Dynamic Traffic Assignment Evaluation of Hurricane Evacuation ...

6 downloads 0 Views 2MB Size Report
The Houston–Galveston, Texas, region has experienced several major hurricanes in recent years. During the evacuation for Hurricane Rita in 2005, the Texas ...
Dynamic Traffic Assignment Evaluation of Hurricane Evacuation Strategies for the Houston–Galveston, Texas, Region Praprut Songchitruksa, Russell Henk, Steven Venglar, and Xiaosi Zeng have been made in the technical tools developed to aid decision makers, the events of 2005 served to highlight the important role of cooperation between jurisdictions, timely and relevant information, and consistent communications with the public (3). With populations in hurricane-prone regions projected to continue increasing, and the number of elderly Texans—a significant portion of the special-needs population—expected to triple by 2050, the challenge of mass evacuations of Texas coastal regions will only increase in complexity and magnitude over time (3; R. Henk, unpublished report for the Texas Department of Transportation, June 2007). Since the 1970s, evacuation modeling techniques have improved significantly. Many of the early models were developed to plan for other emergencies involving civil defense, such as nuclear missile attacks and nuclear power plant accidents. Among these programs were NETVAC, TEDSS, and DYNEV. When these programs are applied for hurricane evacuation purposes, the data that feed many of these programs have come from the inventory of Hurricane Evacuation Studies, initiated in the late 1980s by the Federal Emergency Management Agency (FEMA) to integrate key aspects of hurricane evacuation planning and to assist in disaster preparedness (2). Several models have been developed for or include hurricane evacuation traffic flow analysis, or both. Such tools include the Evacuation Traffic Information System and the Evacuation Travel Demand Forecasting System. Today, simulation programs are used to model weather, flooding, traffic flow, and evacuation travel behavior, among others. An effort is also under way to develop a computer-based incident management decision aid system. More recent research describes a simple, rapid method for calculating evacuation time estimates (ETEs) that is compatible with research findings about evacuees’ behavior during approaching hurricanes. The revised version of the empirically based time estimate method for large-scale evacuations uses empirical data derived from behavioral surveys and allows local managers to calculate ETEs by specifying route system, behavioral, and evacuation scope and timing parameters (4). Data used to estimate and test the models of evacuation travel demand were derived from a household survey conducted in southwest Louisiana, with information related to Hurricane Andrew (in 1992).

The Houston–Galveston, Texas, region has experienced several major hurricanes in recent years. During the evacuation for Hurricane Rita in 2005, the Texas Department of Transportation (DOT) decided to implement contraflow operations on I-45 to relieve massive evacuee congestion departing Houston to the north. The decision to implement contraflow was a difficult one because it involved multiple jurisdictions and required extensive coordination of manpower and resources from various entities. After the Hurricane Rita experience, the Texas DOT implemented a new strategy, referred to as “evaculane,” in which evacu­­ ation traffic could use the outside paved shoulder as a traveling lane when an evacuation was under way and evaculane signing beacons were activated. The objective is to increase capacity along key evacuation routes while avoiding the need for full-scale contraflow operation whenever possible. The evaculane on I-10 was successfully put into use during the Hurricane Ike evacuation in 2008. With the widening and completion of evaculanes on I-10 and US-290 as well as a partial contraflow plan for the I-45 corridor, the Texas DOT sponsored a study to develop a decision support tool to help determine whether these strategies would adequately handle the evacuation demand for various Houston–Galveston region evacuation scenarios. This paper describes the quantitative assessment of the performance of alternative evacuation strategies using a dynamic traffic assignment model, DynusT. The evaluation results indicated the evacu­lanes on I-10 and US-290 can sufficiently handle high evacuation demand on both routes without contraflow operation. In addition, a partial contra­ flow plan for I-45 was shown to provide sufficient capacity to handle high evacuation demand in lieu of full-scale contraflow operation.

The deadliest natural disaster in U.S. history was the Galveston Hurricane of 1900, which claimed more than 8,000 lives when the storm inundated the entire island of Galveston, Texas. Recent developments in warning time and evacuation procedures have significantly reduced the number of U.S. fatalities due to storm surge (1, 2). Unfortunately, as illustrated during the 2005 hurricane season and the evacuations surrounding Hurricanes Katrina and Rita, many improvements can still be made. Although many improvements P. Songchitruksa, Texas A&M Transportation Institute, and X. Zeng, Department of Civil Engineering, Texas A&M University System, 3135 TAMU, College Station, TX 77843-3135. R. Henk and S. Venglar, San Antonio Office, Texas A&M Transportation Institute, Texas A&M University System, 1100 Northwest Loop 410, Suite 400, San Antonio, TX 78213. Corresponding author: P. Songchitruksa, [email protected].

Decision Support Tools for Evacuation A hurricane evacuation is an areawide evacuation for an event that has a high probability of occurrence and for which there is sufficient lead time. Most of the emergency management agencies determine alternate evacuation routes a priori. Modeling a large-scale evacuation

Transportation Research Record: Journal of the Transportation Research Board, No. 2312, Transportation Research Board of the National Academies, Washington, D.C., 2012, pp. 108–119. DOI: 10.3141/2312-11 108

Songchitruksa, Henk, Venglar, and Zeng

109

such as a hurricane event is a complex and difficult task requiring efficient use of available roadway capacities and advanced traveler information systems as well as effective and coordinated evacuation schemes. Current challenges for accurate modeling of hurricane evacuation include the following (5): • Constructing an optimization model that properly incorporates the objectives for optimal evacuation schedules and evacuation routes (e.g., minimizes casualties, exposure, or other relevant measures) and • Determining accurate estimations of traffic conditions on the basis of traffic loadings resulting from varying evacuation time and route scenarios. There are four major components that need to be considered in the process of developing a decision support tool for evaluating hurricane evacuation strategies: • Infrastructure modification—infrastructure changes as part of hurricane evacuation strategies, such as capacity addition and ramp access control; • Information provision—procedures to model the effects of various information provision strategies and advisory compliance behavior of evacuees; • Real-time data—procedures to identify critical real-time traffic data elements and predicted hurricane forecast, followed by their incorporation into the decision-making process; and • Hurricane evacuation operations—how to represent various hurricane evacuation strategies to be tested, such as departure scheduling and deployment timing of hurricane evacuation tools. Researchers developed a decision support tool to capture qualitative and quantitative aspects of the decision-making process. This paper focuses on the quantitative component of this study, in which a mesoscopic dynamic traffic assignment model was used to predict the effects of various management strategies, thus allowing more effective management and providing better traffic information than is currently possible. A recent FHWA-sponsored study produced a guide for decision makers to develop a successful evacuation modeling analysis (6). In that report, Hardy et al. summarized several case studies using different tools to analyze and evaluate evacuation strategies (6). This section summarizes recent applications of the analysis tools and their findings from the case studies. The study location, event type, and selected tool are tabulated in Table 1.

A review of the literature indicated that mesoscopic analysis tools are the most commonly used type of tool for the analysis and evaluation of large-scale evacuations. Table 2 summarizes the applications of mesoscale analysis tools and findings for evacuation applications noted in Table 1.

Objectives and Scope Two major evacuation strategies considered by researchers during the development and refinement of a hurricane evacuation decision support tool were the “evaculane” (EL) and “contraflow” (CF) strategies. The Texas Department of Transportation (DOT) developed a special pavement marking symbol to designate shoulders for use as an additional travel lane during evacuation. These lanes are referred to as evaculanes. In a recent study by the Texas A&M Transportation Institute, researchers developed guidelines for various hurricane evacuation signs and markings, including route signs, contraflow signs, emergency shoulder lane signs, and pavement markings (15). Contraflow is a form of reversible traffic operation in which one or more travel lanes of a divided highway are used for the movement of traffic in the opposing direction. Contraflow is one of the hurricane evacuation tools that will be extensively modeled and evaluated in the current research effort. Contraflow is more practical on freeway facilities because they do not have at-grade intersections, which can interrupt flow or permit unrestricted access into the reversed segment. Several recent studies have examined the characteristics of contraflow operations. The highest flow rates measured by the South Carolina DOT during the Hurricane Floyd evacuation were between 1,500 and 1,600 vehicles per hour per lane (vphpl) (16). Traffic flows measured during the evacuations for Hurricanes Ivan and Katrina on I-55 in Louisiana were somewhat lower, at 1,230 vphpl and 820 vphpl on normal and contraflow lanes, respectively, on average, during the peak 10 h of the evacuation (17). Preparation of contraflow can take at least 6 h in addition to the time to plan and acquire equipment for traffic control. Inadequate designs at the upstream and downstream ends can further limit the effectiveness of the contraflow operation. In addition, traffic incidents and work zones on evacuation routes can affect planned operations. Therefore, these characteristics must be appropriately captured in the modeling process to realistically simulate and analyze evacuation strategies.

TABLE 1   Summary of Analysis Tools and Study Locations (7) Location

Network Size

Tool

Event Type

Washington, D.C. Houston–Galveston, Tex. Umatilla & Morrow Counties, Ore. Houston–Galveston Hampton Roads, Va. Nags Head, N.C. New Orleans, La. New Orleans–Baton Rouge, La. Daytona Beach, Fla.

124 intersections >2,000 signals 9-mi radius covering 6 cities 8 counties 192 signals 15 residential zones 11-mi segment New Orleans metro area 1,309 nodes and 3,264 links

CORSIM DYNASMART-P OREMS Cube Avenue VISSIM TransModeler CORSIM TRANSIMS DYNASMART-P

No notice No notice No notice Hurricane evacuation Hurricane evacuation Hurricane evacuation Hurricane evacuation Hurricane evacuation Planned

Reference (7) (8) (6) (9, 10) (11) (6) (11) (12) (6)

110

Transportation Research Record 2312

TABLE 2   Applications of Mesoscale Decision Support Tools for Evacuation Tool

Applications

Findings

DYNASMART-P

Evaluated contraflow and phased evacuation strategies for the Houston–Galveston area Evaluated and developed a new evacuation plan for the Daytona Speedway in Daytona Beach, Florida

OREMS

Analyzed evacuation time estimates (ETEs) for chemical plant emergency evacuation plan in Oregon Identified bottlenecks in the transport system and policies that could more effectively move evacuees during the natural disaster in the Houston–Galveston area

The analysis identified problematic locations resulting from surge of evacuation traffic. The simulation run time can be significant. For instance, the Houston–Galveston area simulation took 20 h for an evacuation period of 24 h. The model can successfully evaluate effects of lane closure, vehicle–pedestrian conflicts, additional one-ways, and added capacity. The network size is limited by computer memory. The model was a useful tool in quick analysis of different operational scenarios but the data input, error checking, and validation can be burdensome. The network size is limited to 1,999 internal nodes and 999 external nodes. A hybrid model process with aggregate zones and network was found to be too complex. Only a strategic-zone simplification was used in final model approach. Network coding and correct representation of operational features (signals) is crucial when using dynamic traffic assignment. The model was reported to handle the network successfully with 3,000 zones and 25,000 links (14). Integrated transportation analysis packages such as TransModeler can provide useful simulation capabilities for evacuation operations planning. TransModeler is a powerful tool for the before-and-after visualization and animation of traffic from a simulation. The network size is limited by computer memory.

Cube Avenue

TransModeler

Evaluated traffic operations under a hypothetical scenario of forced evacuation from Nags Head, N.C. The strategies examined were reverse lanes, shoulder lanes, and modified signal timing

Since Hurricane Rita in 2005 and Hurricane Ike in 2008, segments of I-10 and US-290 have been widened and shoulders have been paved and signed for evaculane operation to support evacuation from the Houston–Galveston region. In addition, a partial contraflow plan was developed on I-45 to circumvent known evacuation bottlenecks along that route. Dynamic traffic assignment simulation modeling using DynusT served as the means of experimenting with these evacuation strategies under various evacuation demand scenarios, with the goal of realistically identifying which strategies are required to accommodate the evacuation demand from events of varying magnitude.

Researchers selected the DynusT mesoscopic model because most of the Houston–Galveston network had been coded into DynaSmart-P, the predecessor of DynusT, for a previous research effort (19). The DynusT software, which is developed and maintained by the University of Arizona, provides an enhanced user interface but uses the same traffic assignment engine as DynaSmart-P. This allowed researchers to expedite the simulation task by leveraging existing models and maximizing the use of existing resources.

Model Development Demand and Supply Scenarios

Model Selection Simulation modeling has become an increasingly popular and effective tool for analyzing the operational performance of transportation facilities because of its ability to model complex interactions between various system elements in a realistic manner. Researchers selected a mesoscopic model for this study to account for potential changes in demand patterns (e.g., mode shift, route choice) resulting from implementation of capacity enhancement or transportation demand management strategies. This type of model generally aims at describing individual route and departure time choice adjustment in response to changes to the network-level performance through simulation. The algorithms of these models involve iterative traffic assignment procedures for travelers departing at different times to select different routes that offer minimal experienced travel time. Most of these models tend to describe a corridor- or regionwide mode or route shift at a larger geographic scope and a longer time horizon than a microscopic simulation model (18). In comparison with microscopic models, the mesoscopic models typically use a much coarser network representation and involve simplified car-following logic or macroscopic traffic flow relationships without representing detailed intervehicle interactions (18).

Researchers prepared the experimental design for examining the effects of two primary factors that can affect the network performance during hurricane evacuation—evacuation demand and hurricane evacuation strategies. Table 3 summarizes the demand scenarios used in the simulation models. The demand scenarios range from low to high levels. Because the base model was originally developed for Hurricane Rita, researchers generated demand levels for the simulation as percentages of the values observed from Hurricane Rita’s evacuation. Demand was prepared for a 24-h period beginning at 72 h before Rita’s landfall, thus bracketing the time period when peak traffic volume was observed. Two types of origin–destination (O-D) matrices considered in this study are evacuation and background O-Ds. The evacuation O-D matrix was based on a household phone survey conducted in early 2006 after Hurricane Rita (19). The background O-D matrix represents the trip activities generated by nonevacuees and was estimated on the basis of modifications to the travel demand model O-D data of the Houston–Galveston Area Council (HGAC). The background O-D matrix was included to avoid the underestimation of total traffic effects on the network. The detailed procedure on estimating and calibrating the background and evacuation demand

Songchitruksa, Henk, Venglar, and Zeng

111

TABLE 3   Demand Scenarios Demand Scenarios Low

TABLE 5   Simulation Scenarios Supply

Evacuation Demand

Background Demand

50% of Hurricane Rita’s evacuation demand

30% increase from Hurricane Rita’s background demand 15% increase from Hurricane Rita’s background demand Hurricane Rita’s background demand level (approximately 2.7 million vehicles over 24-h period)

Moderate

75% of Hurricane Rita’s evacuation demand

High

Hurricane Rita’s evacuation demand level (approximately 400,000 vehicles over 24-h period beginning at 72 h before landfall) 125% of Hurricane Rita’s evacuation demand

Super high

Demand

BC

EL

PC

CF

Low Moderate High Super high

LO_BC MD_BC HI_BC SH_BC

LO_EL MD_EL HI_EL SH_EL

LO_PC MD_PC HI_PC SH_PC

LO_CF MD_CF HI_CF SH_CF

direction. For the PC strategy, instead of reversing all I-45 southbound main lanes, the plan calls for the select contraflow of I-45 inbound links in the vicinity of Conroe, Texas. In the case of the CF strategy, all inbound lanes on I-10, I-45, and US-290 are reversed to provide additional capacity all the way to San Antonio, Dallas, and Hempstead, Texas, respectively. The shoulders on reversed main lanes are assumed to be used as ELs under the CF strategy. Table 5 summarizes the simulation scenarios analyzed. Researchers developed a total of 16 modeling scenarios, representing a full factorial experimental design with four demand and four supply scenarios. DynusT models were developed for each scenario. The outputs from the models were further used to calibrate and develop the predictive models for estimating the effects of hurricane evacuation on network performance.

25% decrease from Hurricane Rita’s background demand

was discussed elsewhere (7). It is reasonable to assume that the background demand level will be inversely correlated with the evacuation demand level; therefore, the background demand level was reduced when the evacuation demand is expected to be very high, and vice versa (Table 3). Table 4 describes the evacuation (supply) strategies considered in this study. These strategies were developed and refined on the basis of feedback from stakeholders, which included the Texas DOT, HGAC, and the Texas Department of Public Safety. All three evacuation routes have at least two main lanes in each direction. Both I-45 and US-290 have three main lanes in each direction for segments in the vicinity of the Houston area. I-10 after expansion has five main lanes plus two additional HOV lanes in each direction on segments between Houston, Texas, and Katy, Texas. Outside shoulders on I-10 and US-290 have been paved and designated as evaculanes on both inbound and outbound directions. The four evacuation strategies selected represent the most realistic evacuation options under different demand scenarios. The base case represents the existing supply conditions, that is, no additional capacity is added to the network. The full CF strategy is the scenario in which the roadway capacities are maximized by using all paved shoulders and inbound freeway links as travel lanes for (outbound) evacuation. The EL and partial CF (PC) strategies represent the intermediate supply levels between the two extremes. Under the EL strategy, all the evaculanes on US-290 and I-10 are activated for the outbound

Network Modifications To model the evacuation strategies in DynusT, analysts modified network links to represent the EL and CF operation. The base case model consists of 8,532 nodes and 24,119 links. To model the EL, researchers added an extra lane on the outbound links of I-10, US-290, or both. Because of the large number of links that had to be modified, researchers developed a Python-based link modification utility tool to assist with this task. The DynusT network model encompasses the Houston–Galveston area and all the major routes to three destination cities in Texas: Dallas, Austin, and San Antonio. Figure 1a shows the ELs along US-290 in the westbound and eastbound directions. The ELs in the eastbound direction are designed to be operated during the contraflow operation. Figure 1b shows an example of a spreadsheet-based configuration file used to update the number of lanes on each link in the network. Each link is

TABLE 4   Evacuation Strategies: Supply Scenarios Evacuation Strategies Scenario Name

Activate US-290 EL

Activate I-10 EL

Activate I-45 PC

Activate US-290 CF

Activate I-10 CF

Activate I-45 CF

Base case Evaculane Partial contraflow Full contraflow

No Yes Yes Yes

No Yes Yes Yes

No No Yes Part of I-45 CF

No No No Yes

No No No Yes

No No No Yes

Note: EL = evaculane; PC = partial contraflow; CF = full contraflow.

112

Transportation Research Record 2312

Evaculane

Evaculane for Contraflow

(a)

(b) FIGURE 1   Evaculane in DynusT: (a) evaculanes on US-290 and (b) evaculane representation in DynusT.

Songchitruksa, Henk, Venglar, and Zeng

NB2

Closed

(a)

NB2

Closed

NB1

Closed

Closed

SB1

SB2

Contraflow

SB2

113

SB1

NB1 (b)

two travel behaviors, the background traffic is modeled as Class 3, or user equilibrium, and the evacuation traffic is modeled as Class 5, or pretrip information. The simulation procedure was discussed in detail in a technical report by Henk et al. (20). One simulation run takes approximately 4 days to complete on a Windows XP 64-bit computer with dual Xeon processors and 12 GB of RAM. Simulation Results For each simulation scenario, researchers extracted the travel time for each individual vehicle and aggregated them separately by demand types (evacuation versus background), temporally (time of departure), and spatially (evacuation routes). The results from each simulation scenario were populated in a Microsoft Access database. The purpose of the simulation database was twofold. First, the database was designed to simplify the process of querying the simulation outputs for subsequent modeling and analysis. Second, the database was designed for subsequent changes and expansion of modeling scenarios. The analyst can simply update any changes in the simulation outputs as well as add results from new simulation scenarios with minimal effort.

FIGURE 2   CF coding scheme: (a) base scenario and (b) CF scenario.

Demand Summary defined by paired start and end nodes. The user specifies the number of lanes in the normal operation and the evaculane mode. The user can toggle between these two modes by running a utility tool developed specifically for this purpose. The tool will modify the number of lanes in the DynusT network on the basis of this configuration file. CF operation is represented in DynusT by coding a redundant bidirectional link for every pair of nodes along the main lanes of all three major evacuation routes (US-290, I-10, and I-45). Figure 2 shows the representation of the base versus CF cases in DynusT. Every pair of nodes has bidirectional links. As shown in Figure 2a, under the normal operation the southbound link of the northbound direction and the northbound link of the southbound direction are closed to traffic. To convert this scenario into a CF mode with all traffic going northbound, the southbound link in Figure 2b must be closed and the northbound link must be opened instead. To implement this in DynusT, a set of incidents is placed on appropriate links to replicate the closing of the links to traffic. Configuration files consisting of a set of incidents for each evacuation strategy were created to model the normal and CF operation on the three major evacuation routes. Simulation Runs Background and evacuation traffic are the two types of travelers modeled in this study. Both traffic patterns are modeled after the calibrated original-destination trip tables from the previous Hurricane Rita evacuation study (19). These two types of traffic behave differently when travel paths are chosen during evacuation. The background traffic refers to those trips made as part of daily routines, for example, commuter and grocery trips. These travelers have the knowledge of alternative routes and will seek paths that minimize their travel times. However, evacuees rely on pretrip best-path information and thus choose the best path to avoid the congestion at the time of departure. DynusT can model multiple user classes to represent different travel behavior and responses. To replicate these

Table 6 summarizes the number of vehicles generated under each demand scenario. The totals for evacuation and background vehicles are also reported separately. The level of demand under the high demand scenario represents the peak 24-h demand level estimated for the Hurricane Rita evacuation. The total number of vehicles is reduced when the evacuation demand is higher because it is assumed that the background activities will be reduced under scenarios in which high evacuation demand is anticipated. Networkwide Performance Measures Average Networkwide Travel Time The average travel times of evacuation and background vehicles are summarized by using box plots shown in Figure 3. The average evacuation travel times of all vehicles range from 3 to 5 h, depending on the demand and supply scenarios. Figure 3a shows a steeper slope pattern with the changes in the levels of demand, which implies that the demand level has a more significant influence on the average evacuation travel time than the level of supply (evacuation strategies). In addition, the mean background travel times in Figure 3b vary only slightly with the supply scenarios. The implication is that the background traffic conditions are less likely to be affected by the evacuation strategies deployed.

TABLE 6   Demand Summary Demand Scenario Low Moderate High Superhigh

Evacuation Vehicles

Background Vehicles

Total Vehicles

200,168 300,285 398,748 498,248

3,291,856 2,908,544 2,530,937 1,894,154

3,492,024 3,208,829 2,929,685 2,392,402

LO

MD

HI

42 40 38 36 34 32

32

34

36

38

40

42

Mean Background Travel Time (min)

Transportation Research Record 2312

Mean Background Travel Time (min)

114

SH

BC

Demand Scenario

EL

PC

CF

Supply Scenario

LO

MD

HI

SH

250 230 210 190

Mean Evacuee Travel Time (min)

250 230 210 190

Mean Evacuee Travel Time (min)

(a)

BC

Demand Scenario

EL

PC

CF

Supply Scenario (b)

FIGURE 3   Network travel time: (a) mean travel time of background vehicles and (b) mean travel time of evacuation vehicles.

Table 7 shows the percentages of savings in average evacuation travel time with respect to the base travel time when an alternative evacuation strategy is implemented. For example, under a moderate demand level, the average travel time for all evacuation vehicles to complete their trip is 213 min according to the current network configuration (i.e., neither EL nor CF is deployed). At the same demand level, the expected travel time savings are 3.9% for all evacuation vehicles if PC is implemented. This reduction in travel time could be significant considering that this is the aver-

TABLE 7   Average Evacuation Travel Time Savings

Demand Level Low Moderate High Superhigh

Base Case Travel Time (min) 197 213 226 253

Savings in Travel Time Versus Base Case (%) EL

PC

CF

0.5 0.3 2.0 5.0

2.9 3.9 5.4 6.7

5.9 8.0 8.8 11.7

age savings for all vehicles that evacuate during the 24-h analysis period.

Networkwide Travel Time Profile Figure 4 shows the dual peak patterns for the evacuation vehicle travel time for all demand levels regardless of the evacuation strategies deployed. The peak-and-valley patterns are more apparent at higher demand levels. The travel time profiles also reveal that peak travel times can be reduced significantly when high-capacity evacuation strategies such as PC and CF are deployed. The amount of reduction increases with the demand levels. Note that the high scenario resembles the demand level experienced during the Hurricane Rita evacuation. The average evacuation travel time at the peak was approximately 6.5 h under the base case. The corresponding figure was reduced by 30 to 60 min when the CF strategy was implemented. The base case travel time reported in Table 7 (high demand) was lower than that reported in the previous Hurricane Rita study (7, 19). In addition, these estimated figures are lower than those generally reported by Hurricane Rita’s evacuees because (a) the network conditions in the model have been upgraded to reflect the expanded US-290 and I-10 and the completion of evaculanes along I-10 and

Songchitruksa, Henk, Venglar, and Zeng

115

BC

EL

PC 0

5

CF 10

HI

15

20

SH 300

Mean Evacuation Travel Time (minutes)

250

200

150

LO

MD

300

250

200

150

0

5

10

15

20

Departure Hour

FIGURE 4   Evacuation travel time profile.

US-290 and (b) the network conditions are assumed to be incidentfree throughout the entire 24-h modeling period. Evacuation Trips Completed For the 24-h period, it was found that the total numbers of evacuation trips completed are slightly different across all the strategies considered. The small differences observed here implied that the network has sufficient capacity to handle all levels of the evacuation demand during the 24-h period. The peak-and-valley pattern observed from the evacuation travel time profiles is a result of the fact that the traffic demand does not spread out evenly over the entire period. Facility-Based Performance Measures Networkwide performance measures describe network conditions by aggregating the data from all evacuation vehicles regardless of their travel paths. Facility-based performance measures, however, focus on the travel conditions on specific evacuation routes, such as US-290, I-10, and I-45 in this study. To evaluate the effects on evacuation routes, the researchers used only the travel times collected from vehicles with travel paths on each of these three freeways. Researchers developed a Python-based tool that allows the analyst to place a spe-

cific number of probe vehicles into the network with preassigned paths at fixed intervals. This procedure ensures that there will be vehicles completing the trip using the path of interest at every time interval of interest. Because the lengths of these routes are unequal, the travel times were converted into travel speed to facilitate comparison. Speed Profiles Figure 5 displays the speed profiles during the 24-h evacuation simulation period on US-290, I-10, and I-45. Each block consists of four lines representing four supply scenarios [base case (BC), EL, high (HI), and SH (super high)]. The profiles reveal that I-45 remains the most congested evacuation route of the three. The speed drops on US-290 tend to be more severe than on I-10 but last for a shorter period of time. On the contrary, the congestion experience on I-10 appears to be moderate and more spread out over the course of the evacuation period. Speed Contour Researchers plotted speed contours for each route (US-290, I-10, and I-45) for all 16 scenarios. A total of 48 speed contours (16 scenarios × 3 routes) were prepared as part of the simulation results

116

Transportation Research Record 2312

BC

EL 5

10

15

PC

CF

20

5

10

15

US290 LO

US290 MD

US290 HI

US290 SH

I45 LO

I45 MD

I45 HI

I45 SH

20

70 60 50 40 30

Mean Speed (mph)

20

70 60 50 40 30 20

I10 LO

I10 MD

I10 HI

I10 SH

70 60 50 40 30 20 5

10

15

20

5

10

15

20

Departure Hour FIGURE 5   Speed profiles on US-290, I-10, and I-45.

database. Decision makers can consult these speed contours to identify the location, time, and extent of the congestion on each route from implementing any evacuation strategy. Figure 6 shows an example of speed contours on I-45 under the base case versus the full contraflow scenario under high demand conditions. Summary of Findings The level of congestion observed on all three routes under the lowdemand scenario was consistent with what the stakeholders had experienced during Hurricane Ike’s evacuation (the I-10 widening project was already completed at the time) and thus validates the findings from the simulation model. In addition, the examination of the simulation results across all 16 scenarios reveals the following: • I-45 is the most congested freeway of the three routes regardless of the demand–supply scenarios. This is because of a long-term construction bottleneck between the towns of Huntsville and Conroe. • US-290 is the least congested evacuation route. The speed drops on US-290 were observed to last for relatively shorter periods, although slightly more severe, compared with the other routes. • The benefit from the EL strategy on all routes is minimal under low demand conditions. The EL strategy can significantly alleviate

the congestion on US-290 and I-10 when the demand is moderate or higher. • The results also suggest that the benefit gained from implementing contraflow versus evaculane on I-10 and US-290 may be minimal. The results indicated that the EL strategy can provide sufficient capacity increase on these two routes, even under high demand conditions. • The results on I-45 suggest that the PC strategy can provide tangible improvement under moderate demand conditions or higher. The CF strategy is still the best strategy for I-45, because the CF strategy helps reduce both the scale and extent of (congestion) speed drops. However, the improvement in travel condition does not appear to be significant when compared with improvements achieved from the PC strategy. Prototype Predictive Models The results from the simulation model were used to develop predictive models for providing a quantitative assessment of Houston– Galveston traffic conditions as a result of evacuation. Researchers developed regression models for predicting the travel time and speed during evacuation under different demand scenarios and evacuation strategies. These prototype models can provide the required predic-

Songchitruksa, Henk, Venglar, and Zeng

117

(a)

(b) FIGURE 6   Speed contour example on I-45: (a) base case and high-demand scenario (BC_HI) and (b) full CF strategy and high-demand scenario (CF_HI).

118

Transportation Research Record 2312

tions for networkwide and facility-based levels. [More details about these models can be found in the literature (20).] For example, the analyst can use Equation 1 to predict hourly average speed on US-290, I-10, and I-45. The functional form of Equation 1 is a modification of the widely used Bureau of Public Roads volume–delay function. The DE/C ratio approximates the volume-to-capacity (v/c) ratio, which in this case corresponds to the ratio of evacuation demand to route capacity for a given evacuation strategy. The equation for predicting average speed on route i at hour j is of the following form: Vij =

Vf D   1 + exp  α ij + β ij ln E   C 

(1)

where Vij = average speed (mph) of all vehicles traveling on route i at hour j [route i refers to one of the three major freeways evaluated in this study, i.e., US-290, I-10, and I-45, and hour j ranges from 1 to 24 (24-h prediction period)], Vf = free-flow speed (mph) (Vf used in the calibration is 72.6 mph, which is the maximum hourly average speed observed from the simulation study), DE = expected evacuation demand during 24-h period for the Houston–Galveston region (vehicles), C = approximate ratio of expected supply capacity to base case capacity, and αij, βij = model coefficient estimates from the regression analysis of simulation results (coefficients were calibrated separately for each route i and hour j). To estimate the value of C, consider the capacity of the US-290 evacuation route as an example. Under the base case, the capacity is best approximated by the narrowest segment (a bottleneck that restricts traffic flow), which is a two-lane-wide segment. Now, consider the EL as an alternative strategy. This strategy would add one de facto travel lane to this route. Therefore, the C ratio can be estimated as 3/2 = 1.5. To illustrate, assume that the analyst wishes to predict the speed on I-45 at 3 p.m. should the PC strategy be implemented. The calibrated values of αij and βij for I-45 at 3 p.m. are −18.04 and 1.45, respectively. Therefore, the general form of the model for predicting the average travel speed on I-45 at 3 p.m. can be written as V1- 45,3 PM =

72.6 D   1 + exp  −18.04 + 1.45 ln E   1. 5 

(2)

The prototype predictive model is currently implemented in a spreadsheet-based format with Microsoft Excel. The prediction includes the evacuation travel time networkwide as well as the hourly average speed profiles on US-290, I-10, and I-45.

Concluding Remarks This paper presents a quantitative assessment of hurricane evacuation strategies for the Houston–Galveston region to assess whether the current network infrastructure and available evacuation strategies

would be able to accommodate varying levels of hurricane evacuation demand. The authors evaluated the effects of four potential evacuation strategies ranging from the existing configuration to full-scale CF with evaculanes deployed on normal and contraflow lanes. The evaluation was conducted by using the mesoscopic dynamic traffic assignment model, DynusT. The simulation results indicated that the average networkwide evacuation travel time using the existing infrastructure (expanded I-10 and US-290) will be approximately 30% lower than those experienced during the Hurricane Rita evacuation (i.e., before those facilities were improved). The results also indicated that the EL strategy on I-10 and US-290 can save up to 2% to 5% in the average evacuation travel time in the high- and very-high-demand scenarios. In addition, when the EL strategy is deployed in conjunction with partial CF on I-45, this combination can reduce the average networkwide evacuation travel time up to 5% to 7% without the need to implement the full-scale CF operation. However, this study did not account for events such as lane-blocking incidents or vehicle breakdowns during evacuation. Therefore, CF may still be needed in light of high evacuation demand and potential capacity-reducing events, which are not addressed in the current modeled scenarios. To incorporate the results into the prototype decision support tool being developed, the researchers also calibrated the models for predicting evacuation travel time and speed profiles on the evacuation routes. The current study is built on the proposed 16 modeling scenarios, which do not account for the variation in the background demand and capacity-restricted events such as lane-blocking incidents. To address the effects of those factors, recommendations are being made to expand the modeling scenarios and recalibrate the predictive models before full-scale implementation of the decision support tool for the Texas DOT.

Acknowledgments The authors thank Yi-Chang Chiu and Eric Nava of the University of Arizona for their helpful support and valuable advice concerning the DynusT model throughout the course of this study. This research was performed in cooperation with the Texas DOT.

References  1. Using Highways during Evacuation Operations for Events with Advance Notice—Routes to Effective Evacuation Planning Primer Series. FHWA, U.S. Department of Transportation, 2006.   2. Wolshon, B., E. Urbina, C. Wilmot, and M. Levitan. Review of Policies and Practices for Hurricane Evacuation. ASCE Natural Hazards Review, Aug. 2005, pp. 129–142.  3. Impacts of Climate Variability and Change on Transportation Systems and Infrastructure—Gulf Coast Study. Joint Study for U.S. Department of Transportation and U.S. Geological Survey, March 2008.   4. Lindell, M. EMBLEM 2: An Empirically Based Large Scale Evacuation Time Estimate Model. Transportation Research Part A, Vol. 42, No. 1, 2008, pp. 140–154.   5. Chiu, Y.-C. Traffic Scheduling Simulation and Assignment for AreaWide Evacuation. Presented at IEEE Intelligent Transportation Systems Conference, Washington, D.C., Oct. 3–6, 2004.   6. Hardy, M., K. Wunderlich, and J. Bunch. Structuring Modeling and Simulation Analyses for Evacuation Planning and Operations. Report DTFH61-05-D-00002. Noblis, Falls Church, Va., May 2009.   7. Chen, M., L. Chen, and E. Miller-Hooks. Traffic Signal Timing for Urban Evacuation. ASCE Journal of Urban Planning and Development—Special Emergency Transportation Issue, Vol. 133, No. 1, 2007, pp. 30–42.

Songchitruksa, Henk, Venglar, and Zeng

  8. Chiu, Y.-C., H. Zheng, J. A. Villalobos, W. Peacock, and R. Henk. Evaluating Regional Contra-Flow and Phased Evacuation Strategies for Texas Using a Large-Scale Dynamic Traffic Simulation and Assignment Approach. Journal of Homeland Security and Emergency Management, Vol. 5, No. 1, 2008, pp. 1–27.   9. Brown, C., W. L. White, J. Benson, and C. V. Slyke. Development of Strategic Hurricane Evacuation–Dynamic Traffic Assignment Model for the Houston, Texas, Region. In Transportation Research Record: Journal of the Transportation Research Board, No. 2137, Transportation Research Board of the National Academies, Washington, D.C., 2009, pp. 46–53. 10. Lam, C. P., C. V. Slyke, and H. Wang. Applying Meso-scopic Simulation to Evacuation Planning for the Houston Region. Proceedings of the 12th TRB National Transportation Planning Applications Conference, Houston, Tex., May 23, 2009, pp. 17–21. 11. Edara, P. K., S. Sharma, and C. McGhee. Simulation Analysis of the Hampton Roads Hurricane Evacuation Traffic Control Plan. Presented at 88th Annual Meeting of the Transportation Research Board, Washington, D.C., 2009. 12. Wolshon, B., and G. Theodoulou. Modeling and Analyses of Freeway Contraflow to Improve Future Evacuations. Presented at 83rd Annual Meeting of the Transportation Research Board, Washington, D.C., 2004. 13. Wolshon, B. Adaptation and Application of TRANSIMS for the Simulation of Regional Model Evacuation. Proceedings of the 12th TRB National Transportation Planning Applications Conference, Houston, Tex., May 23, 2009. 14. Citilabs. Modeling Traffic in Detail with Cube Avenue. http://www. citilabs.com/sites/default/files/files/IS_CubeAvenue.pdf. Accessed Nov. 14, 2011.

119

15. Ullman, B. R., N. Trout, and A. J. Ballard. Guidelines for Hurricane Evacuation Signing and Markings. Report FHWA/TX-08/0-4962-P1. Texas Transportation Institute, College Station, Tex., Dec. 2007. 16. Post, Buckley, Schuh, and Jernigan (PBS&J). Hurricane Floyd Assessment—Review of Hurricane Evacuation Studies Utilization and Information Dissemination. U.S. Army Corps of Engineers report. Tallahassee, Fla., 2000. 17. Wolshon, B., and B. McArdle. Temporospatial Analysis of Hurricane Katrina Regional Evacuation Traffic Patterns. ASCE Journal of Infrastructure Systems, Vol. 15, No. 1, 2009, pp. 12–20. 18. Transportation Research Circular E-C153: Dynamic Traffic Assignment: A Primer. ADB30 Transportation Network Modeling Committee, Transportation Research Board of the National Academies, Washington, D.C., 2010. 19. Henk, R. H., A. J. Ballard, R. L. Robideau, W. G. Peacock, P. Maghelal, M. K. Lindell, C. S. Prater, L. Loftus-Otway, P. Siegesmund, R. Harrison, L. Lasdon, Y.-C. Chiu, J. Villalobos, J. Perkins, C. Lewis, and S. Boxill. Disaster Preparedness in Texas. Technical memorandum. Texas Transportation Institute, College Station, June 2007. 20. Henk, R. H., P. Songchitruksa, S. P. Venglar, S. Samant, and G. Lim. Prototype Design for a Predictive Model to Improve Evacuation Operations. Technical Report 0-6121-1. Texas Transportation Institute, College Station, 2011. The contents of this paper reflect the views of the authors and do not necessarily reflect the official views or policies of the Texas Department of Transportation. The Transportation Safety Management Committee peer-reviewed this paper.