See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235782447

Modeling Emergency Evacuation of Individuals with Disabilities in a Densely

Populated Airport

Article  in  Transportation Research Record Journal of the Transportation Research Board · October 2011
DOI: 10.3141/2206-05

CITATIONS READS
27 832

4 authors, including:

Yong Seog Kim Keith Christensen

Utah State University Utah State University
134 PUBLICATIONS   1,277 CITATIONS    47 PUBLICATIONS   650 CITATIONS   

SEE PROFILE SEE PROFILE

Anthony Chen
The Hong Kong Polytechnic University
208 PUBLICATIONS   5,738 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Resilience of transportation networks View project

DEVELOPING A MULTI-OBJECTIVE SEQUENCING MODEL FOR AIRCRAFT ARRIVAL SCHEDULING: CASE OF SUVARNNABHUMI AIRPORT IN THAILAND View project

All content following this page was uploaded by Anthony Chen on 20 May 2014.

The user has requested enhancement of the downloaded file.

Modeling Emergency Evacuation
of Individuals with Disabilities
in a Densely Populated Airport
Matthew Manley, Yong Seog Kim, Keith Christensen, and Anthony Chen

Emergency evacuation from airports is an important consideration, given Agent-based modeling (ABM) is a popular microscopic simulation
the continuing occurrence of natural and human-caused disasters affect- technique for representing social dynamic systems characterized by a
ing these locations. These incidents have focused attention on the needs of wide variety of motives, actions, and interactions. ABM has been used
individuals with disabilities, who are more likely to suffer during emer- to study diverse topics such as infectious disease epidemiology (4),
gency situations. The agent-based model presented in this paper can be stock trading strategies (5), and traffic congestion (6). In this study,
used by engineering and management professionals in design and plan- ABM is used to assess airport evacuation performance in the event of
ning efforts to estimate the evacuation performance of heterogeneous a bomb discovery within a terminal. The results are important for
populations from airports. The model classifies the environment accord- engineering and management practitioners because they demonstrate
ing to accessibility characteristics encompassing various conditions that the limitations of the building design, identify the individuals most
have a disproportionate effect on the behavior of individuals with disabil- at risk, and reveal areas prone to bottlenecks or clogging.
ities during an evacuation. The results of a simulation experiment demon- The remainder of the paper is organized as follows. The first
strate some of the limitations of the pier airport design and identify which section reviews the relative advantages and disadvantages of three
individuals are most at risk: those with lower stamina and those using general approaches to pedestrian crowd simulation research: macro-
wheelchairs. The results also reveal areas of the airport that are prone to scopic, microscopic, and mesoscopic models. The second section
bottlenecks or clogging. explains the system architecture in terms of organization, implemen-
tation, and underlying assumptions. The model calibration process
and validation results using a university campus building are also
Safely evacuating people from airports is an important consideration
described. The third section presents the results and practical impli-
given the continuing occurrence of emergencies arising from both nat-
cations of a simulation experiment. The fourth section concludes
ural and human-caused disasters. For example, Terminal 2 of the San
with a discussion of the limitations of the study and directions for
Diego, California, airport was evacuated on April 4, 2010, in response
future research.
to an earthquake that occurred earlier that morning (1). Additionally,
the international terminal of the Los Angeles, California, airport
was evacuated on July 4, 2002, after a gunman opened fire at one
LITERATURE REVIEW
of the ticketing counters (2). Unfortunately such events are not rare
occurrences. There are three general approaches to simulating pedestrian crowd
At the same time, incidents such as these have focused attention on behavior: macroscopic, microscopic, and mesoscopic. Macroscopic
the challenges of evacuating individuals with disabilities. More than modeling is characterized as a top-down approach in which collec-
12% of the working population has some form of physical, sensory, tive pedestrian dynamics such as spatial density or average veloc-
or mental disability (3) and these individuals are more likely to suffer ity are related to model parameters through a closed-form formula.
during a disaster situation. A disability is a condition that limits major Macroscopic models are reductions of real-world systems offering
life activities such as walking, lifting, hearing, seeing, remembering, researchers the ability to focus only on specific aspects of the model
or concentrating. Consequently, studies examining evacuations of without worrying about potentially confusing details. Examples of
heterogeneous populations, people with and without disabilities, from macroscopic models include queuing models (7 ) and stochastic
airports are needed to better ensure their safety during emergency models (8).
situations. A limitation of this approach, however, is the inability to explain
several self-organizing collective behaviors empirically observed in
M. Manley, Jon M. Huntsman School of Business, and Y. S. Kim, Department of pedestrian crowds (9). Self-organizing behaviors are naturally occur-
Management Information Systems, Jon M. Huntsman School of Business, Utah ring behaviors that arise without external influence from signals or
State University, Logan, UT 84322-3515. K. Christensen, Department of Land-
conventions. Examples of self-organizing collective behavior include
scape Architecture and Environmental Planning, Utah State University, Logan, UT
84322-4005. A. Chen, Department of Civil and Environmental Engineering, Utah lane formation and clogging at narrow passages (9).
State University, Logan, UT 84322-4110. Corresponding author: M. Manley, Microscopic modeling is characterized as a bottom-up approach in
[email protected]. which pedestrians are modeled as individual entities having several
attributes such as speed and size. Formulae encapsulating spatial tran-
Transportation Research Record: Journal of the Transportation Research Board,
No. 2206, Transportation Research Board of the National Academies, Washington,
sition probabilities are repeatedly applied leading to temporal changes
D.C., 2011, pp. 32–38. in state or behavior. Like macroscopic models, microscopic models
DOI: 10.3141/2206-05 are simplifications of real-world systems and thus enjoy the same

32
Manley, Kim, Christensen, and Chen 33

benefits in terms of ease of understanding and control. Examples of Environment Component

microscopic models include particle models (10) and cellular automata
(CA) models (11). The environment component is responsible for creating the virtual
Microscopic models have been shown to successfully reproduce world. This is accomplished by importing Raster geographical infor-
self-organizing phenomena, such as lane formation, as demonstrated mation system (GIS) data. Raster GIS data are an abstraction of the
in Henderson (10) and Burstedde et al. (11). In particular, the social real world in which spatial data is expressed as a two-dimensional grid
force model developed by Helbing (12) has been shown to repro- of cells having a finite length and width with the spatial position of
duce lane formation, clogging, and several other phenomena (9, 13). each cell implicit in their ordering. Thus, BUMMPEE’s virtual envi-
However, a limitation of this model is its computational intensity, ronment is a discretized version of the real world and similar to many
which makes simulation of large crowds difficult on traditional single- other models based on CA theory.
processor systems. Quinn et al. (14) utilized a parallel-computing Unlike other models, however, BUMMPEE classifies the environ-
approach to overcome this obstacle. Song et al. (15) followed with ment in terms of accessibility characteristics encompassing various
the CA with force essentials model (CAFE), a rule-based translation conditions which have been shown to have a disproportionate effect
of social forces, demonstrating comparable results in a discrete envi- on the behavior of individuals with disabilities during an emergency
ronment (16). In all cases microscopic models are limited by the evacuation (23). The accessibility characteristics are (a) exit charac-
realism of the underlying spatial movement algorithms. Quantify- ter, (b) route character, and (c) obstacle character, each of which refers
ing human behavior is challenging because the factors involved vary to the functional demand imposed by the environment on an individ-
significantly between people and environments. ual’s competency to meet it in relation to his or her disability. Specif-
A third approach to pedestrian crowd simulation is mesoscopic ically, for each cell, accessibility characteristics are defined by ordinal
modeling, which may be viewed as a combination of both macro- position within a seven-digit number. For example, at ordinal posi-
scopic and microscopic techniques. A good example is Hoogendoorn tion, one exit character e is defined for cell x as all exit [e(x) = 1], lim-
and Bovy’s (17) gas-kinetic model. In this model, though pedestrian ited exit [e(x) = 2], no exit [e(x) = 3], or refuge [e(x) = 4]. Similarly,
spatial movement is individually specified, it is still dependent on at ordinal position five, route character r is defined for cell x as default
aggregate flow conditions rather than interactions with other pedestri- speed [r(x) = 1], stair speed [r(x) = 2], obstacle speed [r(x) = 3],
ans. Unfortunately, mesoscopic models suffer from the same limita- and so on.
tion as macroscopic models. In simulation experiments involving
bidirectional flows, Hoogendoorn and Bovy (17) found that pedestri-
ans did not demonstrate lane formation or any preferences in relation Population Component
to individuals traveling in an opposing direction.
Based on the available literature, microscopic models appear best The population component is responsible for creating virtual evacuees
suited for realistically representing self-organizing collective behav- or agents. Agent populations are specified as distributions of six dis-
iors observed in empirical crowd studies. ABM is a particular sub- tinct types of people: (a) nondisabled, (b) motorized wheelchair users,
category of microscopic modeling focused on the emergent process. (c) nonmotorized wheelchair users, (d) visually impaired, (e) hearing
Agent-based models often reveal novel and coherent structures aris- impaired, and ( f ) stamina impaired. Each type is distinguished
ing at the level of the aggregate system that cannot be seen by exam- by physical and psychological characteristics that address several
ining agents in isolation (18). For example, a crowd clamoring to established criteria for describing the functional competency of
purchase an item might be moving in a generally different direc- people with disabilities in the general population [e.g., speed, size,
tion than many of the individuals comprising it. According to ability to negotiate terrain, and others (23)], the expression of which
Bonabeau (19), “ABM is, by its very nature, the canonical approach is realized through an object-oriented programming structure.
to modeling emergent phenomena.”
Several advantages have contributed to ABM’s growing popular-
ity in recent years, including its simplicity and flexibility (20). Per- Visualization Component
haps ABM’s most significant advantage, however, lies in the ability
to incorporate multiple perspectives (21). Other modeling approaches The visualization component is responsible for generating the user
are often forced to make limiting assumptions that may contradict interface and graphical map display. In addition to the structure itself,
real-life behavior. ABM, however, is ideal for modeling problems users can selectively highlight and view important features such as
where conflicting interests are essential (20), which is why the ABM exits, connections, and stairwells. The accessibility characteristics of
approach has been adopted for this study. individual cells can also be displayed.

RESEARCH MODEL Simulation Component

Model Architecture The simulation component is responsible for controlling the progress
of the evacuation by sending movement signals to agents at regular
The ABM developed for this study is called BUMMPEE (Version time intervals. Upon receiving a signal, an agent calculates the prob-
3.0) (i.e., bottom-up modeling of mass pedestrian flows—implications ability for movement p as the its velocity modified by the route char-
for the effective egress of individuals with disabilities). A thorough acter for the current location vrij in relation to the signal interval t,
description of BUMMPEE is provided in Manley et al. (22). How- which is compared to a real-valued random number u generated from
ever, a condensed version is provided here for the reader’s conve- a uniform distribution. Thus the decision to move is defined by the
nience. The model architecture consists of four logical components: following algorithm:
(a) environment, (b) population, (c) visualization, and (d) simulation.
A discussion of each component follows. p = vrij t (1)
34 Transportation Research Record 2206

if ( p > u ( 0, 1)) move (2) Friction refers to the necessity of physically slowing down when
an agent is in contact with an obstacle or another agent. Friction is
Given the decision to move, agents exhibit three behaviors: expressed as the probability of moving from the current location to the
(a) response delay, (b) destination choice, and (c) direction choice. direction choice cell. The friction probability is determined in a man-
Response delay refers to the human tendency to wait some time ner similar to repulsion probability or influenced by the hardness of
before responding to an evacuation alert (24). At the beginning of a adjacent cells. For example, an agent who desires to move into a cell
simulation, each agent is assigned a random delay generated from a next to a wall is less likely to do so than one who desires to move into
normal distribution having M = 29s and S2 = 9s (25). Thereafter the a cell with no surrounding obstacles. The application of friction
agent checks to see if current time has exceeded the delay. If so, the probability to agent movement over time results in slower speeds in
agent executes destination and direction choice behaviors. If not, narrow passageways and so forth.
the agent waits until the next opportunity for movement.
Destination choice behavior refers to the selection of intermediate Model Validation
and final destinations based on available exits according to agent type
preferences. When the simulation starts, each agent chooses a final The model was validated by comparing the mean evacuation time of
destination from all available exits based on proximity to the agent’s several simulations to the evacuation time of a physical exercise held
current location. If the agent is not on the same level as the final in the same structure. Since the airport has never had occasion to con-
destination, an intermediate one is chosen from the set of available duct a complete evacuation of all terminals, the model was validated
connecting features, such as a stairways or elevators, using the same using a four-story office complex on the university campus for which
mechanism. such data were available. A physical evacuation of this building
Direction choice behavior refers to the selection of the next prospec- occurred on September 14, 2005, during which 71 participants evac-
tive location or cell. Given the decision to move, an agent evaluates uated by way of three exits on the ground floor. Sixty-five evacuees
the obstacle character of each cell corresponding to the set of four were nondisabled, one was visually impaired, one used a motorized
directional choices {north, south, east, west}. If a cell is impenetra- wheelchair, and four had lower stamina. The time at which the last
ble, it is removed from the set of available choices according to agent evacuee left the building was 155 s. The results of 100 simulations
type characteristics (e.g., the cell is part of a wall or occupied by using the same building and population distribution demonstrated
another agent). Thereafter, the agent selects from the remaining pos- comparable results at a mean evacuation time of 159 s.
sible choices and updates the location accordingly. The final direction The correctness of the model was also visually validated by observ-
choice is dependent on the interaction of social forces as described ing individual agent paths through the building. Trace images gener-
in the next section. ated by the simulator demonstrated realistic path-finding behavior.
Figure 2 presents an example of an individual agent’s path beginning
in the northwest corner of the fourth floor, down the staircase past the
Implementation of Social Forces
third, and out via the main exit on the second. Minor aberrations are
An agent’s direction choice is influenced by three essential forces: explained by interactions with other agents during the evacuation.
(a) attraction, (b) repulsion, and (c) friction (see Figure 1). Attraction
refers to the desire to move toward a destination. Attraction is quanti- EXPERIMENT: ESTIMATION OF THE
fied by means of floor fields, which are duplications of the spatial grid
IMPACT OF BOMB PLACEMENT
for each exit and connecting feature on each level of the virtual envi-
ON EVACUATION OUTCOMES
ronment. Floor field cells contain values indicating the time to destina-
tion (TTD) from their location. For example, an exit cell will have a Objective
TTD value of 0, while one further away is greater. Agents selecting a
direction will first choose the one having the lowest TTD value. The objective of the simulation experiment was to estimate the impact
Repulsion refers to the desire to avoid injury resulting from prox- of bomb placement on evacuation times for individuals with and with-
imity to obstacles or other agents. Repulsion is expressed as the out disabilities. The experiment was conducted using the map of an
probability of accepting or rejecting the first choice cell. Repulsion international airport from the intermountain west region of the United
probability is influenced by the hardness of adjacent cells. For exam- States. The airport complex is a two-story structure consisting of three
ple, a cell situated next to a wall has a greater chance of being rejected terminals and five concourses patterned after the pier terminal design.
than one next to another agent or one without any surrounding obsta-
cles at all. Given rejection of the first choice cell, an agent will choose
the option having the next lowest TTD value.

Repulsion

Attraction

Friction
1st Floor 2nd Floor 3rd Floor 4th Floor
FIGURE 1 Illustration of essential forces
influencing pedestrian movement. FIGURE 2 Agent path from fourth to second floor (ground level).
Manley, Kim, Christensen, and Chen 35

Concourse D
Concourse C

Terminal 3 Terminal 2 Concourse B

Terminal 1
Concourse A
1st Floor 2nd Floor

FIGURE 3 Map of airport.

In this design the passenger processing sequence, including ticketing, cally result in a controlled evacuation of the entire airport or one in
baggage check, and security screening, is centralized in one area within which participants evacuate on the terminal side of the airport as
a terminal while access to aircraft occurs along both sides of long piers opposed to the ramp, where aircraft and other heavy equipment pose
that extend away from it (see Figure 3). Note that the fifth pier or con- additional safety threats. In this case, the airport emergency plan also
course connected to Terminal 3 was not included when the airport was requires the establishment of a 300-ft perimeter around the bomb
mapped. As a result, it is not included in the simulations. itself, or hot zone, which no one is allowed to enter. Note that the hot
Six thousand agents were evacuated during the simulations. Based zone potentially prohibits evacuees from using certain exits or other
on an interview with airport management, this value represented a building features to escape depending on its size and location.
reasonable midpoint in terms of the number of people present at any Four simulation scenarios were specified based on the presence and
given time (R. Berg, personal communication, April 8, 2010). The location of the bomb. Scenario 1 was conducted without a bomb to
diversity and prevalence of individuals with disabilities was assigned establish a baseline for subsequent comparisons. Scenarios 2, 3, and
based on distributions defined by U.S. Census Bureau values (26). 4 simulated a bomb discovery in one of the Explosive Detection Sys-
As a result, 5,161 were nondisabled, 180 used motorized wheel- tem (EDS) devices found in Terminals 1, 2, and 3, respectively. Fifty
chairs, 180 used nonmotorized wheelchairs, 90 had a visual impair- simulations were conducted for each scenario.
ment, 89 had a hearing impairment, and 300 had lower stamina.
The criteria for speed, size, and ability to negotiate obstacles was Results
assigned according to empirical data when available (27, 28) and
accessibility axioms when not. It is assumed that individuals using Scenario 1
wheelchairs are accompanied by a nondisabled assistant to help
negotiate stairs because elevators are typically not accessible during Simulation performance was measured by evacuation time or the time
an emergency evacuation. at which the last agent exited the airport. Both mean evacuation time
The simulation scenarios were based on the discovery of a dirty (MET) and maximum evacuation time (MaxET) were recorded.
bomb within the airport. A dirty bomb is a radiological weapon The MET for all terminals in Scenario 1 was 2,169.84 s (see Table 1).
that combines radiological material with conventional explosives Nondisabled agents displayed the largest MaxET and second
to potentially contaminate a large area. Such a discovery would typi- largest MET for all terminals. However, these results are attributed

TABLE 1 Evacuation Times for Scenario 1 (seconds)

Terminal All Nondisabled Motorized Nonmotorized Visual Lower Stamina Hearing

All
MaxET 2,745.32 2,745.32 2,637.63 2,641.82 2,345.72 2,701.43 2,295.93
MET 2,169.84 2,135.29 1,937.76 1,936.41 1,990.03 2,165.31 1,911.54
SD 148.98 153.91 133.05 129.58 140.20 145.29 164.95
One
MaxET 2,745.32 2,745.32 2,637.63 2,641.82 2,345.72 2,701.43 2,295.93
MET 2,169.08 2,135.29 1,937.76 1,936.41 1,990.03 2,164.56 1,911.54
SD 149.55 153.91 133.05 129.58 140.20 145.84 164.95
Two
MaxET 1,119.53 965.45 900.35 1,119.53 950.15 1,070.85 891.25
MET 1,021.72 855.49 638.37 673.86 837.33 1,017.66 766.47
SD 33.11 34.10 99.90 148.97 42.93 28.54 49.61
Three
MaxET 2,076.70 1,170.93 996.90 1,056.83 1,054.03 2,076.70 991.30
MET 1,120.61 1,020.04 913.08 914.60 931.92 1,108.43 850.24
SD 264.71 43.03 53.47 53.36 62.23 267.31 99.91

NOTE: MaxET = maximum evaluation time; MET = mean evaluation time; SD = standard deviation.
36 Transportation Research Record 2206

Evacuation Curve Evacuation Curve Evacuation Curve

(Nondisabled) (Motorized) (Nonmotorized)
100% 100% 100%

80% 80% 80%

Evacuated Agents
Evacuated Agents
Evacuated Agents

60% 60% 60%

40% 40% 40%

20% 20% 20%

0% 0% 0%
0 1200 2400 3600 0 1200 2400 3600 0 1200 2400 3600
Time (s) Time (s) Time (s)
All Nondisabled All Motorized All Nonmotorized

Evacuation Curve Evacuation Curve Evacuation Curve

(Visual) (Stamina) (Hearing)
100% 100% 100%
80% 80% 80%

Evacuated Agents
Evacuated Agents
Evacuated Agents

60% 60% 60%

40% 40% 40%
20% 20% 20%
0% 0% 0%
0 1200 2400 3600 0 1200 2400 3600 0 1200 2400 3600
Time (s) Time (s) Time (s)
All Visual All Stamina All Hearing

FIGURE 4 Evacuation rates for different disability types in Terminal 1.

to the proportion of nondisabled agents within the overall population during emergency evacuations. Namely, evacuees may be required to
(i.e., at a ratio of 6:1, nondisabled agents were more likely to be ran- walk long distances from the aircraft boarding gates to the exits in the
domly placed farther from an exit than disabled ones). Notwithstand- terminal. Additionally, the particular ease with which piers are length-
ing these values, the slowest agents to evacuate according to MaxET ened or added may exacerbate the situation. The effect on individuals
and MET were those with lower stamina, nonmotorized wheelchairs, with disabilities, specifically those with lower stamina, is especially
and motorized wheelchairs, supporting prior results found in Manley pronounced. Consequently, other configurations such as the linear
et al. (22). design should be considered. In this design, passenger processing
Agents exiting through Terminal 1 evacuated significantly slower leads directly to aircraft access, resulting in significantly shorter
than those exiting through Terminals 2 and 3 (see Table 1). This is walking distances, which is an important advantage during total
understandable given that two concourses feed Terminal 1, whereas evacuation scenarios such as the one presented here.
only one concourse each feeds Terminals 2 and 3. In other words, From an emergency management perspective, the results are use-
more agents were required to exit through Terminal 1 than Terminals ful because they identify the most vulnerable individuals during an
2 and 3 (2,209 mean agents compared to 987 and 2,061, respectively).
Note that Concourse B is also approximately 30% longer than Con-
courses C and D. Furthermore, agents originating in the farthest half
of Concourse B were required to negotiate an extra stairway to reach
the rest of the concourse and an exit. The effect of these building char-
acteristics is more pronounced for agents with lower stamina who
move slower or those using wheelchairs who require assistance or a
special device to traverse stairs. The mean evacuation curves in Fig-
ure 4 demonstrate the rate of evacuation for different disability types
in Terminal 1.
Observation of individual simulations revealed another side effect
of these factors. Unusually large queues of waiting agents were often
observed at the stairways in Terminal 1 (see Figure 5). Periodic
clogging lasting several seconds at a time was also observed at these
locations.
From an architectural engineering perspective, the results are use- FIGURE 5 Screen shot demonstrating queuing at stairways in
ful because they demonstrate a significant limitation of the pier design Terminal 1.
Manley, Kim, Christensen, and Chen 37

evacuation. In particular, evacuees with lower stamina, visual impair- The results are useful from an emergency management perspective
ments, and those using wheelchairs were at the greatest risk. The because they identify potential bottleneck areas within the airport,
results add further emphasis to recommendations made by authors namely the stairways in Terminals 1 and 3. Note that the majority of
of previous studies with similar results. Specifically, while individ- stairways are only wide enough to accommodate two evacuees abreast,
uals using wheelchairs are easily identifiable, those with other dis- whereas one wheelchair user with a stair-climbing device occupies
abilities, such as lower stamina, may not be. Such people may include almost the entire width. As a result, those behind the wheelchair are
the elderly, pregnant women, and those with chronic health condi- forced to descend at the same speed, slowing the evacuation rate fur-
tions or temporary injuries. Therefore, managers should adopt a broad ther. In this case, managers may be wise to consider routing some
approach in terms of identifying individuals with disabilities early and individuals from Concourses A, B, or D to Terminal 2, where the
often to better ensure their safety (22). extra capacity gained from the second-floor exit allows for an overall
quicker escape.
The results also reveal another limitation of the pier airport design.
Scenarios 2, 3, and 4 In all cases, agents originating in a concourse were required to move
toward the hot zone or hazard area before exiting the terminals. The
The MET for Scenario 2 at 2,038.95 s was significantly faster than the danger faced by evacuees in doing so was demonstrated during an
MET for Scenario 1 (alpha = 0.05) (see Table 2). Note that the hot incident which occurred in the airport on October 14, 1989. A Boe-
zone in Scenario 2 effectively disabled all of the stairways in Termi- ing 727 aircraft parked at a gate near the intersection of the con-
nal 1 as the EDS device containing the bomb was located in front of course and terminal caught fire, forcing the passengers and crew to
the ticketing counter immediately adjacent to them. As a result, agents evacuate (29). The rest of the passengers waiting in the concourse
originating on the second floor in Concourses A and B were forced to shortly followed, some of them walking through the smoke from the
find exits in Terminal 2 instead of Terminal 1. In this case, even with fire as they exited toward the main terminal (D. Korzep, personal
twice the average number of agents using Terminal 2, it was still communication, April 8, 2010).
quicker to leave the airport through that location. This finding is attrib-
uted to the extra capacity provided by a wide exit located on the sec-
ond floor of Terminal 2. Agents choosing this exit did not have to wait CONCLUSION AND FUTURE RESEARCH
for stairway access to leave the building.
In contrast, the MET for Scenario 4 at 2,262.94 s was significantly The ABM presented in this study has been shown to simulate emer-
slower than the MET for Scenario 1 (alpha = 0.05). The bomb’s hot gency evacuations of heterogeneous populations in a realistic manner.
zone in Scenario 4 only disabled three stairwells nearest the ticketing The results of a simulation experiment with a densely populated air-
counters in Terminal 3, leaving three accessible stairways in the oppo- port demonstrated behavior consistent with expectations regarding
site corner of the building. Thus, agents originating in Concourse D individuals with and without disabilities. Additionally, several find-
were still able to find an egress from Terminal 3 instead of another ings which inform both architectural engineering and emergency
terminal. However, the reduction in stairway capacity signifi- management practice were revealed, including (a) limitations of
cantly lengthened the time required to exit the building (1,628.23 s the pier airport design during emergency evacuations, (b) identi-
in Scenario 4 compared to 1,120.61, 1,132.52, and 1,071.49 s in fication of the individuals most at risk or those with lower stam-
Scenarios 1, 2, and 3, respectively). ina, and (c) potential bottleneck areas at the stairways located
The MET for Scenario 3 was not significantly different from Sce- in Terminals 1 and 3.
nario 1. In this case, the hot zone prevented agents from using some It is important to acknowledge the primary limitation arising from
of the exits on the first floor of Terminal 2 but did not affect the stair- the exclusion of the fifth concourse connected to Terminal 3 on the
ways leading to the second floor of the building. In this terminal, bag- airport map. Note, however, that its inclusion would result in a
gage is checked at the ticketing counters and routed via a conveyor configuration very similar to that of Terminal 1. The expectation
belt to an automated EDS facility on the far side of the main building. is that doing so would lend further credence to the findings already
Consequently, agents used the exits located near the baggage claim presented. Another limitation arises from demographic differences
carousels on the opposite side of the terminal without affecting the between U.S. and airport populations affecting the distribution of
overall evacuation time. individuals with disabilities. For example, airports such as Miami,
Florida; Palm Beach, Florida; and Seattle–Tacoma, Washington, are
recognized as having older populations, resulting in higher pro-
portions of individuals with lower stamina. Yet another limitation
TABLE 2 MET (seconds) and Mean Number of Agents by
Terminal for Scenarios 1, 2, 3, and 4
arises from the speed and mobility restrictions imposed by floor
covering on individuals with disabilities. For example, people
Terminal Scenario 1 Scenario 2 Scenario 3 Scenario 4 using wheelchairs travel faster over hard surfaces than the carpet
often found at concourse gates and other waiting areas. Unfortu-
MET nately, data to address these issues were unavailable at the time the
All 2,169.84 2,038.95 2,168.79 2,262.94
1 2,169.08 883.82 2,168.79 2,227.59
study was conducted.
2 1,021.72 2,037.37 1,014.85 1,033.91 Last, the findings presented here suggest several avenues for future
3 1,120.61 1,132.53 1,071.49 1,628.23 research. In particular, studies comparing evacuation performance of
Agents different airport designs, such as pier, linear, and satellite, would more
All 6,000.00 6,000.00 6,000.00 6,000.00 clearly reveal the advantages and disadvantages of each during crisis
1 2,295.44 911.98 2,299.58 2,369.16 situations. Additionally, empirical studies examining the performance
2 1,643.52 3,024.32 1,638.80 1,713.54
3 2,061.00 2,063.68 2,061.56 1,917.24
of different equipment, such as stair chairs or lifts, would lend further
realism to this and other evacuation models.
 38 Transportation Research Record 2206

ACKNOWLEDGMENTS 14. Quinn, M. J., R. A. Metoyer, and K. Hunter-Zaworski. Parallel Implemen-

tation of the Social Forces Model. Proc., Second International Conference
The authors thank TRB advisory panel members, including Randy in Pedestrian and Evacuation Dynamics, London, 2003, pp. 63–74.
15. Song, W., Y. Yu, X. Xu, and H. Zhang. Evacuation Analysis of a Com-
Berg, Paul Friedman, Michael Drollinger, Katharine Hunter-Zaworski, mercial Plaza with CAFE Model. International Journal on Engineering
Keith Mew, James Briggs, and Larry Goldstein, all of whom con- Performance-Based Fire Codes, Vol. 7, No. 4, 2005, pp. 182–191.
tributed to this study. The authors are particularly grateful to air- 16. Song, Y., Y. Yu, B. Wang, and W. Fan. Evacuation Behaviors at Exit in
port management and staff for their collaboration. This work was CA Model with Force Essentials: A Comparison with Social Force
Model. Physica A, Vol. 371, 2006, pp. 658–666.
supported by a grant from the Airport Cooperative Research Pro-
17. Hoogendoorn, S. P., and P. H. L. Bovy, Gas-Kinetic Modeling and Sim-
gram (Project 11-04) through the Transportation Research Board of ulation of Pedestrian Flows. In Transportation Research Record: Jour-
the National Academies. nal of the Transportation Research Board, No. 1710, TRB, National
Research Council, Washington, D.C., 2000, pp. 28–36.
18. Corning, P. The Re-emergence of Emergence: A Venerable Concept in
Search of a Theory. Complexity, Vol. 7, No. 6, 2002, pp. 18–30.
REFERENCES 19. Bonabeau, E. Agent-Based Modeling: Methods and Techniques for
Simulating Human Systems. Proceedings of the National Academy of
1. Siegfried, J. Earthquake Leads to Evacuation of San Diego Terminal. Sciences, Vol. 99, 2009, pp. 7280–7287.
Airlines/Airport Examiner, April 4, 2010. http://www.examiner.com/ 20. Wang, J., K. Gwebu, M. Shanger, and M. Troutt. An Application of
x-18134-AirlinesAirport-Examiner∼y2010m4d5-Earthquake-closes-San- Agent-Based Simulation to Knowledge Sharing. Decision Support Sys-
Diego-Airport. tems, Vol. 46, 2009, pp. 532–541.
2. Feldman, C. Los Angeles Airport Shooting Kills 3. CNN.com, July 5, 21. Jennings, N., K. Sycara, and M. Wooldridge. A Roadmap of Agent
2002. http://archives.cnn.com/2002/US/07/04/la.airport.shooting/. Research and Development. Autonomous Agents and Multi-Agent Sys-
3. Rehabilitation Research and Training Center on Disability Demographics tems, Vol. 1, No. 1, 1998, pp. 7–28.
and Statistics. 2005 Disability Status Reports. Cornell University, Ithaca, 22. Manley, M. T., Y. S. Kim, K. Christensen, and A. Chen. An Agent-
N.Y., 2005. Based Model for Emergency Evacuation of Heterogeneous Popula-
4. Huang, C., C. Sun, and J. Hsieh. Simulating SARS: Small-World Epidemi- tions. Accepted for publication, 2010.
ological Modeling and Public Health Policy Assessments. Journal of Arti- 23. Christensen, K. M., S. D. Collins, J. M. Holt, and C. N. Phillips. The Rela-
ficial Societies and Social Simulation, Vol. 7, No. 4, 2004, pp. 100–131. tionship Between the Design of the Built Environment and the Ability to
5. Luo, Y., K. Liu, and D. Davis. A Multi-Agent Decision Support System Egress of Individuals with Disabilities. Review of Disability Studies,
for Stock Trading. IEEE Network, Vol. 16, No. 1, 2002, pp. 20–27. Vol. 2, No. 3, 2006, pp. 24–34.
24. Proulx, G., and R. F. Fahy. Toward Creating a Database on Delay Times
6. Burmeister, B., A. Haddadi, and G. Matylis. Applications of Multi-Agent
to Start Evacuation and Walking Speeds for Use in Evacuation Modeling.
Systems in Traffic and Transportation. IEE Transactions on Software
2nd International Symposium on Human Behaviour in Fire, London,
Engineering, Vol. 144, No. 1, 1997, pp. 51–60.
2001, pp. 175–173.
7. Lovas, G. G. Modeling and Simulation of Pedestrian Traffic Flow.
25. Purser, D., and M. Bensilum. Quantification of Behavior for Engineering
Transportation Research Part B, Vol. 28, 1994, pp. 429–443. Design Standards and Escape Time Calculations. Safety Science, Vol. 28,
8. Ashford, N., M. O’Leary, and P. D. McGinty. Stochastic Modeling of No. 2, 2001, pp. 157–182.
Passenger and Baggage Flows Through an Airport Terminal. Traffic 26. U.S. Census Bureau. Census 2000 Summary File 4. 2003. http://www.
Engineering and Control, Vol. 17, 1976, pp. 207–210. census.gov/prod/cen2000/doc/sf4.pdf.
9. Helbing, D., L. Buzna, A. Johanssen, and T. Werner. Self-Organized 27. Boyce, K., T. J. Shields, and G. W. Silcock. Toward the Characterization
Pedestrian Crowd Dynamics: Experiments, Simulations, and Design of Building Occupancies for Fire Safety Engineering: Prevalence, Type,
Solutions. Transportation Science, Vol. 39, No. 1, 2005, pp. 1–24. and Mobility of Disabled People. Fire Technology, Vol. 35, No. 1, 1999,
10. Henderson, L. F. On the Fluid Mechanics of Human Crowd Motion. pp. 35–50.
Transportation Research, Vol. 8, 1974, pp. 509–513. 28. Wright, M. S., G. K. Cook, and G. M. Webber. Emergency Lighting and
11. Burstedde, C., K. Klauck, A. Schadschneider, and J. Zittartz. Simulation Wayfinding Provision Systems for Visually Impaired People: Phase I of
of Pedestrian Dynamics Using a 2-Dimensional Cellular Automaton. a Study. Lighting Research Technology, Vol. 31, No. 2, 1999, pp. 35–42.
Physica A: Statistical Mechanics and Its Applications, Vol. 295, No. 3–4, 29. National Transportation Safety Board. Accident Report: DCA90MA002.
2001, pp. 507–525. 1990. http://www.ntsb.gov/NTSB/brief.asp?ev_id=20001213X29575
12. Helbing, D. A Mathematical Model for the Behavior of Pedestrians. &key=1.
Behavioral Science, Vol. 36, 1991, pp. 298–310.
13. Helbing, D., and P. Molnar. Social Force Model for Pedestrian Dynam- The ACRP Selection Panel for the Graduate Research Award Program on Public-
ics. Physical Review E, Vol. 51, 1995, pp. 4282–4286. Sector Aviation Issues peer-reviewed this paper.

View publication stats