galea, E., R.,Lawrence, L. and Filippidis, L., “Modelling Occupant interaction with fire
conditions using the buildingEXODUS evacuation model”, Fire Safety Journal, 36, pp 327-357, 2001.
3. Park J., Gwynne S., Galea E.R., and Lawrence P., Validating the buildingEXODUS evacuation model
using data from an unannounced trial evacuation. Proc 2nd Int Pedestrian and Evacuation Dynamics
Conference, Ed: E.R.Galea, CMS Press, Greenwich, UK, ISBN 1904521088, pp295-306, 2003.
4. Cauchon D., Not found or not existing, 40 names to leave WTC death toll, USA Today 29/10/03.
5. Galea E.R. and Blake S.J, Collection and Analysis of Data relating to the evacuation of the Word Trade
Centre Buildings on 11 September 2001, Report produced for the UK ODPM, to appear 2004.
6. Blake, S.J., Galea, E.R., Westeng H. and Dixon, A.J.P, An Analysis of human behaviour during the
WTC disaster of 11 September 2001 based on published survivor accounts. To appear in this
proceedings, 2004.
7. Federal Emergency Management Agency, Federal insurance and Mitigation Administration,
Washington, DC, FEMA Region II, New York, New York, FEMA 403 / May 2002, Chapter 1 & 2;
World Trade Center Building Performance Study: Data Collection, Preliminary Observations, and
Recommendations
8. The New York Times; Interactives: Inside the Towers.
http://www.nytimes.com/2002/05/26/nyregion/26WTC.html
9. USAToday; Interactive documentary: Getting out alive
http://www.usatoday.com/news/sept11/year-later-index.htm
10. Architectural Record, Windows On The World, May 1977.
11. Private Communication, 2002.
