Content uploaded by Thomas Cova
Author content
All content in this area was uploaded by Thomas Cova on Oct 22, 2017
Content may be subject to copyright.
A network flow model for lane-based evacuation routing
Thomas J. Cova
*
, Justin P. Johnson
Center for Natural and Technological Hazards, Department of Geography, University of Utah, 260 S. Central
Campus Dr., Rm. 270, Salt Lake City, UT 84112, USA
Received 22 August 2001; received in revised form 28 November 2002; accepted 8 December 2002
Abstract
Most traffic delays in regional evacuations occur at intersections. Lane-based routing is one strategy for
reducing these delays. This paper presents a network flow model for identifying optimal lane-based
evacuation routing plans in a complex road network. The model is an integer extension of the minimum-
cost flow problem. It can be used to generate routing plans that trade total vehicle travel-distance against
merging, while preventing traffic crossing-conflicts at intersections. A mixed-integer programming solver is
used to derive optimal routing plans for a sample network. Manual capacity analysis and microscopic
traffic simulation are used to compare the relative efficiency of the plans. An application is presented for
Salt Lake City, Utah.
Ó2003 Elsevier Science Ltd. All rights reserved.
Keywords: Emergency management; Evacuation routing; Network flow modeling; Optimization modeling
1. Introduction
Evacuation is a common strategy in emergency management. In many hazardous events, the
best option is to relocate threatened populations to safer areas. This is a complex problem with
many behavioral and management facets (Perry, 1985; Vogt and Sorensen, 1992; Dow and Cutter,
1998; Drabek, 1999). A zone to evacuate must be agreed upon (Sorensen et al., 1992), shelters and
exits must be designated (Sherali et al., 1991), and evacuees must be routed to safety under dy-
namic hazard and traffic conditions (MacGregor-Smith, 1991; Southworth, 1991). Any number of
transportation problems can arise during an evacuation. For example, notifying evacuees may be
difficult, traffic delays are common, and transportation lifelines are often compromised by the
hazard.
*
Corresponding author. Tel.: +1-801-581-7930; fax: +1-801-581-8219.
E-mail addresses: cova@geog.utah.edu (T.J. Cova), justin.johnson@geog.utah.edu (J.P. Johnson).
0965-8564/03/$ - see front matter Ó2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0965-8564(03)00007-7
Transportation Research Part A 37 (2003) 579–604
www.elsevier.com/locate/tra
A central challenge in managing a regional evacuation is routing people to safety. An efficient
routing plan is valuable because evacuations routinely result in travel demand that exceeds the
available network capacity. This can occur at scales from an individual neighborhood to an entire
urban region, as in the Oakland–Berkeley Fire evacuation in 1991 and the Hurricane Floyd
evacuation in 1999. Traffic delays may range from inconvenient to catastrophic. Important factors
that affect the outcome of an evacuation include the time available before (or after) the hazardÕs
impact, the expected travel demand, and the consequences of not clearing the area in a timely
manner. As transportation infrastructure is a limited resource both in terms of directional ac-
cessibility and capacity, it is worth pursuing a routing plan that makes optimal use of this in-
frastructure.
Most traffic delays during an evacuation occur at intersections (Southworth, 1991). Lane-based
routing is a valuable strategy for reducing these delays. In a lane-based routing plan, select turning
options at intersections are restricted to improve traffic flow away from a hazardous area. A plan
might require vehicles in the right lane of an intersection approach to turn right while requiring
vehicles in the left lane to continue straight. One benefit of this type of routing is that intersections
with potentially significant delays can be temporarily transformed into an uninterrupted flow
facility. This increases intersection approach capacities in directions favorable for evacuating a
defined area. Lane-based routing can also be used to reduce (or eliminate) intersection crossing
and merging conflicts. In general, restricting intersection turning-latitude inconveniences some
drivers in the name of increasing system-wide transportation network throughput.
The 2000 Cerro Grande Fire evacuation in Los Alamos provided a valuable example of lane-
based routing (LAC, 2001). Evacuees north of the main transportation artery in Los Alamos, Dia-
mond Boulevard, were instructed to enter the westbound right lane of this Boulevard and remain in
that lane. Evacuees originating south of Diamond Boulevard were instructed to enter the westbound
left lane and remain in that lane. For this reason, vehicles north of Diamond did not merge with
vehicles from the south. Furthermore, major intersections were placed in a state of uninterrupted
flow. This allowed emergency managers to evacuate more than 11,000 residents in a few hours with
only one major exit, a feat that would have taken many more hours under normal traffic control.
This paper presents a network flow model for identifying optimal lane-based evacuation
routing plans in a complex road network. We review prior work in prescriptive evacuation routing
and outline criteria for identifying an optimal lane-based routing plan. A subsequent section
presents the model formulation. A mixed-integer programming solver is used to derive routing
plans for sample networks. Manual capacity analysis and microscopic traffic simulation are used
to evaluate the efficiency of the plans. Lane-based routing plans are generated for an area in
downtown Salt Lake City, Utah. The paper concludes with a discussion of the results and areas
for further research.
2. Prescriptive evacuation routing
2.1. Prior work
There are few examples of prescriptive routing in regional evacuation modeling. Evacuation
routing research has emphasized the dynamic flow of evacuees and route-choice modeling (Sheffi
580 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
et al., 1982; Southworth, 1991; Hobeika et al., 1994; Pidd et al., 1996; Urbanik, 2000). Dunn and
Newton (1992) proposed a maximum-flow approach to evacuation routing with the objective of
moving the most flow (evacuees) from a source to a sink through a capacitated network (Ford and
Fulkerson, 1962). An estimate of the population at source nodes is required along with the
available links and their capacities. This approach is very valuable in assessing network capacity
for long-term evacuation planning, but the results can be difficult to implement in practice. Traffic
managers are limited in their ability to accurately estimate origin flow volumes during an emer-
gency, although methods to address this problem have been proposed (Glickman, 1986). It is also
difficult for emergency managers to precisely allocate flow volumes to links when drivers are
generally free to select a route to a designated destination.
Yamada (1996) used the minimum-cost flow problem to assign pedestrian evacuees to shelters
at the city scale. Yamada defined the shortest evacuation plan (SEP) as one where the total
distance from all evacuees to all shelters is minimized. This approach can also be used in a road-
network context, where each vehicle is routed to its nearest evacuation zone exit under a shortest
network-distance assumption. This can be very valuable in assessing metrics like total travel
distance for a given scenario. However, in a road-network context, assigning vehicles to their
closest exit can result in significant merging and congestion. Also, a vehicular SEP can be difficult
to implement because emergency managers must communicate closest exits and shortest paths in a
complex road network to all drivers. In cases where it is possible to communicate this information
using an intelligent transportation system (ITS), successful execution generally depends on a high
level of voluntary participation in the plan.
Recently, Campos et al. (2000) presented a k-shortest path method for identifying optimal
independent evacuation routes. In their method, an individual path is better if it has a greater
capacity and lower travel time. The optimal set of disjoint routes between an origin and desti-
nation node is defined as the one with the greatest sum of the capacity/time ratios for each route.
The application that Campos et al. refer to is identifying routes to evacuate the area surrounding a
nuclear power plant. The scale of the analysis is the same as that of Dunn and Newton (1992) in
that a node represents a relatively large traffic-generating region surrounding a nuclear power
plant.
2.2. Lane-based routing
Although it is difficult to accurately estimate source flows into a network, allocate flows to
links, and communicate best routes to evacuees in an emergency, it is relatively straightforward to
control turning latitude at intersections. Traffic managers routinely improve vehicle flow away
from a hazardous area by blocking entry into an evacuation zone and placing barriers to restrict
lanes at intersections to turning left, right, or continuing straight. This is a complex network-
based decision space. A first step toward representing it in a network optimization model is de-
fining a set of criteria for rating lane-based evacuation routing plans.
A central goal in evacuation routing is to transform critical intersections into uninterrupted
flow facilities. An uninterrupted flow facility does not require vehicles to come to a full stop. Fig.
1a depicts the conflict points in a standard four-leg single-lane approach intersection. Removing
conflict points can reduce intersection delays and total evacuation time. The same action also
reduces potential intersection accident points (Poch and Mannering, 1996; Rao and Rengaraju,
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 581
1997). Fig. 1b shows one example where 13 of the 16 crossing conflicts (81%) and 2 of the 4
merging conflicts (50%) in an intersection are eliminated by blocking traffic at two upstream
intersections. This is analogous to permanently converting the two intersecting streets to one-way
streets (ITE, 1993). Fig. 1c shows how traffic pylons (or barriers) can be set up to eliminate all
crossing conflicts and further reduce merging conflicts to 1. Fig. 1d represents the ideal case where
the pylons are configured to remove all intersection conflict points.
The prior example highlights two objectives for lane-based evacuation routing plans: mini-
mizing intersection crossing conflicts and merging conflicts. A third objective in evacuation
routing is to minimize lane changing along multi-lane arterials. If a significant percentage of
vehicles changes lanes along a link, traffic delays and evacuation time may increase. It might not
be possible to physically restrict lane changing during an evacuation due to time constraints in
setting up temporary barriers, but instructions can be included in an evacuation order to achieve
this end. An order might stipulate that, ‘‘Vehicles from the north should enter the right lane of
Boulevard atraveling west and remain in that lane.’’ Complete driver participation is not nec-
essary, as any reduction in lane changing can reduce traffic delays.
Reducing the number of crossing and merging conflicts in an evacuation routing plan can
increase the distance that vehicles must travel to reach an evacuation zone exit. Another objective
is to minimize the total evacuee travel distance (i.e. shortest network distance). This is the shortest
evacuation plan (SEP) criterion (Yamada, 1996). The SEP criterion establishes a fundamental
trade-off between total travel-distance and merging. At one extreme, evacuees will be routed to
their closest exit under a shortest network-distance assignment, but merging will be at its greatest.
At the other extreme, merging is minimized, but evacuees may be routed away from their closest
exit to achieve this end.
Fig. 1. Evacuation routing plans for a four-leg, single-lane appoach intersection with various crossing and merging
conflict levels.
582 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
Fig. 2 depicts the expected effect of this trade-off on evacuation times under various traffic
volume scenarios. Under heavy volumes, the decrease in network clearing time afforded by re-
ducing the number of intersection crossing and merging conflicts should be greatest. Under
moderate traffic volumes, the decrease would be less because the inherent delays in having traffic
cross and merge is less. In scenarios with light traffic volumes, the trend in evacuation time re-
verses, and it becomes quicker to send vehicles to their closest exit (shortest network distance)
because there is little to no delay associated with having traffic cross and merge. The reduction in
network clearing time afforded by removing merging is likely to be a function of the efficiency
within which merging can be conducted. If this process is very time consuming (e.g. four-way
stop), then removing intersection conflicts in a routing plan should reduce network clearing time
more than in cases where intersection conflicts are managed very efficiently (e.g. adaptive signal
control). In a subsequent section, we use manual capacity analysis and microscopic traffic sim-
ulation to demonstrate these various cases.
3. Model formulation
The routing problem described in this section is based on the minimum-cost flow problem
(Ford and Fulkerson, 1962). The min-cost flow problem holds a central position among network
optimization models because it encompasses a broad class of applications (Hillier and Lieberman,
1990). The objective is to minimize the cost of transporting all supply (source nodes) to meet all
demand (destination nodes) in a capacitated network. It is a linear programming problem that can
Fig. 2. The theoretical effect of removing intersection conflict points (crossing and merging) from an evacuation routing
plan on network clearing time.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 583
be solved optimally in a very efficient manner using the network simplex method, a streamlined
version of the simplex method (Ahuja et al., 1993; Eiselt and Sandblom, 2000).
The evacuation routing problem (ERP) is an integer extension of the min-cost flow problem.
The primary objective is to route vehicles to their closest evacuation zone exit. A secondary
objective is to minimize the number of intersection merging-conflicts. Furthermore, the model
prevents intersection crossing-conflicts. It can be specified as follows:
iindex of network nodes
i!jdirected arc from node ito node j
binet flow at node i
dij distance along i!j
uij capacity for i!j
Mupper bound on the number of merges
Decision variables:
xij vehicle flow on i!j
yij
1 if the flow on arc ij is positive
0 otherwise
zithe number of traffic streams that merge at node i
Evacuation routing problem:
Minimize:Z¼X
iX
j
dijxij ð1Þ
Subject to:X
j
xij X
j
xji ¼bifor all ið2Þ
xij 6uijyij for all i!jthat cross k!lð3Þ
xkl 6uklð1yij Þfor all k!lcrossed by i!jð4Þ
X
j
yji 6ziþ1 for all iwith a potential merge ð5Þ
X
i
zi6Mð6Þ
06xij 6uij for all i!jð7Þ
The objective (1) is to minimize total travel distance. Constraint (2) is the standard flow con-
servation constraint in a network flow problem: the flow out of a node minus the flow into the
node must be equal to the net flow at the node. Typically, biis a constant that is positive for source
nodes, zero for intermediate nodes, and negative for destination nodes. In this model, biis positive
for source nodes, zero for intermediate nodes but a variable for evacuation zone exit nodes. This
allows the net outflow at exits to be resolved endogenously along with the routes. Constraints (3)
and (4) prevent intersection crossing-conflicts. Constraint (5) records a merge at node ifor each
traffic stream above 1 that terminates at the node. The yivariables are binary, so the zi
Õs can be
relaxed linear variables, and they are guaranteed to have integer values. To prevent three-way
merging into a single lane, an optional upper bound of 1 can be placed on the zivariables.
584 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
Constraint (6) places an adjustable upper bound on the total number of merges allowed in a
routing plan. This is the secondary objective of minimizing the number of merging conflicts
represented using the constraint method for multi-objective programming (Cohon, 1978). Con-
straint (7) places bounds on all flow variables.
4. Lane-based network data modeling
Lane-based network data modeling is a finer level of geographic detail than typically relied
upon in network flow modeling. In recent years, ITSs and other transportation applications have
highlighted the need for lane-based network data models (Miller and Shaw, 2001). In the context
of evacuation routing, lane connectivity is very valuable, as abstracting an intersection into a
single node conceals important traffic flow details that might cause delays (Ziliaskopoulous and
Mahmassani, 1996). This can lead to routing ambiguity because an intersection might be in any
number of states at the lane level. Fig. 3 depicts a standard node-per-intersection network flow
data model that can be interpreted in different ways. This data model cannot distinguish the upper
case from the lower case. This is important because the potential traffic delay in the upper case is
greater under elevated travel demand, as traffic is merging into a single lane.
To represent lane connectivity, a node-per-lane model can be used. Fig. 4 depicts a standard
four-leg, single-lane approach intersection with one lane in each direction (Garber and Hoel,
1997). There are eight nodes that represent intersection entry and exit points and 12 possible
turns. Thus, 12 linear arc variables ðxijÞare required per intersection using this data model. There
are 16 crossing conflict-points, and four merging conflict points. A binary crossing matrix in Fig. 4
depicts the arcs that cross. Because arc acrossing bis the same as bcrossing a, only the 1Õs above
the diagonal are shown. Six matrix rows contain a 1, so six binary variables ðyij Þand six con-
straints would be required to represent constraint set (3). There are sixteen 1Õs in the matrix,
leading to 16 constraints per intersection for constraint set (4). Capturing merges requires binary
variables for each arc that might merge with other arcs. This leads to 12 binary variables ðyijÞ, four
linear merge variables ðziÞ, and four constraints for constraint (5). Although six binary variables
can prevent traffic stream crossings, all 12 arcs require a binary variable to detect merges.
Network data modeling can significantly affect model solution time. A representation that
reduces the number of variables and constraints in a problem instance is preferable. Fig. 5 depicts
more compact representations for a four-leg single-lane approach and double-lane approach in-
tersection. This model eliminates right-hand turn arcs, which reduces the number of linear flow
Fig. 3. A node-per-intersection network flow data model cannot represent the difference between a lane merge and
non-merge.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 585
variables per intersection by 4. The number of binary variables remains 12 because variables must
be added to the four approaching arcs in each intersection to detect merges with a right-hand turn.
The number of crossing conflicts remains 16, and the number of merging variables ðziÞand
constraints both remain 4. There are three potential crossing conflicts that might occur at each
node. For example, at node 1 in Fig. 5, arcs 1 and 2 cross, 1 and 3 cross, and 2 and 4 cross.
Overall, the reduction in variables and constraints afforded by this representation warrant its use.
There are many other intersection and link types in real road networks (TRB, 1992), but they can
be represented with variations of the data models shown.
An additional issue that arises in lane-based data modeling is exclusive turning lanes. Although
these lanes are beneficial in normal traffic operations, they do not add any turning latitude in an
evacuation routing context. In other words, given that intersections are to be placed in state of
uninterrupted flow, exclusive left and right-hand turning lanes do not add any turns that would
not otherwise be possible without them. For this reason, it is not necessary to add additional
logical nodes and arcs to represent these lanes. This substantially reduces data-model complexity
and the number of variables necessary to represent the model in equations (1)–(7).
Single-lane data models may suffice in neighborhoods and rural areas, but in many urban
ERPs, a multi-lane model is required. There are a number of issues that arise in a multi-lane
evacuation routing context. Central to these is how additional lanes along a link should be uti-
Fig. 4. A four-leg, single-lane approach data model and crossing matrix.
586 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
lized. For example, there are three ways to enter a left (or center) lane along a link in a routing
scheme. In the first case, the vehicle might enter from a driveway or other intermediate access
point along a link. This can be modeled as a direct source (origin) flow into the left lane. The
second entry method is to change lanes from the right lane into the left lane. If lane changing is
allowed along a link, it can be represented with two additional arcs that connect the lanes. The
third entry method is to make a left-hand turn at an intersection into a left lane from an inter-
secting leg. A standard left-hand turn arc can handle this option. Fig. 6 depicts these three cases
along with the addition case of a reversible lane. Note that explicit lane-changing arcs introduce a
potential crossing and two potential merging points into a model (Fig. 6b). Given that the model
in (1)–(7) minimizes merging, lane changing in a routing plan will naturally be avoided if this
option is used. In the reversible lane case, Fig. 6d, additional binary variables must be added to
the model because only one direction is possible in a routing scheme.
Any turn that a decision maker deems possible in a real evacuation should be represented in a
lane-based network data model. This is a larger set than allowed under normal network opera-
tions. For example, in an emergency, a right hand lane in a four-leg double-lane approach
Fig. 5. Compact data models for four-leg single-lane approach and double-lane approach intersections.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 587
intersection might be routed using a left hand turn into the right lane of the departing leg. The
additional variables required to represent this class of turn make a problem instance more difficult
to solve optimally. For this reason, the recommended approach is to add additional turns as
needed. If an acceptable routing plan can be identified with standard traffic rules, this is the
preferred option. Additional intersection turns and lane changing can be added to the data model
in stages until a suitable routing plan emerges.
5. Computational experiments
5.1. Solving the model
This section demonstrates the modelÕs output and solution time with controlled computational
experiments on simple data sets. CPLEX 7.0 was used to derive optimal solutions to the model
(ILOG, 2000). Fig. 7 depicts a nine-intersection evacuation zone of two-way streets represented
using the compact four-approach intersection data model. The network has 60 nodes, 120 arcs,
144 potential crossing conflicts, and 24 nodes at which merging might occur. Eliminating
boundary arcs and nodes entering the sample data set can reduce the number of variables and
constraints in a problem instance. These arcs will never have positive flow if emergency personnel
block all entrances to the evacuation zone. The 1Õs in intermediate nodes along links represent
source flow (bi¼1 in Eq. (2)). This can be interpreted as the, ‘‘presence of source flow in lane i.’’
The four regions bounded by intersections might be city blocks, urban areas, or rural areas of any
shape.
Fig. 6. Data modeling methods for utilizing higher-order lanes in a network.
588 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
All arcs in the sample network have unit travel distance (dij ¼1 in Eq. (1)) and unlimited
capacity (uij ¼100 in Eq. (7)). Thus, traveling between intersections has cost 2, left turns and
traveling straight through an intersection have cost 1, and right-hand turns have no cost. There
are 12 exits from the network that may be selectively blocked by a hazard or emergency personnel.
For example, closing the four southwest exits results in directional egress to the northeast, and
closing all but the five northern most exits results in directional egress to the north.
Fig. 8 depicts routing solutions for varying degrees of merging and exit availability. There are
no intersection crossing-conflicts in any plan. The routing plan in the upper left-hand corner
corresponds to the shortest-distance plan for the 12-exit case (eight merges). Given eight merges,
the minimum travel distance to route vehicles out of the evacuation zone is 48. Because of the
regularity in the sample data set, multiple optima exist for all cases in the scenario matrix. For the
upper right-hand case (no merges), 16 optima were identified. Additional criteria can be used to
distinguish these solutions from one another. For example, minimizing the number of left hand
turns is beneficial in a routing plan because right hand turns are easier for emergency personnel to
direct. An additional constraint was added to control the maximum number of left-hand turns
allowed in a routing plan using an upper bound L. The following constraint was added to the
problem in (1)–(7):
X
iX
j
yij 6Lif arc i!jis a left turn ð8Þ
Of the 16 optima identified for the upper right-hand case using the initial criteria, only one had 4
left hand turns, the minimum. All other routing plans shown in Fig. 8 represent the solution with
Fig. 7. A sample network with nine intersections and 12 exits.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 589
the least number of left-hand turns selected from the set of optima using only the distance and
merging criteria. For example, the solution in the center is the eight-exit case, where up to four
merges and five left-hand turns were allowed. The minimum travel distance to route all evacuees
out of the area was 96 (D¼96, M¼4, L¼5). Reducing the number of left hand turns in this
problem instance to 4 would result in a greater total travel-distance.
The trade-off between merging and travel distance in Fig. 8 can be observed across each row.
For example, in the case where the hazard blocks all but the northern most exits (bottom row), if
seven merges are allowed and eight left hand turns, the travel distance is 126. As the number of
allowable merges is decreased to 4 and then 2, the travel distance increases to 139 and 153 because
vehicles have to travel further from their closest exit to avoid merging. The plans in the right hand
column represent the routing plans with minimal merging. In the 8-exit case (middle row), there
was no solution with less than one merge, and in the five-exit case there was no solution with less
than two merges. The plans in the left hand column represent the SEP. In this column, each
vehicle travels directly to the nearest exit (given no crossing conflicts), but merging is at its
Fig. 8. Lane-based routing plans for various combinations of distance, merging, and left turns (D, M, L) for three levels
of exit availability.
590 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
greatest. The plans in the middle column represent a compromise between merging and travel
distance. These trade-offs are quantified in Fig. 9 for the given exit scenarios.
Table 1 presents solution times for a 16-intersection (4 4) and 25-intersection (5 5) network
using the same data model given in Fig. 7. The solution times are depicted as a function of the
maximum number of merges and the number of available exits. In each case, the available exits
were a contiguous group on the fringe of the sample network.
Solution times tend to decrease as the number of exits is decreased and egress is reduced, but
this is not universal. This can be seen in the column means at the bottom of Table 1a and b.
Furthermore, solution times tend to increase and then decrease as the number of allowable merges
is increased from the minimum. This can be seen in the row means. In general, the computational
burden of identifying solutions with few merges or many merges is lower than that of identifying
solutions with a moderate amount of merges, but this is also not universal.
5.2. Assessing network clearing time
5.2.1. Manual capacity analysis
Manual capacity analysis was used to compare the relative efficiency of the various evacuation
routing plans in the nine-intersection network. The performance measure that we selected to
represent efficiency is network clearing time. This measure places greater importance on trans-
portation network performance than human behavior, but this is our focus. To simplify the
analysis, we assumed that network clearing time could be estimated using one side of the sample
network because the major routing feature in each plan is repeated on each side. In other words,
all four sides should take approximately the same time to clear.
On a given side of each routing plan, there is a principle intersection that determines how long it
will take to clear the network. Each of these intersections has a critical volume-to-capacity ratio
Fig. 9. The trade-off between distance and merging for three levels of exit availability.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 591
ðv=cÞcthat represents the approach with the greatest demand to capacity. If the intersection is
signalized, then the capacity of each approach is given by (TRB, 1992):
ci¼siðgi=CÞ ð9Þ
where si, is the saturation flow capacity of approach i,giis the effective green time for approach i,
and Cis the cycle length. In practice, many factors enter in to the calculation of the saturation
flow capacity, but our analysis is theoretical, so we will forgo discussing these factors to simplify
the comparison.
Fig. 10 depicts the major routing feature of the three plans in black. When the critical v=cratio
of the principle intersection exceeds 1, it can be used to estimate the time that it will take to clear
the area for a given plan. If the critical v=cratio is less than or equal to 1, there is no significant
delay, and network clearing time is slightly more than the time horizon twithin which vehicles
enter the network. When the ratio exceeds 1, it can be multiplied by the time horizon tto estimate
network clearing time T:
T¼ðv=cÞctð10Þ
For example, assume that each single-lane approach in the study area has a saturation flow rate si
of 1800 vph regardless of the movement. In the case of the minimum-merging plan, there are no
Table 1
The 16 intersection (a) and 25 intersection (b) solution time tests (in seconds)
Merges Number of exits Mean time
16 10 4
Time Iterations Time Iterations Time Iterations
(a) 16-intersection sample network (4 4)
1 3 1215 Infeas. Infeas. Infeas. Infeas. 3
6 10 12,459 33 40,302 8 6359 17
11 59 85,866 13 14,301 2 1464 25
16 5 5762 6 5360 3 2679 5
21 1 373 2 995 1 522 1
Mean 16 21,135 14 15,240 4 2756
(b) 25 intersection sample network (5 5)a
20124
4 199 191,544 Infeas. Infeas. Infeas. Infeas. 199
10 400 395,974 149 131,643 Infeas. Infeas. 275
16 94 88,106 33 22,899 67 62,638 65
22 14 9904 10 4799 16 11,622 13
28 161 152,988 2 1063 9 5379 57
34 4 809 4 1590 2 1135 3
40 1 306 1 596 1 889 1
Mean 125 119,947 33 27,098 19 16,333
Platform: Sun Enterprise 3500 (366 MHz) with 1024MB of RAM running Solaris 2.6.
a
MIP termination gap of 2% for 25 intersection problems.
592 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
merges, so vehicles entering from four block faces have a capacity of 1800 vph in the critical
approach because there is no need to interrupt the flow (i.e. protected left-hand turn out of the
area). If vehicles enter the network at a rate of 600 vph from each of the four block faces for 15
min, then the estimated network clearing time for the routing plan would be (row 1 in Table 2):
ð600 vph 4Þ=1800 vphÞ15 min ¼1:33 15 ¼20 min ð11Þ
In the minimum distance and balanced plans, there are varying degrees of merging. For this
reason, the capacity of the intersections depends on the type of intersection control. For example,
Fig. 10. The major routing feature in each of three evacuation routing plans (in black) and example intersection ca-
pacities for the principle intersection under demand-insensitive and demand-sensitive signal timing.
Table 2
Network clearing time calculations using manual capacity analysis
Routing plan Vph per
origin
Origin
zones
Signal timing Capacity
(vph)
v=cTime
horizon
Clearing
time (min)
Minimum merging 600 4 None 1800 1.33 15 20.0
Balanced 600 3 Equal 900 2.00 15 30.0
Minimum distance 600 2 Equal 600 2.00 15 30.0
Balanced 600 3 Proportional 1350 1.33 15 20.0
Minimum distance 600 2 Proportional 900 1.33 15 20.0
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 593
if green time is divided equally between two protected turns, each approach will have a capacity of
approximately 900 vph because they depart into a lane with a capacity of 1800 vph (Fig. 10). In
the balanced plan, the critical v=cratio occurs on the approach that has vehicle flow arriving from
three block faces. The network clearing time would then be 30 min, as shown in row 2 of Table 2.
For the shortest distance plan, where two block faces rely on the critical approach and three turns
must be merged (600 vph each), the network clearing time is also 30 min, as depicted in row 3 of
Table 2. In this simple comparison with relatively inefficient intersection control, the minimum-
merging plan has the lowest network clearing time of 20 min. The shortest distance plan and
balanced plan take 50% longer at 30 min.
The relative efficiency of the plans is sensitive to the capacity of the principle intersection. For
example, if intersection capacity is improved so that the green time is proportional to the arriving
demand on each approach (i.e. demand sensitive), then the capacity of the critical v=cratio in the
balanced plan would be 0:75 1800 vph ¼1350 vph because three of four block faces use the
critical approach (Fig. 10). Similarly, the capacity for the critical approach in the shortest-distant
plan under proportional timing would be 0:51800 vph ¼900 vph because two of four origin
zones rely on the critical approach. Using the same demand as in the prior examples, the network
clearing time for the balanced and shortest-distance plans under proportional signal timing is
20 min. This is depicted in rows 4 and 5 of Table 2.
Therefore, when (protected) merging is controlled in a relatively efficient manner, the three
plans have the same network clearing time: 20 min. This means that the benefit of minimizing
merging in a routing plan depends upon the efficiency within which merging can be performed. If
intersection control is very inefficient at merging traffic streams, as in the case of four-way stops or
flashing-red signals from a power failure, then a routing scheme that minimizes merging will likely
result in a lower total network clearing time. However, if the signal timing is highly demand-
sensitive, as in a fully actuated signal, then the benefit of avoiding merging is likely to be negli-
gible.
5.2.2. Microscopic traffic simulation
Microscopic traffic simulation (microsimulation) was also used to compare the efficiency of the
sample routing plans generated in the prior section. Fig. 11 depicts the nine-intersection micro-
simulation study area with 24 origin zones and 12 destination zones. Our intent was to compare
network clearing times for the plans that the routing model generates under various demand and
intersection-capacity scenarios. Microsimulation allowed us to compare the routing plans to the
no routing plan (null case) in addition to increasing our confidence in the simple manual capacity
calculations presented in the prior section. Paramicseby Quadstoneewas used to run the ex-
periments (Cameron and Duncan, 1996).
Seven evacuation routing cases were compared. The first, the no-routing case, assumed that
flows were not channeled at intersections. Drivers departing from each of the 24 origin zones
randomly selected one of the 12 destination zones. This is a near worst-case destination-choice
behavior on the part of evacuees that results in many crossing and merging conflicts. This case was
included as a theoretical baseline for assessing the benefit of implementing any routing plan with
flow channeling at intersections. The no-routing case was divided into signalized (equal timing)
and uncontrolled intersections resulting in two no-routing scenarios. The three lane-based routing
594 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
plans evaluated were the minimum-distance, balanced, and minimum-merging plans depicted in
the top row of Fig. 8.
For the minimum-distance and balanced routing plans, intersection control was introduced to
safely control merging. For these cases, we developed demand-insensitive and demand-sensitive
intersection signal timings. In the demand-insensitive case, the green time was equal (and fixed)
for each approach regardless of the demand or routing plan. This represents a relatively inefficient
(low capacity) intersection. In the demand-sensitive signal-timing case, the green time for each
approach was proportional to the vehicle demand on that approach for a given plan. This is
representative of a fully-actuated signal or very efficient manual traffic-direction. In both signal-
timing cases, all turns were protected (i.e. no left-hand turns on green or right-hand turns on red),
so green time rotated through the various approaches. No green time was allocated to an ap-
proach entering the evacuation zone. This generated four more evacuation-routing scenarios. The
last case evaluated was the minimum-merging plan with no crossing or merging conflicts. The
chief capacity constraint in this plan was the saturation flow capacity of a left-hand turn out of
the evacuation zone.
For experimental simplicity, the departure rate for each origin zone was uniform. This is un-
realistic, but we assumed that the departure rate would not affect the relative network clearing
times of the routing cases. The time horizon for demand input into the network was 15 min. So,
for the case where the volume is 120 vph per origin zone, the simulation model uniformly input
30 vehicles per origin zone into the network over 15 min. Each simulation scenario was run 10
times and the mean network clearing time was calculated. Network clearing time was defined as
Fig. 11. The network for the microsimulation experiments.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 595
the point when the last vehicle cleared the network, and vehicles departing from an origin zone
were restricted from crossing street centerlines in all cases.
Table 3 depicts the result of the simulation experiments. The columns in the table are the
routing plan, average vehicle distance (AVD) in block faces, existence of crossing and merging
conflicts, signal timing, mean network clearing times (in minutes) as a function of travel demand
per origin zone, and the percent decrease from the prior plan (600 vph case). For the no-routing
case with uncontrolled intersections, gridlock occurred at input volumes greater than 240 vph per
origin zone. Network clearing time is not reported for these cases. The rows are sorted in de-
scending order by network clearing time for the 600 vph per origin zone case.
The most significant decrease in network clearing time occurred in shifting from the no routing
case with equal signal timing to a routing plan that prevents traffic crossing conflicts (i.e. 40%
decrease for the 600 vph per origin zone case). The next significant decrease occurred in shifting to
either a demand-sensitive signal timing strategy or the minimal merging plan (i.e. additional 32%
decrease in 600 vph case). These results confirm the manual capacity analysis results. Namely,
minimizing merging-conflicts has approximately the same network clearing time as a shortest-
distance or balanced plan when intersection control is relatively efficient at merging traffic. As in
the manual capacity analysis, if the intersections in the evacuation zone are very efficient in
conducting merges, then the benefit of removing merging conflicts becomes negligible. However, if
the critical intersections in the evacuation are four-way stops, or they have been reduced to four-
way stops because the signals are not working due to a power failure, then a minimal merging plan
may hold the most promise for reducing network clearing times. In any case, channeling flows to
remove crossing conflicts appears to be beneficial regardless of the level of merging.
To verify that the manual capacity estimates provide similar results to the microsimulation
experiments, we calculated the time that a scenario in Table 3 should take for the 600 vph case
using the manual capacity formula. The saturation flow capacity of a lane in Paramicseis ap-
proximately 1620 vph. For example, in the minimum-distance plan with equal signal timing (i.e.
demand insensitive), the estimate would be:
ðð600 vph 2Þ=ð1620 vph=3ÞÞ 15 minÞ¼2:215 ¼33:3 min ð12Þ
This was within 30 s ð33:332:9¼0:4Þof the estimated network clearing time using micro-
simulation. The far right of Table 3 depicts similar manual capacity calculations for the 600 vph
Table 3
Mean network clearing time for the sample network using microsimulation
Routing plan AVD Cross Merge Signal Vph per origin zone % De-
crease
Manual capacity
120 240 360 480 600 600 600 Difference
None Yes Yes None 16.0 21.6
None Yes Yes Equal 16.8 23.2 35.0 47.1 54.9
Minimum distance 0.75 No Yes Equal 15.8 15.9 20.1 26.6 32.9 40.1 33.3 0.4
Balanced 1.25 No Yes Equal 15.5 15.6 20.2 26.3 32.8 0.3 33.3 0.5
Minimum distance 0.75 No Yes Proportional 15.8 16.1 16.3 18.2 22.3 32.0 22.2 0.1
Minimum merging 2.00 No No None 15.6 15.7 15.7 18.1 22.3 0.3 22.2 0.1
Balanced 1.25 No Yes Proportional 15.5 15.9 16.1 17.8 21.9 1.8 22.2 0.3
596 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
per origin zone case. The greatest difference (diff.) between the mean network clearing time using
microsimulation and the manual capacity calculation was 30 s (0.5 min).
Although this was a very simple experiment, it supports the theory that reducing the number of
merging conflicts in a routing plan reduces network clearing time when traffic volumes are high
and intersections capacities are relatively low. We did not find the minimum-distance plan to be
the quickest alternative under light to moderate traffic volumes, as predicted in Fig. 2. Under light
traffic volumes, the difference between the plans was insignificant. In general, if traffic managers
do not foresee merging at intersections as causing significant delays, then a shortest-distance plan
is the most straightforward option. If intersection delays due to merging are expected to be sig-
nificant, then the opportunity arises to minimize merges as a means for reducing these delays.
However, more tests need to be performed to verify the cases where this holds, particularly in
more complex networks.
6. Lane-based evacuation routing in Salt Lake City, Utah
From 1995 to 2000, four notable evacuations occurred in downtown Salt Lake City. Two were
due to armed assailants, one was the result of a tornado, and one was a bomb threat at the State
of Utah Office of the Regents that was later determined to be a diversion for a bank robbery
(Hoard, 2001). All of the evacuations were conducted in the same manner. An emergency plan-
ning zone containing the incident was blocked to entering traffic by police and designated a pe-
destrian evacuation zone (PEZ). People in this zone were instructed to proceed out of the zone on
foot, abandoning all vehicles. Traffic surrounding this internal zone was instructed to leave the
area of its own accord. This represents the surrounding vehicle evacuation zone (VEZ). In all four
evacuations, traffic gridlock occurred on the fringe of the PEZ, as drivers were unable to use the
sub-network blocked by police to leave the area. An approach to alleviating the gridlock sur-
rounding the PEZ is to develop a lane-based evacuation routing plan for the surrounding in-
tersections. This would require a team of emergency personnel to direct traffic, but it holds the
potential to reduce intersection delays and network clearing time. The model described in this
paper can be applied to the problem of routing vehicles out of this external VEZ.
The study area is a 20-intersection area of downtown Salt Lake City where the four evacuations
occurred. This area includes prominent office buildings, hotels, an events center, a conference
center, tourist attractions, a shopping mall, light rail stations, restaurants, and numerous parking
lots above and below ground. The area is capable of generating many thousands of vehicle trips in
a daytime evacuation. It is also clear from prior evacuations that the v=cratios for all intersection
approaches are much greater than 1. This is the basis for applying the routing model presented in
Section 3. Furthermore, it is common for power failures to render signals unusable in a disaster.
For this reason, removing merging conflicts in a routing plan is also likely to be beneficial because
there is no efficient manner in which to conduct merges under this scenario.
A first step in applying the routing model is developing a digital representation of the area of
interest. Digital lane-level information is scarce. Most network information is a street-centerline
representation with the number of lanes represented as an attribute of each arc in the database
(Miller and Shaw, 2001). One method for acquiring lane connectivity is to use high-resolution
imagery. An air-photo mosaic was commissioned for the Salt Lake City vicinity at 1-foot
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 597
resolution to plan for the Winter Olympics in 2002. This information was used and later field-
checked to generate a lane-based network model.
Fig. 12 is a schematic of the data model. This representation exaggerates the scale of the in-
tersections to reveal lane-turning connectivity. The area is comprised of 20 intersections joined by
48 lane groups (i.e. adjacent lanes between two intersections with the same direction of travel) and
111 individual lanes. The node-arc representation of the area has 314 nodes, 443 directed arcs, and
607 crossing conflicts. Thirty-five nodes on the fringe of the network represent evacuation zone
exits. Each lane is represented using two directed arcs with a node at the lane midpoint. The
midpoint node is assigned a bivalue of 1 in constraint (2). This represents the presence of origin
flow in the lane. Volume inputs are not represented because we know that the v=cratio for all
intersection approaches will be well above 1. Also, as stated, our goal is to develop a routing
model that can be applied without this information because emergency managers do not have
time to collect it in an emergency.
Lane capacities in the study area were assumed to be unlimited, so the node-arc model has the
same format as in Fig. 7. Drivers entering the network from parking lots, alleys, and other entry
points along a link are free to seek any starting lane. However, once a driver selects a lane,
emergency personnel at each intersection determine the vehicleÕs route. Given that the lane groups
connecting all intersections are the same length in the study area, all arcs were assigned unit
length. Therefore, traveling between intersections has cost 2 (two arcs), left turns and traveling
straight through an intersection have cost 1, and right turns have no cost.
Fig. 12. A traffic-lane schematic for the 20-intersection Salt Lake City study area.
598 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
As noted, the four recent evacuations in downtown Salt Lake City involved an internal, con-
tiguous pedestrian evacuation zone. An example PEZ was established under the assumption that
an incident is in progress at the intersection of 200 West and South Temple (ÔIÕin Fig. 13). The
four links entering this intersection are considered blocked by police and unusable by vehicles.
The boundary of a PEZ is the line where emergency managers deem that the incident no longer
poses a threat to vehicles. A bomb threat is a good example of such a hazard. Fig. 13 depicts a
routing plan given the defined 1-block PEZ. This plan limits the number of merges to 2, the
minimum, and the number of left turns to 11. For this reason, some vehicles would be routed
significantly further than their closest exit to avoid merging, left turns, and crossing conflicts.
As complex as the overall routing scheme appears, the focus should be the turns at each in-
tersection. This is where route direction occurs. The actual route that a vehicle will take to leave
the area emerges as the result of controlling lane-level turns at each intersection approach. Fig. 14
depicts the trade-off between merging and total distance for the study area given the PEZ. Re-
laxing the number of merges from two to four improves the balance between merging and travel
distance. Fig. 15 depicts a routing plan where the maximum number of merges is 4 and the
number of left-hand turns is 17. This plan also requires evacuees to travel further than their closest
exit in some cases. If this is not acceptable, then Min constraint (6) can be increased.
This raises the question of the best position on the non-inferior trade-off curve between total
travel-distance and merging. In general, if traffic congestion is expected to be severe and inter-
sections are very inefficient at merging traffic (e.g. flashing red or four-way stop), then a plan with
minimal merging is likely to perform best. In this type of plan, some vehicles will be routed away
from their closest exit, but intersection delays will be minimized with no crossing conflicts,
minimal merging conflicts, and minimal left turns. If traffic volumes are expected to be moderate,
and the situation is less urgent, then traffic managers can opt for a plan with moderate merging. If
traffic congestion is not a factor, then a shortest-distance routing plan is the most straightforward
Fig. 13. A lane-based evacuation routing plan, given a pedestrian evacuation zone.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 599
approach. These are only guidelines, and experienced emergency managers must decide which
routing plan is best in a unique hazardous situation.
7. Discussion
Many issues arise in implementing a lane-based routing plan. In general, a traffic manager or
traffic operations center would run the model at the time of a recognized threat or event. The
Fig. 14. The trade-off between merging and distance, given a pedestrian evacuation zone.
Fig. 15. A routing plan with an improved balance between merging and total travel-distance.
600 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
routing plan would then be relayed (e.g. Palm Pilot) to emergency responders in the field who
would set up barriers (e.g. pylons, vehicles, or temporary barricades) to prevent traffic from en-
tering the evacuation zone and channel flows at critical intersections. This would require a sig-
nificant amount of coordination among emergency personnel as well as cooperation from
evacuees. If there are not enough personnel to direct traffic at every intersection at once, then a
plan can be implemented in stages. For example, stage 1 might be preventing traffic from entering
the evacuation zone and directing traffic on the fringe to turn right and exit. This pattern is
evident in many plans generated by the model (Figs. 13 and 15); traffic on the fringe is directed to
turn right out of the zone at the first opportunity. This alone could improve the situation, as
vehicles could not turn toward the center of the zone. As each intersection is placed under control
and personnel become available, they can move toward the center of the evacuation zone, setting
up flow-channeling patterns at the most important intersections until most (or all) intersections
are functioning as uninterrupted flow facilities. In many contexts, it may not be necessary to
channel flows at all intersections. The most critical points are major intersections where v=cratios
are expected to be significantly above 1.
The application in the prior section did not require accurate vehicle input and lane capacity
estimates to generate efficient evacuation routing schemes. A binary parameter was used to define
the presence of vehicle flow in each lane, and lane capacities were considered unbounded. This is
possible because the definition of ‘‘efficient’’ in the routing model is couched more in terms of
potential traffic stream interactions at intersections than in accurately minimizing the distance that
all vehicles will travel. For this reason, the model is less sensitive to these input values than the max-
flow and min-cost models (Dunn and Newton, 1992; Yamada, 1996). In special cases, like the upper
right-hand plan in Fig. 8 (M¼0, L¼4), there is only one solution regardless of the vehicle source
flow volumes. In other words, any values can be entered for source flow biconstants in (2), and the
optimal routing scheme will be the same. This is only true when merging and left hand turns are
constrained to a minimum and link capacities are not restricted. When merging and left hand turns
are allowed to increase in the name of minimizing total evacuee travel distance, the vehicle input
values will affect the locations where merging and left-hand turns occur in a routing scheme. This is
the case because the model minimizes the total distance that all evacuees will travel, and the amount
and locations of merging affect this distance. Therefore, the detail and accuracy of the population
input estimates increases in relevance as merging and left-hand turns are allowed to increase.
Our focus in this paper is short-term, on-the-fly evacuation route planning when vehicle inputs
are not available. In longer-range planning, a simple improvement in these estimates would be to
examine the block faces (or other geographic input area) in the study area to determine which
have large parking lots and which might result in no entering traffic. The bisource flow values
could be adjusted to model vehicle flow in a relative fashion like, ‘‘Block face awill likely result in
three times the vehicle input of block face b.’’ If block face aand bboth have three lanes, each lane
in a could be assigned a source flow value of 3, and each lane in block face bcould be assigned a
source flow value of 1. Further refinement of these values could be achieved with a detailed
parking study. Similarly, lane capacity estimates could be improved if lanes in the area of interest
vary substantially in this regard. The main point is that accurate vehicle trip and lane capacity
inputs for the proposed routing model, though valuable, are not required to generate potentially
useful evacuation routing schemes. This is beneficial because these data are problematic to collect
in an emergency.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 601
8. Conclusion
This paper presented a network flow model for identifying optimal lane-based evacuation
routing plans in a complex road network. The potential for lane-based route modeling to reduce
traffic delays at intersections in evacuations is promising. There are many hazardous areas with
relatively poor access that do not have an evacuation plan (Church and Cova, 2000; Cova and
Johnson, 2002). These communities would likely benefit from an application of this routing model.
The central assumption underlying the model is that placing intersections in a state of uninterrupted
flow and minimizing merging will result in fewer traffic delays and lower network evacuation
clearing times under moderate to heavy volumes. This can require the cooperation of evacuees and
many emergency personnel if there are a large number of intersections to control. For this reason,
the model presented in this paper may identify routing plans with complexity beyond that which can
be implemented either from a management perspective or that of evacuee compliance.
The microsimulation results indicate that channeling flows at intersections to remove crossing-
conflicts can significantly decrease network clearing time over no routing plan (i.e. random des-
tination choice). A 40% reduction was reported, but this amount is likely to vary depending on the
road-network context and scenario. A shortest-distant routing plan has no crossing-conflicts but
may involve a significant amount of merging. The benefit of channeling flows at intersections to
remove merging depends on traffic volumes and the efficiency within which merging can be
performed. If this process is very inefficient and intersection v=cratios exceed 1 (e.g. four-way stop
or flashing red signals from a power failure), then a plan with minimal merging can further de-
crease network clearing time. An additional 32% reduction was reported, but again, this amount is
likely to vary depending on the road-network context and scenario. If merging can be conducted
in a very efficient manner, as in the case of demand-sensitive signal control, then reducing the
amount of merging in a routing plan appears to have little or no benefit. In this case, a shortest-
distance plan would be the most straightforward option.
There are a number of directions to pursue in the context of this research. Reversing the di-
rection of lanes was briefly discussed but not tested, and their role in evacuation routing is po-
tentially valuable. Heuristic algorithms to solve larger lane-based routing problems in less time are
also needed. There is no limit to the time pressure that emergency managers might be under to
route traffic out of an area in an emergency. The model in this paper might also serve as the
centerpiece for an evacuation routing decision support system. This would involve extensive
collaboration with emergency management experts to develop a system to meet their needs. Fi-
nally, although the Cerro Grande Fire evacuation provided evidence for the effectiveness of lane-
based evacuation routing, more studies on this type of routing are needed.
Acknowledgements
The authors thank two of the anonymous reviewers for their comments. Peter Martin and
Joseph Perrin of the Utah Traffic Lab. (UTL) provided valuable input regarding intersection
capacities. This research was funded by the US DOT, Research and Special Programs Admin-
istration (RSPA) as part of the National Center for Remote Sensing in Transportation Hazards
Consortium (NCRST-H) administered as Separate Transactions Agreement #DTRS56-00-T-003.
602 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
References
Ahuja, R.K., Magnanti, T.L., Orlin, J.B., 1993. Network Flows: Theory, Algorithms, and Applications. Prentice Hall,
Englewood Cliffs, NJ.
Cameron, G., Duncan, G., 1996. Paramics: parallel microscopic simulation of road traffic. Journal of Supercomputing
10, 25–53.
Campos, V.B.G., da Silva, P.A.L., Netto, P.O.B., 2000. Evacuation transportation planning: a method of identifying
optimal independent routes. In: Surcharov, L.J. (Ed.), Proceedings of Urban Transport V: Urban Transport and the
Environment for the 21st Century. WIT Press, Southampton, pp. 555–564.
Church, R.L., Cova, T.J., 2000. Mapping evacuation risk on transportation networks using a spatial optimization
model. Transportation Research C 8, 321–336.
Cohon, J.C., 1978. Multiobjective Programming and Planning. Academic Press, New York.
Cova, T.J., Johnson, J.P., 2002. Microsimulation of neighborhood evacuations in the urban-wildland interface.
Environment and Planning A 34, 2211–2229.
Dow, K., Cutter, S.L., 1998. Crying wolf: Repeat responses to hurricane evacuation orders. Coastal Management 26,
237–252.
Drabek, T.E., 1999. Understanding disaster warning responses. Social Science Journal 36, 515–523.
Dunn, C.E., Newton, D., 1992. Optimal routes in GIS and emergency planning applications. Area 24, 259–267.
Eiselt, H.A., Sandblom, C.L., 2000. Integer Programming and Network Models. Springer, Berlin.
Ford, L.R., Fulkerson, D.R., 1962. Flows in Networks. Princeton University Press, Princeton.
Garber, N.J., Hoel, L.A., 1997. Traffic and Highway Engineering. PWS Publishing Company, Boston.
Glickman, T.S., 1986. A methodology for estimating time-of-day variations in the size of a population exposed to risk.
Risk Analysis 6, 317–324.
Hillier, F.S., Lieberman, G.J., 1990. Introduction to Operations Research. McGraw-Hill, Inc., New York.
Hoard, C., 2001. Personal communication. Utah Division of Comprehensive Emergency Management, Salt Lake City,
Utah.
Hobeika, A.G., Kim, S., Beckwith, R.E., 1994. A decision support system for developing evacuation plans around
nuclear power stations. Interfaces 24, 22–35.
ILOG, 2000. CPLEX 7.0 userÕs manual, ILOG Corporation, Reno.
ITE, Institute for Transportation Engineers, 1993. The traffic safety toolbox. Institute for Transportation Engineers,
Washington, DC.
LAC, Los Alamos County, 2001. Los Alamos County Evacuation Plan, unpublished manuscript, April 19th.
MacGregor-Smith, J., 1991. State-dependent queuing models in emergency evacuation networks. Transportation
Research B 25, 373–389.
Miller, H.J., Shaw, S.L., 2001. Geographic Information Systems for Transportation: Principles and Applications.
Oxford University Press, Oxford.
Perry, R.W., 1985. Comprehensive Emergency Management: Evacuating Threatened Populations. Jai Press, London.
Pidd, M., de Silva, F.N., Eglese, R.W., 1996. A simulation model for emergency evacuation. European Journal of
Operations Research 90, 413–419.
Poch, M., Mannering, F., 1996. Negative binomial analysis of intersection-accident frequencies. Journal of
Transportation Engineering 122, 105–113.
Rao, V.T., Rengaraju, V.R., 1997. Probabilistic model for conflicts at urban uncontrolled intersection. Journal of
Transportation Engineering 123, 81–84.
Sheffi, Y., Mahmassani, H., Powell, W.B., 1982. A transportation network evacuation model. Transportation Research
A 16, 209–218.
Sherali, H.D., Carter, T.B., Hobeika, A.G., 1991. A location-allocation model and algorithm for evacuation planning
under hurricane/flood conditions. Transportation Research B 25, 439–452.
Sorensen, J.H., Carnes, S.A., Rogers, G.O., 1992. An approach for deriving emergency planning zones for chemical
munitions emergencies. Journal of Hazardous Materials 30, 223–242.
Southworth, F., 1991. Regional evacuation modeling: a state-of-the-art review. Oak Ridge National Labs, ORNL/TM-
11740.
T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604 603
TRB, Transportation Research Board, 1992. Highway capacity manual, 2nd ed., rev., National Research Council,
Washington, DC.
Urbanik, T., 2000. Evacuation time estimates for nuclear power plants. Journal of Hazardous Materials 75, 165–180.
Vogt, B.M., Sorensen, J.H., 1992. Evacuation Research: a reassessment, Oak Ridge National Laboratory, Tennessee,
ORNL/TM-11908.
Yamada, T., 1996. A network flow approach to a city emergency evacuation planning. International Journal of Systems
Science 27, 931–936.
Ziliaskopoulous, A.K., Mahmassani, H.S., 1996. A note on least time path computation considering delays and
prohibitions for intersection movements. Transportation Research B 30, 359–367.
604 T.J. Cova, J.P. Johnson / Transportation Research Part A 37 (2003) 579–604
