Tsunami vertical-evacuation planning in the U.S. Pacific Northwest as a geospatial, multi-criteria decision problem Academic research paper on "History and archaeology"

ELSEVIER

Tsunami vertical-evacuation planning in the U.S. Pacific Northwest as a geospatial, multi-criteria decision problem

Nathan Wood a *, Jeanne Jones b, John Schelling c, Mathew Schmidtlein d

a Western Geographic Science Center, United States Geological Survey, 2130 SW 5th Avenue, Portland, OR 97201, USA b Western Geographic Science Center, United States Geological Survey, 345 Middlefield Road, Menlo Park, CA 94025, USA c State of Washington Military Department, Emergency Management Division, Building 20, Camp Murray, WA 98430, USA d Sacramento State University, Department of Geography, 6000 J Street, Sacramento, CA 95819, USA

ARTICLE INFO ABSTRACT

Tsunami vertical-evacuation (TVE) refuges can be effective risk-reduction options for coastal communities with local tsunami threats but no accessible high ground for evacuations. Deciding where to locate TVE refuges is a complex risk-management question, given the potential for conflicting stakeholder priorities and multiple, suitable sites. We use the coastal community of Ocean Shores (Washington, USA) and the local tsunami threat posed by Cascadia subduction zone earthquakes as a case study to explore the use of geospatial, multi-criteria decision analysis for framing the locational problem of TVE siting. We demonstrate a mixed-methods approach that uses potential TVE sites identified at community workshops, geospatial analysis to model changes in pedestrian evacuation times for TVE options, and statistical analysis to develop metrics for comparing population tradeoffs and to examine influences in decision making. Results demonstrate that no one TVE site can save all at-risk individuals in the community and each site provides varying benefits to residents, employees, customers at local stores, tourists at public venues, children at schools, and other vulnerable populations. The benefit of some proposed sites varies depending on whether or not nearby bridges will be functioning after the preceding earthquake. Relative rankings of the TVE sites are fairly stable under various criteria-weighting scenarios but do vary considerably when comparing strategies to exclusively protect tourists or residents. The proposed geospatial framework can serve as an analytical foundation for future TVE siting discussions.

Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/3.0/).

Contents lists available at ScienceDirect

International Journal of Disaster Risk Reduction

journal homepage: www.elsevier.com/locate/ijdrr

CrossMark

Article history: Received 10 January 2014 Received in revised form 18 April 2014 Accepted 18 April 2014 Available online 26 April 2014

Keywords: Tsunami

Multi-criteria decision analysis Vertical evacuation Cascadia Vulnerability

1. Introduction

Many coastal communities throughout the world are threatened by tsunami hazards that could inundate low-lying areas only minutes after being generated by a local earthquake, landslide, or other water-column disturbance. Given this small time horizon for taking protective action,

* Corresponding author. Tel.: +1 503 251 3291. E-mail addresses: nwood@usgs.gov (N. Wood), jmjones@usgs.gov (J. Jones), John.Schelling@mil.wa.gov (J. Schelling), schmidtlein@csus.edu (M. Schmidtlein).

at-risk individuals are often taught to self-evacuate on foot to natural high ground after recognizing environmental cues of potentially imminent waves. Vehicular-based evacuations are not likely due to damaged roads after an initial earthquake [6] or prohibited in some jurisdictions due to possibility of traffic congestion and traffic-signal failures [11].

In areas where high ground is unattainable (e.g., [38]), tsunami vertical-evacuation (TVE) strategies may be warranted to minimize loss of life. For example, artificial berms, towers, buildings, and platforms have been built to provide vertical-evacuation refuges in several Japanese

http://dx.doi.org/10.1016/j.ijdrr.2014.04.009

2212-4209/Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

coastal communities [11,19] and are being planned in the United States [8] and Sumatra [13]. During the 2011 Tohoku Mw 9.0 earthquake and tsunami, a large number of TVE buildings provided safe refuge to thousands of people in Japanese coastal communities [12].

Deciding where to build TVE sites is a difficult policy question and is a topic largely undocumented in the United States [22]. One issue is the general lack of evacuation modeling to identify areas where TVE refuges may be needed [22]. Once the need for a refuge is identified, elected or appointed officials must then grapple with technical, administrative, political, legal, economic, and environmental issues to establish a TVE refuge, as they would with any development project. Community engagement is essential to ensure (1) local buy-in of a TVE site as a plausible refuge, (2) multiple community benefits beyond a refuge that can be attributed to a TVE site, (3) guarantee of 24-h access, (4) awareness of evacuee conditions, and (5) local willingness to help maintain the site [12]. Vertical-evacuation siting decisions are further complicated by the fact that multiple sites within a community and across a region may be warranted. Public officials may be confronted with multiple TVE options across multiple communities, each representing a suite of advantages and disadvantages. Reports by the U.S. Federal Emergency Management Agency [2,3] provide engineers and public officials with general guidance on issues to consider when developing a TVE strategy at one site (e.g., structural loads, design criteria, and site planning considerations) but lack any substantive discussion of how to compare multiple locations either within one community or among multiple communities.

Public officials likely lack the resources to develop TVE strategies to protect every at-risk individual and will therefore need mechanisms for understanding the implications of various options across a region. Rarely will there be a unique and optimal solution, given the large areas located in tsunami hazard zones and the varying priorities, values, and economic capital for deciding which at-risk populations to protect first. Therefore, decision makers could benefit from a spatially explicit, multi-criteria framework for evaluating TVE alternatives. This type of framework has been used for land suitability studies that involved multiple stakeholders and conflicting priorities (e.g., [4,9]). Geospa-tial, multi-criteria decision analysis (MCDA) has a rich literature (e.g., [9,20,21], for reviews) but to date, we are not aware of efforts to explore this type of evaluative framework for TVE planning. Park et al. [24] introduce the concepts and potential applicability of genetic algorithms for TVE siting in coastal Oregon (USA), but exclude several important aspects, such as community engagement, siting constraints, non-residential populations, model accessibility, and open travel across a landscape to high ground.

The objective of this paper is to demonstrate a mixed-methods application of MCDA concepts for comparing TVE options in communities that lack natural high ground. Because TVE planning is in its infancy in the United States, there are still substantial financial, legal, ethical, environmental, and land-use planning aspects to be addressed at multiple levels of government. Therefore, we do not profess to offer an exhaustive solution to TVE siting, but

instead focus on introducing MCDA concepts and the potential role of geospatial modeling in their future implementation. To demonstrate this approach, we focus on the coastal community of Ocean Shores, Washington (United States), which is one of many coastal communities in the U.S. Pacific Northwest that is threatened by local tsunamis associated with Cascadia subduction zone earthquakes. First, a series of TVE options (including capacity estimates and preferences) were proposed at community workshops [10]. After the workshops, geospa-tial pedestrian-evacuation models were developed to assess the evacuation potential of at-risk individuals and to evaluate the changes in this potential due to the proposed TVE options. Metrics to compare the multiple TVE options were developed to characterize changes in the number and type of at-risk populations that may be able to reach high ground before tsunami-wave arrival. Because all models are simplified representations of reality, we discuss areas for continued research of geospatial MCDA for TVE siting, ethical considerations of TVE siting, and benefits of using GIS-MCDA in future TVE workshops. This information supports public officials in their efforts to develop tsunami risk-reduction strategies that reflect the varied needs of an at-risk population.

2. Study area

Our case study of multi-criteria TVE planning focuses on the city of Ocean Shores in Grays Harbor County along the open-ocean, southwest coast of Washington (Fig. 1a). Ocean Shores, like all other coastal communities between northern California (USA) and southern British Columbia (Canada), is threatened by earthquakes and subsequent tsunamis associated with the Cascadia subduction zone (CSZ; [6]). In addition to relatively instant geomorphic changes to the landscape (e.g., liquefaction, lateral spread, subsidence), a future CSZ earthquake (likely magnitude 8 or greater) could create a series of large tsunami waves, the first of which reaching the Ocean Shores coast approximately 25 min after the initial earthquake [33]. Tens of thousands of people live in the tsunami-hazard zone associated with a CSZ earthquake in southwest Washington and thousands of them, in addition to thousands of employees and customers at local stores, may have insufficient time to reach natural high ground before the first tsunami waves arrive. Ocean Shores has the most significant evacuation challenges in southwest Washington given its high number of residents and greatest distances to natural high ground [38]. Although high ground is too far for most at-risk individuals to reach by foot, vehicular-based evacuations are also not probable because road networks would likely be compromised due to cracked roads, sand boils, and downed electrical lines [6]. Tsunami education efforts in the region emphasize the need for individuals in tsunami hazard zones to self-evacuate after observing natural cues (e.g., prolonged ground shaking, shoreline recession) because tsunami warnings for this area will be issued approximately 5 min after the earthquake [23], which in Ocean Shores' case represents 20% of the time they likely would have to evacuate.

Fig. 1. (a) Study area map of Ocean Shores, Washington, including modeled pedestrian travel times to safety, vertical-evacuation sites proposed during Project Safe Haven meetings, and regional map, and (b) frequency of votes for site from Project Safe Haven participants.

3. Multi-criteria decision analysis (MCDA)

MCDA is a collection of techniques for (1) structuring a decision problem that lacks a unique or optimal solution, (2) designing and evaluating a suite of feasible alternatives each with various tradeoffs, and (3) identifying the most preferred solution [21]. A MCDA that leverages geographic information system (GIS) software (noted as GIS-MCDA) provides a process that combines spatially explicit data with decision makers' preferences for evaluation criteria and various weighting techniques [16]. The structured process provides a feedback mechanism to quickly identify and visualize the implications and policy tradeoffs of various alternatives.

GIS-MCDA can be multi-attribute problems that satisfy a single objective or multi-objective problems with a continuous set of options and multiple objectives [7,21]. We assume that TVE planning is a multi-attribute problem for this case study, in that solutions center on selecting refuges among a discrete suite of potential locations with the sole objective of saving lives. TVE sites can have multiple attributes related to design considerations and site benefits, but possibilities are not infinite within a community due to property rights, environmental restrictions, or other land-use limitations. Other complementary TVE objectives (e.g., year-round recreational opportunities or environmental habitats) are not modeled here, nor are conflicting objectives, such as the use of sites for non-refuge purposes.

Another point of distinction in GIS-MCDA is whether there are single or multiple goal preferences. A single decision-maker's problem involves only one goal preference (e.g., saving lives of children above all other population types) for the individuals or groups involved. If multiple interest groups are involved, each with their own preferences, then it becomes a participatory, group decision process [17]. TVE planning will likely entail group decision making with multiple preferences because of the involvement of community members and public officials (ranging from local to federal) in the site-selection, design, approval, and funding processes.

The role of data and decision-rule uncertainty is also an important MCDA element [9,21]. We discuss decision-rule uncertainty in Section 4.3 where we compare weighting scenarios for population criteria. Data uncertainty can be treated deterministically or stochastically. One source of data uncertainty is the spatial extent, height, and arrival time of tsunami waves due to the range in source parameters (e.g., earthquake attributes) that govern these variables. Another area of data uncertainty is the spatial and temporal distribution of at-risk populations because of the highly variable and dynamic nature of coastal populations due to time of day, season, and mixed populations (e. g., residents, employees, schoolchildren, tourists). One could stochastically frame the tsunami-refuge problem by developing probabilistic maps for both tsunami inundation and population distributions. Although there are probabilistic tsunami-hazard mapping efforts (e.g., [14]), we are not aware of efforts to probabilistically map populations. Because of the exploratory nature of this case study, we approach the TVE problem from a deterministic perspective, which translates to the use of maximum tsunami-hazard zones (based on credible scenarios) and population distributions (regardless of time of day or season). This approach follows that of most emergency-management procedures in that it prepares a community for a worst-case possible scenario, instead of an actuarial approach more appropriate for economic mitigation efforts that focus on more probable (and likely smaller) events over a given time period.

4. Methods

Based on our brief overview of MCDA concepts, we frame TVE planning in our study area as a multi-attribute, group-decision process under conditions of relative certainty. This section illustrates a mixed-methods approach to steps in a MCDA process including (1) problem definition through geospatial evacuation modeling, (2) identification of stakeholders, alternatives, and criteria through community workshops, and (3) evaluation of alternatives and weighting of criteria using statistics, and (4) decision and sensitivity analysis. We introduce aspects of decision and sensitivity analysis but do not fully execute them, given the exploratory nature of this article and the unresolved land-use and other societally relevant issues of TVE planning, such as ethical considerations. In this case study, the community workshops were held first due to the timing of practitioner involvement and geospatial modeling was done later to further examine TVE siting

issues. Throughout the paper, we highlight areas where future workshops could benefit from having the modeling done before or during the workshops. The following sections summarize additional information on each of the analytical steps and the various input data that were used.

4.1. Modeling the current evacuation landscape

The first step in a MCDA is problem definition. In our case, there is considerable evidence that significant tsunamis have repeatedly inundated the U.S. Pacific Northwest coast and will strike again due to earthquakes associated with the Cascadia subduction zone (CSZ). Previous work has established that Ocean Shores has a substantial life-safety problem because of the thousands of residents and tourists in tsunami-hazard zones that would only have minutes to reach natural high ground after experiencing a CSZ earthquake (Wood and Soulard, [36], [38]).

To further demonstrate the evacuation problem and to assess tradeoffs of TVE alternatives, we created a baseline of pedestrian-evacuation potential under current landcover conditions and population distributions. This was done using an anisotropic, least-cost-distance (LCD) model implemented in ESRI's ArcMap 10.1/SP1 software, following methods described in greater detail in Wood and Schmidtlein [37,38]. This approach focuses on landscape characteristics related to elevation and land cover to calculate the most efficient path to safety from every location in a hazard zone, with the difficulty of traveling through each location represented as a cost surface. Anisotropy incorporates direction of travel (e.g., the influence of a given slope will vary whether travel is uphill, downhill, or perpendicular to the slope). The path distance approach within LCD modeling calculates distances and slopes between cells of varying elevations. The modeling estimates travel times based on optimal routes; therefore, actual travel times may be greater due to evacuee route preferences and environmental conditions during an evacuation. We use a LCD-based approach instead of other methods, such as agent-based models (e.g., [18,39]), because our focus is on understanding the spatial distributions of evacuation times within a community, rather than individual evacuee behavior.

Data required for the pedestrian-evacuation modeling include a hazard zone, elevation, and land cover. The tsunami-hazard zone by Walsh et al. [33] for the southwest Washington coast delineates likely areas of inundation associated with a magnitude (Mw) 9.1 earthquake along the Cascadia subduction zone. The safe zone is any remaining land in the study area, denoted by a shoreline layer (Washington Department of Ecology, [34]), that is not in the tsunami-hazard zone. A 2010 1-m, LiDAR-derived digital elevation model ([35]) was used to derive slope, which was then coupled with a lookup table based on Tobler's [27] hiking function that converts slope to speed conservation values (SCV). SCV represent the proportion of maximum travel speeds that are expected on areas with given conditions. 2009 1-m pixel resolution, orthorectified color imagery [30] was used to manually classify and map land cover, which was then reclassified

into SCV based on terrain-energy coefficients discussed in Soule and Goldman [25]. Values include "No Data" to note where travel is not possible (e.g., over water and through fences or buildings) and then a range from 0.5556 to 1.0 to note the percentage of the base travel speed (assuming constant energy expenditure). SCVs were mapped for impervious surfaces (1.0), grass, dirt/gravel surfaces, and other developed areas (0.9091), light brush (0.8333), heavy brush (0.6667), and wetlands, sand, and shoreline (0.5556). Cost surfaces that integrate land cover and elevation SCV maps were generated using ESRI's Path Distance tool and then converted to maps of pedestrian travel times using a slow-walking speed of 1.1 m/s, which is conservative but appropriate given a mixed population with ranges in age and physical mobility [31].

Because of the slow travel-speed assumption and the likelihood that many at-risk populations will move faster, estimates of population exposure as a function of travel time should not be interpreted as definitive mortality estimates. Other factors that will influence travel time include the time needed to decide to evacuate, to get out of a building or car, and to navigate unfamiliar surroundings. Some individuals in tsunami-hazard zones also may wait to receive a warning from official tsunami warning system, which may or may not be effective given the potential for damage to communication systems from the preceding earthquake and delays in warning dissemination as it travels through various agency channels. Because individuals may be moving faster than modeled travel speeds but also will likely be delayed in their evacuation for various reasons, we believe the use of a slower travel speed provides a good overall approximation of travel times to safety.

Various datasets were assembled to characterize the at-risk population. Residential estimates were created by manually identifying residential structures as points from the 2009 imagery and then disaggregating block-level population estimates in the 2010 U.S. Census Bureau count [29] to these residential points. Employee points were developed using a 2011 version of the Infogroup Employer Database [15], which is a proprietary database that includes business locations, employee counts, and type based on the North American Industrial Classification System (NAICS). We used NAICS codes to classify certain businesses as public venues (e.g., museums, overnight accommodations, and parks or other outdoor venues), dependent-population facilities (e.g., child services, elderly services, medical centers, and K-12 schools), and other community businesses that would likely have substantial numbers of customers (e.g., banks or credit unions, civil or social organizations, gas stations, government offices, grocery stores, libraries, and religious organizations). We estimated the number of customers at public venues, dependent-population facilities, and community businesses because equal units of measurement (i.e., number of people instead of a mix of people and facility counts) are needed when we explore the use of weighting criteria on TVE priorities. Based on our work experience in this study area for many years, we estimated 100 customers at each public venue in our study area, which included hotels, churches, movie theaters, and RV parks, except for one

substantially larger hotel where we estimated 318 people based on an assumption of two-person occupancy for 159 rooms [28]. Twenty customers were estimated for each community business, which included tourist shops, art galleries, barber shops, city offices, and restaurants. For dependent-population facilities, we estimated 20 customers at the various medical offices and clinics, but then used 2012-2013 enrollment figures for the elementary (248), junior high (111), and senior high (216) schools [26]. Each population layer was overlaid on the evacuation-time maps to estimate the number of individuals or facilities in terms of travel time to safety. Population estimates are not mutually exclusive and should not be combined because residents are also business customers and may also work or attend school in the area. Results should be interpreted as maximum values at specific locations.

4.2. Project Safe Haven

Following problem definition, a MCDA involves the identification of stakeholders, alternatives, and comparative criteria. This can be based on participatory approaches to engage at-risk populations or solely on expert opinion in areas where participatory processes are less common. In our study area, stakeholder engagement was accomplished through a series of community workshops that were collectively called Project Safe Haven and were held in Ocean Shores, as well as other coastal communities in the region. Described in greater detail in Engstfeld et al. [10], Project Safe Haven involved several steps to develop and compare alternatives. The first step involved a site visit by a project team to help community leaders identify opportunities for, and barriers to, potential TVE projects. Landscape characteristics were noted, including potential tsunami inundation depths and vacant parcels. A public meeting was then held to introduce the concept of TVE planning and to solicit community ideas of the possible strengths and weaknesses of various TVE options. Some participants were specifically invited due to technical or local expertise (e.g., local officials, scientists, and engineers), but the majority were self-selecting after reading advertisements in various local media outlets and seeing notices posted in U.S. Post Offices. Participants included adult-residential-care business owners, a disability advocate, representatives of the school district, and several elderly residents. The first meeting in Ocean Shores had 75 participants and was held March 10, 2011, one day before the 2011 Tohoku earthquake and tsunami disaster in Japan.

At the first meeting, participants used interactive hazard maps with acetate overlays to estimate distances from a point within 15 min of travel time to discuss conceptual locations for the structures, and the advantages and disadvantages of each structure type at a particular location. Alternatives for TVE sites were generated through open, collaborative discussions among community participants. Preference was for empty public lots that could be easily converted to refuges without requiring buy-outs from existing property owners. The primary criterion were spatial proximity to perceived concentrations of at-risk populations with subsequent criteria related to population types, such as residents, employees, schools and day-care

centers, hospitals, adult-residential-care facilities, hotels, and public venues that cater to tourists. A second meeting was held where project members presented a consolidated summary of alternatives to community members and collectively the group conducted an analysis of strengths, weaknesses, opportunities, and threats (commonly called a SWOT analysis) of each alternative. The second meeting in Ocean Shores had 150 participants, which is twice the number from the first meeting and likely due to its timing of two weeks after the 2011 Tohoku tsunami disaster.

Once a community developed a preferred strategy, community-wide meetings also open to the public were held to present the final strategy and review the strategies for comprehensiveness, redundancy, coordination of efforts, and to solicit input on community priorities for future implementation. Collaborative design sessions with architects were organized to identify specific structure locations and how the structures could best fit into the community context, as well as multipurpose components that could be incorporated. Other criteria that entered the discussions at this point included cost considerations and potential long-term maintenance issues. Potential day-to-day uses for each vertical evacuation structure at each proposed site were incorporated into the overall vertical evacuation strategy and final sketches were presented back to the community as hand-drawn conceptual designs. Initial cost estimates were then completed and in some communities, participants voted on the various options to help prioritize future development.

4.3. Evaluation of alternatives using comparative maps and metrics

Geospatial efforts to characterize differences in evacuation potential due to TVE alternatives focused on two elements. The first element portrays spatial variability in evacuation potential. A series of maps were generated to show what parts of the community could reach high ground given a specific alternative developed previously at the Project Safe Haven workshops. This was done to demonstrate that no one single alternative could effectively cover the entire study area and that decisions based on tradeoffs would need to be made. A second set of evaluative metrics focused on describing variations in population exposure as a function of travel time to safety based on the various TVE alternatives. The primary evaluative criteria were the additional number of residents, employees, dependent populations, public venue visitors, and customers at local businesses and offices that could

reach a proposed TVE refuge in less than 25 min (i.e., predicted wave-arrival time). The effectiveness of an individual TVE refuge was gauged by changes in the population exposure of these various population groups relative to current landscape conditions. It did not take into account TVE capacity because we felt that could be determined at a later time during project design based on our analysis.

Once evaluative criteria are established, decision rules are implemented within a MCDA to help the decision maker to select the most preferred solution. At this early stage of TVE discussions, workshop participants were not asked to develop weighting criteria (e.g., school children are twice as important as business employees). Instead, for this exploratory analysis, we implemented a series of simple weighted sum models to demonstrate the potential for stakeholder-derived criteria in future discussions. A weighted sum model is considered one of the most common and simplest approaches in the GIS-MCDA literature [21] and is defined as:

Ai = 2 jj

where n denotes the number of criteria (which is five in this study, namely residents, employees, public venue visitors, community business customers, and dependent populations), Wj denotes the relative weight of importance for a population metric, aij denotes the value of that population metric for a TVE refuge option, and Ai is the calculated total score for one TVE refuge alternative that considers all five of the population metrics. In our case study, all criteria provide benefits to protecting lives; therefore, higher Ai values represent relatively better options.

The role of weighting criteria is demonstrated using ten scenarios that prioritize different populations (Table 1). The estimated number of people within 25 min of a TVE site for each of the five population groups is multiplied by the appropriate criteria weight for each scenario. The five values for a given TVE site are then added and normalized to the maximum number of people at a location for each scenario. The normalization was done because of differences in weighted-sum totals between the scenarios. TVE options are then compared based on their relative scores (between 0 and 1) for the weighting assumptions of a given scenario.

Scenarios include one assuming equal weighting, several that focus only on one population group and ignore the other groups, and several that emphasize one group

Table 1

Criteria weights for various scenarios of prioritizing populations.

Equal Tourists Dependents Customers Residents Winter

Only Primary Only Primary Only Primary Only Primary

Residents 0.2 0 0.125 0 0.125 0 0.112 1 0.335 0.25

Employees 0.2 0 0.125 0 0.125 0.25 0.222 0 0.166 0.25

Public venues 0.2 1 0.5 0 0.125 0.25 0.222 0 0.166 0

Community offices and businesses 0.2 0 0.125 0 0.125 0.25 0.222 0 0.166 0.25

Dependent populations 0.2 0 0.125 1 0.5 0.25 0.222 0 0.166 0.25

but don't exclude the remainder. For the equal weighting scenario, each variable has a weight factor (wj) of 0.20 (100% divided by five classes). For the scenarios that focus exclusively on one group (listed as "only" for public venues, dependent populations, or residents in Table 1), the group of interest has a weight of 1.0. The "customers only" scenario excludes residents but equally divides the weight among the other groups that will have tourists or customers. For the scenario that emphasizes customers but doesn't exclude the others, we doubled the weight of these businesses and include a single weight for residents. For the scenarios that emphasize tourists or dependent populations but do not completely exclude others (listed as "primary" in Table 1), we assigned a value 0.5 for the primary group based on the assumption that tourists or school children may be on site for half of the day and then divide the other 0.5 evenly across the remaining classes. For the scenario that emphasizes residents but doesn't exclude the others, we used the same time-based reasoning for assigning residents a weight of 0.335 (assuming people are at their house for at least one-third of the day) and dividing the remaining 0.66 among the other classes. The "winter" scenario focuses on all population categories except for public venues because tourism is low in the study area during winter months but residents will still be in the area, as will employees at stores, children at schools, and patients at hospitals.

4.4. Sensitivity of weighting criteria

A final step in a typical GIS-MCDA involves a sensitivity analysis to explore the relative importance of criteria and the robustness of the relative scores. If relative scores vary dramatically based on modeling assumptions, then participants will want to ensure that all assumptions are justified and well thought out. If relative scores do not vary dramatically, then participants may not need to devote as much time to defending their perspectives on difficult and subjective opinions on the relative importance of various modeling inputs and weighting criteria. To examine the sensitivity of the relative scores to various modeling assumptions, we calculated the mean and one standard deviation from the mean of the normalized values for each TVE site for each of the ten weighting scenarios. We also identify the normalized values for the residents-only and tourists-only scenario (i.e., public venues), since these two scenarios may best reflect potentially conflicting perspectives of priorities—namely focusing risk-reduction strategies on year-round residents or on tourists.

5. Results

5.1. Baseline conditions of evacuation potential

Assuming a slow walking speed for evacuees (1.1 m/s), pedestrian travel times to high ground outside of the tsunami-hazard zone in Ocean Shores range from a few minutes to more than 200 min at the southern tip of the peninsula (Fig. 1a). Based on 2010 US Census population counts and 2011 InfoGroup business data, the tsunami-

hazard zone in our study area contains 6234 residents, 1431 employees, 3318 visitors to 30 public venues, 2200 customers to 110 local businesses and agencies, and 715 dependent populations at 10 facilities, which includes three schools and various medical offices. As discussed earlier, these population estimates are maximum values for specific locations and are not mutually exclusive because residents are also business customers and may also work or attend school in the area.

Due to the great distances to high ground and limited time available to reach it, many of the at-risk population in Ocean Shores may have difficulty evacuating before the first wave is expected to arrive 25 min after the earthquake. Merging evacuation travel times with population locations suggests that 5,041 residents (81% of those in the study area), 1001 employees (70%), 1918 visitors to public venues (58%), 1520 local-business customers (69%), and 308 dependent populations (43%) are in areas where pedestrian travel times exceed predicted wave arrival times. If we assume the bridges throughout the community are destroyed from the initial CSZ earthquake, than an additional 69 residents in Ocean Shores would be unable to evacuate prior to wave arrival.

5.2. Changes in evacuation landscapes from TVE options

Participants in community workshops associated with Project Safe Haven identified 52 sites as potential TVE sites throughout Grays Harbor and neighboring Pacific Counties. Twenty of these proposed TVE sites were identified in Ocean Shores. For each of the sites, participants decided upon a construction type (berm, building, or tower) and estimated the required capacity for evacuees based on current and future population derived from local jurisdiction comprehensive plans (Fig. 1a). During the second workshop in Ocean Shores, 86 participants voted on which TVE option would best serve the Ocean Shores community. Site 14 at the southern end of the community received the highest number of votes, followed by site 4 in the center of town (Fig. 1b).

Building all 20 proposed TVE options in Ocean Shores may be ideal to minimize potential loss of life from future Cascadia-related tsunamis but is unlikely over the near term given the limited financial resources at city, State, and Federal government agencies. Our first step in framing TVE site planning as a multi-criteria decision problem was to map changes in the evacuation landscape given the various TVE options to demonstrate that difficult decisions would need to be made. We did this by modeling and mapping travel times to safety for the baseline condition and for each TVE option proposed in Ocean Shores (Fig. 2). Again, the first tsunami waves are predicted to arrive in this area approximately 25 min after the earthquake. We identify areas with travel times of 25-29 min to show areas where additional people may successfully evacuate if they increased travel speeds. Areas with travel times greater than 30 min are in gray in Fig. 2 and denote areas where successful evacuations are unlikely. Under baseline conditions, only areas in the northern part of Ocean Shores have travel times that are less than 30 min. The purpose of the travel-time maps for the 20 proposed options is to

Fig. 2. Maps of modeled pedestrian travel time to safety (minutes) under current conditions and for each of the 20 tsunami vertical-evacuation (TVE) options proposed during the Project Safe Haven workshop.

show that each TVE option could help individuals near the proposed site, but that no one TVE option can help everyone and be considered the only solution to the evacuation problem. Therefore, difficult decisions would need to be made on where to locate TVE sites given limited resources, hence the need for a multi-criteria, decision-analysis framework.

The multiple travel-time maps were then combined to identify what areas may be served by multiple TVE options and which areas may be overlooked (Fig. 3). To do this, we created buffers for each TVE option that delineated travel times less than 25 min. The buffers were added and the value of each 1-m pixel notes the number of TVE options that could provide high ground in less than 25 min of travel time. The yellow areas in Fig. 3 denote sections that can reach natural high ground, regardless of any TVE option. Site 4 is in downtown Ocean Shores and results demonstrate that any one of the other three TVE options near it may also provide a refuge to downtown residents, employees, and store customers. If there is interest in focusing on downtown Ocean Shores, then deciding among the four sites may not be as difficult for stakeholders if they know each of the sites could provide an evacuation refuge for the area.

Another conclusion to be drawn from this frequency map (Fig. 3) is that certain areas may be out of reach for any one of the proposed TVE sites. The most obvious sites are the southern and southeastern tips of the peninsula in

Ocean Shores, which both represent natural areas with little on-site populations, other than occasional beachcombers. However, there are several residential areas that also would be outside the reach of TVE sites, such as areas north of sites 20 and 9 and south of site 6. Natural areas west of sites 3 and 4 are also outside the reach of these sites, but may have substantial numbers of tourists on the beach given its relative proximity to downtown Ocean Shores. Maps similar to Fig. 3 could be used in community discussions to help determine whether these gaps in coverage are considered acceptable to stakeholders. If they are not acceptable, then new TVE options could be proposed to eliminate these gaps. In this case study, geospatial modeling of TVE options occurred after the Project Safe Haven workgroups but could be done dynamically and iteratively during future workshops.

5.3. Gauging capacity expectations for TVE options

In addition to voting on TVE priorities, participants at Project Safe Haven workshops estimated the necessary capacity of evacuees for each proposed TVE site. Geospatial modeling of pedestrian evacuation travel times provides decision makers the opportunity to determine whether or not those estimates are realistic given the evacuation landscape. In Ocean Shores, we compared capacity estimates developed at the workshops with estimates derived by estimating the number of residents, employees, public

I I Natural high ground accessible

□ Safe zone

O Tsunami vertical-evacuation site

□ Water

Number of TVE sites that are within 25 minutes of travel time None

Fig. 3. Number of tsunami vertical-evacuation (TVE) sites within 25 min of travel time. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

venue visitors, local-business customers, and dependent populations that are within 25 min of travel time. Results indicate that capacity estimates developed by workshop participants were adequate for some TVE options (sites 3, 14-16) but significantly underestimated capacities for most of the other sites (Fig. 4). For example, site 5 was estimated by workshop participants to require a capacity

of 350 people, but evacuation modeling done after the workshop suggests that there are approximately 2,435 people within a 25-min window of that proposed site. Other sites where capacities were significantly underestimated include sites 4, 6, 7, 17, and 20. As TVE planning discussions mature throughout the world, pedestrian evacuation modeling using GIS-MCDA could play an important role in supporting community members by estimating realistic capacity requirements during the workshops.

5.4. Population metrics to compare TVE options

Proposed TVE sites were evaluated based on the additional number of residents, employees, public venue visitors, dependent populations, and local-business customers that would be within a 25-min window of travel time if the TVE site existed (Fig. 5). The dotted black line in each graph notes the current percentage of the community that could be saved if evacuees moved to natural high ground. For example, 19% of the residents in the study area currently may be able to reach natural high ground and the construction of TVE site 5 may increase this to 38% (Fig. 5a). Results indicate that certain sites would provide refuges to certain populations but that no one site was universally advantageous to all five of the population metrics. For example, TVE site 5 may be the preferred option if only residents were concerned, but TVE site 4 may be a preferred option if the discussion focused on tourists, customers, and employees. TVE sites 7 and 20 would most benefit dependent populations; in both cases, a TVE refuge would be near the local elementary school. Results also indicate that certain sites would have minimal impact to the community as a whole. For example, site 15, which is located in an area already within 1 to 24 min of high ground, would obviously benefit people near the site by providing them an even closer refuge, but would not increase the number or community-wide percentage of saved individuals or facilities in any category.

The benefits of the various proposed TVE sites are tempered in some cases if one doesn't assume bridges will be functional after being subjected to a magnitude 9.0 earthquake followed by regional subsidence of one meter and liquefaction of unconsolidated material. Fig. 5 also shows how the percentage of each population group that is within 25 min of the various TVE options change if a nearby bridge were destroyed. Results demonstrate that the benefit of site 6 is most compromised with regard to resident protection, in that 635 of the 1090 residents that could be within 25 min of travel time to the proposed TVE site would be unable to reach it if the bridge to its west was destroyed. The benefits of TVE site 5 is also compromised if bridges are not functioning, although more so for employees and customers and less so for residents than site 6.

To examine the relative merit of the various 20 proposed TVE sites in Ocean Shores, we developed a comparative index. Because the maximum number of people that have accessible refuges varies for each TVE scenario, we normalized values in each of the five categories to the maximum number of additional people that would have a refuge within 25 min of travel time (2648 people at site 4). After the minmax normalization, the number of additional people at each

2,000-

1,000-

I I Additional people within 25 min I I Estimated capacity

12 3 4

i i i i i i i i

9 10 11 12 13 14 15 16 17 18 19 20

Tsunami vertical-evacuation option

Fig. 4. Comparison of estimated tsunami vertical-evacuation capacity for each proposed site and the additional number of residents, employees, and facilities within a 25-min travel time from the site.

TVE site is transformed to represent a percentage of the maximum found at site 4 (hence a value of 1.0 at site 4), resulting in a relative assessment of the number of additional residents, employees, or businesses that would be within a 25-min travel window to a certain TVE site (Fig. 6a). The five categories were considered of equal weight so no additional weighting was done at this point. Higher scores reflect sites that potentially provide refuge to greater numbers of at-risk populations. Overall, site 4 would result in the greatest number of additional people (2648) that potentially could be served by that TVE site, including an estimated 372 employees, 700 public venue visitors, and 820 local-business customers. Site 4 would also be near approximately 752 residents and 40 dependent populations, but other sites may provide greater protection to more residents (site 5 with 1198) and dependent populations (sites 7 and 20 near the elementary school). After site 4, sites 5 and 17 may provide refuge to the greatest number of people in the study area. The remaining proposed TVE sites would benefit local populations but would have relatively limited impact to the entire community. If one assumes bridges are too damaged for use in the aftermath of the CSZ earthquake, the relative ranking of the sites change only slightly (Fig. 6b). Because of the number of bridges near site 5, its ranking decreases from third down to fifth. Aside from changes at site 5, the rest of the relative scores and subsequent rankings are fairly stable.

5.5. Sensitivity of weighting criteria

Composite values normalized to the maximum total value for each TVE option do vary when weighting criteria are introduced to prioritize different population groups. Fig. 7 summarizes the average and one standard deviation for the normalized weighted sums of each TVE option across the ten weighting scenarios (Table 1). For discussion purposes, we also highlight the values assuming equal weighting across the five population groups, the scenario that focuses exclusively

on residents, and the scenario focusing exclusively on public venues that are likely to have tourists. In general, the average value for a TVE site across the ten weighting scenarios is a close approximation for the equal weighting scenario (r2=0.99). Some sites (e.g., sites 1-3) demonstrate small variances among the various weighting scenarios, whereas others (e.g., sites 6, 7, and 20) have a wide range of relative scores. For almost all TVE sites, the residents-only and tourists-only scenarios yield relative scores that are on opposite ends of the range of values for a TVE site. For example, the normalized weighted sums at site 6 are 0.14 for a focus on tourists but 0.91 for a focus on residents, meaning site 6 may be the best option for protecting residents and the least effective for protecting tourists.

Although there is variability in relative scores at a particular TVE site, the relative ranking across the study area does not vary substantially. There are three general clusters of TVE sites based on the composite values. Sites 4, 5, and 17 consistently rank in a top tier regardless of how the five population groups are weighted. Site 6 can be considered part of this top tier if a community chose to exclusively focus on protecting residents. For TVE sites in the middle tier (sites 6, 7, 12, 13, and 18-20), the relative rankings of these sites fluctuate somewhat depending on the weighting scenario. For example, sites 18 and 19 are more favorable for a focus on residents, sites 12 and 13 are more favorable for tourists, and as mentioned previously site 20 is favorable for elementary schoolchildren. Relative rankings for TVE sites in the bottom tier (i.e., sites 1 -3, 911, and 16) do not vary substantially across the weighting scenarios and typically have normalized values of less than 0.25.

5.6. Revealed preferences in TVE siting

During the Project Safe Haven meetings, participants were provided the opportunity to vote for certain TVE sites

50 ■

Residents

I I Bridges required □ Bridges not required -- Current situation

I I I "I

1 2 3 4 5 6 7

I I I I I I I I "I

9 10 11 12 13 14 15 16 17 18 19 20

0 I ' I I I I I I I I I I I I I I I I I I I

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

100 ■

Dependent-care populations

I I I I I I I I I I I I I I I I I I r

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

0 I ' I I I I I I I I I I I I I I I I I I I

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

Public venue customers

£ 0 I 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i 1 1 i

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

Tsunami vertical-evacuation option

Fig. 5. Percentage of (a) residents, (b) employees, (c) dependent populations, (d) community-business customers, and (e) public-venue visitors in Ocean Shores, WA, that would have less than 25 min of travel time to safety, given the various proposed TVE options. Percentages include assumptions of functioning bridges and non-functioning bridges. The dashed line represents the current conditions.

Normalized number of additional people that would require less than 25 minutes of travel time to safety

T3 CD .t!

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Tsunami vertical-evacuation option

Normalized number of additional people that would require less than 25 minutes of travel time to safety

e lu al

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Tsunami vertical-evacuation option

□ Dependent-care facilities □ Public venues □ Residents D Community businesses □ Employees

Fig. 6. Comparison of normalized indices for the percentage of additional residents, employees, public venues, community businesses, and public venues that would require less than 25 min of travel time to safety given a proposed TVE site. (a) Assumes all bridges are functioning, whereas (b) assumes bridges would be unavailable for crossing waterways.

to identify their preferences if funds were to become available. Because post-workshop interviews were not conducted, we are unable to document expressed preferences of the participants. Instead, we examined potential revealed preferences by conducting a regression analyses that compared the number of votes for each TVE site with the community percentage of estimated residents, employees, tourists at public venues, dependent populations, and community-business customers that would be within a 25-min walk of a specific TVE site based on

modeling done after the workshops. In addition, we included the current travel time to safety at a proposed TVE site as a geographic indicator of potential need for that site. The number of votes is considered the dependent variable, whereas the other variables are considered independent. Before running the regression analysis, we checked for collinearity between the independent variables. Correlation coefficients for employees compared to estimated populations at public venues and community businesses each exceeded 0.8, suggesting the presence of

Fig. 7. Comparison of weighted normalized population indices summarizing the estimated number of people that would require less than 25 min of travel time to safety given a proposed TVE site. Ten weighting scenarios (Table 1) were used to calculate average and one standard deviation values. Resident-only and tourist-only weighting scenarios are also shown for discussion.

Tsunami vertical-evacuation option

Average □ Residents only J One standard dev.

□ Equal weighting Tourists only

significant collinearity. Therefore, estimated populations associated with public venues and community businesses were excluded from the regression analysis.

Multiple regression analysis of 86 TVE votes (dependent variable) compared to estimated resident, employee, and dependents populations, and travel time to high ground (independent variables) suggests that travel time to safety is the only significant independent variable (p=0.04, coefficient=0.04). The other independent variables had low p-values suggesting that they are not statistically-significant indicators of TVE votes, including residents (p=0.40, coefficient = 16.07), employees (p=0.35, coefficient = 12.35), and dependent populations (p=0.27, coefficient=-10.7). The overall regression model results were moderate but not statistically significant (p=0.18, r2=0.32) considering the small sample size of our case-study perspective and the inherent difficulty in predicting human behavior. The overall regression model in this study may not be statistically significant because other relevant variables may have been left out of the analysis, or perhaps the sample size is too small.

The statistical results identifying geographic distance as potential factor in TVE votes make sense, however, when one visually compares the distribution of votes (Fig. 1b) and the map of modeled travel time to safety (Fig. 1a), where site 14 is the farthest from natural high ground (186 min) but also received the highest number of votes (19). It is unclear why participants voted for the farthest site during the workshop. They may have all come from the neighborhood where TVE site 14 would be located or the whole community felt this neighborhood had the greatest need for a TVE site because it was the farthest from high ground. Regardless of why, results of the regression analysis suggest that geographic distance had some influence on decision making, whereas the actual

number and distribution of at-risk individuals across the study area did not. In future workshops, one could interview participants to gather expressed preferences and then compare them to revealed preferences using statistical and modeling methods described here. This approach could potentially identify gaps between expressed and revealed preferences, or determine if people lacked sufficient information to implement their expressed preferences reliably.

6. Discussion

Recent disasters (e.g., 2004 Indian Ocean, 2010 Chile, and 2011 Tohoku) have raised global awareness of the deadly consequences for coastal communities that lack access to natural high ground to escape tsunami inundation. As research continues to improve our understanding of tsunami hazards and of coastal populations threatened by these hazards, discussions of TVE strategies will become more common. Existing work has focused on advancing the engineering of TVE structures. Our article provides a geographic framework to analyze the trade-offs between various options for TVE locations. In this section, we focus on other topics that warrant additional discussion to support TVE siting. Our goal is not to provide a comprehensive list of issues, as such a task would require input from many fields not currently engaged in TVE siting discussions, including historians, legal experts, ethicists, education experts, and other social scientists. We hope that our limited discussion here broadens and encourages more conversation on how to best protect at-risk populations in coastal communities from local tsunami threats.

One area that deserves additional discussion is ethical considerations. In many public health issues, decisions are

based on utilitarianism, which is a theory that holds that the most ethically correct choice is the one that yields the greatest benefit to the most people [1]. Figs. 5 and 6 summarizing comparative population metrics are our attempt to implement a utilitarian perspective of maximizing community benefits of a TVE refuge. While striving to achieve maximum benefits, public officials are also expected to strive for justice during a decision problem, in which the distribution of benefits and burdens is equitable and decisions are applied fairly and consistently across people and space. Achieving maximum benefits for the community while also providing equitable distribution of benefits will be challenging for TVE siting given the limited resources and the potential for substantial loss of life in areas not served by a TVE refuge. Certain steps can be taken to improve procedural justice, such as transparency of the process, engagement with stakeholders, consistency in the treatment of at-risk populations, impartial decision makers, and accountability for decisions [1]. While these practical steps will help officials to effectively engage at-risk populations in TVE siting decisions, more guidance could be provided on the competing pressures of maximizing the public good while also minimizing inequities within an at-risk population.

In our case study, maximizing the number of lives saved from future tsunamis is considered the primary objective of TVE siting. The number of saved lives as the defining success criteria can be tempered, however, if weighting criteria are introduced (e.g., emphasizing school children over other populations). And as we demonstrated, priorities are indeed sensitive to changes in these weighting scenarios (Table 1, Fig. 7). Because weighting criteria reflect certain values or priorities, community engagement is critical in this process. In this study, we assumed equal weighting of population criteria (Fig. 5) and hypothetical weighting scenarios (Fig. 7) but as TVE refuge discussions mature, different approaches for comparing alternatives, such as analytical hierarchy process (AHP), could be implemented to develop weighting criteria [21]. For AHP-derived weights to have credibility and relevance, TVE workshop organizers need to ensure key decision makers that truly represent the priorities and values of the community are in attendance; however, this raises the ethical questions of who has the power to decide whether or not the appropriate people are being asked to derive AHP weights and whether or not all evaluative criteria are included in the pairwise comparisons. For example, issues such as costs, environmental considerations, land ownership, and site conditions may play significant roles in a decision making process but were not expressly chosen as variables during the Project Safe Haven workshop discussions. It is unlikely that there exists one conclusive list of variables that is universally applicable to all coastal communities; therefore, more research on relevant TVE siting factors is warranted.

One area for improvement related to identifying relevant TVE siting factors is site suitability modeling. In our case study, geospatial evacuation modeling was based on a set of TVE refuges determined by community participants at the Project Safe Haven workshops (Fig. 1). Site suitability therefore was an implied process based on the revealed

preferences of workshop participants. In future workshops, maps of suitable sites based on travel-time maps, land cover conditions, undeveloped parcels, and population hotspots could be used to highlight potentially favorable areas, to initiate TVE siting discussions, and to minimize gaps in refuge coverage (Fig. 3). Suitability maps could also reflect evacuation-route preferences, which may vary for residents and tourists. For example, residents may choose to avoid certain routes known to have menacing dogs or thick, seasonal vegetation and tourists may not follow optimal routes because they lack familiarity with the landscape.

In addition to procedural justice and weighting criteria, other ethical considerations have been raised concerning TVE siting. Velotti et al. [32] discusses the issue of liability in the event of TVE refuge failure, either because of inferior construction or because a tsunami exceeded the design considerations that were based on smaller, possibly more likely, events. If a TVE refuge fails for either reason, then the potential for life loss may be even greater, given the eventual false sense of perceived safety that could contribute to greater development near a TVE refuge, among other variables that could affect development. This is referred to as the "levee effect" or the "safe development paradox" in the case of riverine-flood mitigation strategies (Burby, [5]).

Evacuee welfare, with regard to the length of occupancy, capacity (Fig. 4), and quality of life of evacuees at refuges, is another other ethical question that is largely unexplored in TVE siting [32]. Fig. 4 demonstrates how population capacity may be substantially underestimated for certain TVE options. In addition to the potential for overcrowding, Fraser et al. [11] documents how some survivors of the 2011 Tohoku earthquake and tsunami disaster spent multiple days at TVE refuges that were only designed for up to 6 h of use and therefore had inadequate shelter, food, and water. Discussions regarding quality of life and access are important for officials that are developing TVE sites, but also for at-risk populations to help manage their expectations. Expectations for TVE access are especially important because TVE structures may or may not be required to meet the access requirements of the U.S. Americans with Disabilities Act of 1990 (Public Law 101336, 4.3.10), depending on the requirements of the administrative authority having TVE jurisdiction.

The role of land use planning in directing growth towards areas that are presently or may be served by TVE in the future is another topic that has not received sufficient attention in the literature. Including MCDA and TVE siting as part of a future land use planning process could help ensure current, as well as projected, populations are considered when determining TVE capacities. With the development of maps and MCDA-derived results, local officials may be able to develop appropriate subarea plans or incentive-based zoning strategies to support private sector implementation as part of new development projects. Such an approach may also naturally lend itself to inclusion of multi-purpose functionality of the TVE, such as year-round recreation opportunities (e.g., a TVE berm that also serves as a park). Future research on the applicability of this approach may be beneficial to determine broader practical application as coastal communities manage future growth.

7. Conclusions

This case study of tsunami vertical-evacuation siting in Ocean Shores, Washington, focused on the use of GIS-MCDA to support risk-reduction decision making. Based on our analysis, we reach several conclusions that bear on future tsunami risk-reduction research and application to at-risk communities.

• Populations at risk are widely distributed and no one vertical-evacuation refuge can provide universal protection from tsunami hazards.

• Workshop participants significantly underestimated population capacity for many of the proposed TVE sites, indicating the value of incorporating geospatial modeling into future workshops.

• Workshops participants also prioritized TVE sites that had relatively lower value in terms of potential reductions of loss of life, indicating the need for more discussion on TVE priorities.

• The benefits of each vertical evacuation refuge varied depending on which population group was prioritized (e.g., residents versus tourists), as well as the postdisaster availability of bridges.

• GIS-based, multi-criteria decision analysis that includes community workshops provides a useful tool for federal, state, and local decision makers faced with prioritizing scarce resources for tsunami risk reduction.

• Tighter integration of GIS-MCDA into community participatory approaches could provide a more interactive exploration of alternatives, tradeoffs, priorities, and qualities of TVE sites. Future work could be strengthened by a comparison of revealed vs. expressed preferences to better understand implications of community preferences. Ethical considerations warrant additional discussion as TVE site planning matures.

Acknowledgments

This study was supported by the US Geological Survey (USGS) Land Change Science Program. We thank Mara Tongue of the USGS, Peter Howe of Utah State University, and three anonymous reviewers for their insightful reviews of earlier versions of the article. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the US Government.

References

[1] American Health Care Association. Ethical guidelines for the development of emergency plans, unpublished white paper; 2013. Available at <http://www.ahcancal.org>. Accessed 27 November 2013.

[2] Applied Technology Council. Guidelines for design of structures for vertical evacuation from tsunamis. Washington, D.C: Federal Emergency Management Agency Publication P646; 158.

[3] Applied Technology Council. Vertical evacuation from tsunamis—a guide for community officials. Washington, D.C: Federal Emergency Management Agency Publication P646A; 50.

[4] Bojórquez-Tapia LA, Díaz-Mondragón S, Ezcurra E. GIS-based approach for participatory decision-making and land suitability assessment. Int J Geog Inf Sci 2000;15:129-51.

[5] Burby R. Hurricane Katrina and the paradoxes of government disaster policy-bringing about wise governmental decisions for hazardous areas. The Annals of the American Academy of Political and Social Science 2006;64(1):171-91.

[6] Cascadia Region Earthquake Workgroup. Cascadia subduction zone earthquakes—a magnitude 9.0 earthquake scenario. Portland: Oregon Department of Geology and Mineral Industries; 2013.

[7] Carver S. Integrating multi-criteria evaluation with geographical information systems. Int J Geog Inf Sci 1991;5:321-39.

[8] Doughton, S. Grays Harbor County school to build first U.S. vertical-tsunami refuge, Seattle Times; 2013. Available at (http://seattle times.com/html/localnews/2022051420_shakeoutxml.html>. Accessed March 28. 2014.

[9] Eastman JR, Jin W, Kyem PAK, Toledano J. Raster procedures for multi-criteria/multi-objective decisions. Photogramm Eng Remote Sens 1995;61(5):539-47.

10] Engstfeld, A, Killebrew, K, Scott, C, Wiser, J, Freitag, B,El-Anwar, O. Tsunami safe haven project—report for Long Beach, Washington. Department of urban design and planning, college of built environments, University of Washington; 2010.

11] Fraser, S, Leonard, GS, Matsuo, l,Murakami, H. Tsunami evacuation— lessons from the Great East Japan Earthquake and Tsunami of March 11th 2011. Geosciences New Zealand (GNS) science report 2012/17, 89 p; 2012a.

12] Fraser S, Leonard GS, Murakami H, Matsuo l. Tsunami vertical evacuation buildings—lessons for international preparedness following the 2011 Great East Japan Tsunami. J Disaster Res 2012;7: 446-57.

13] Geohazards lnternational. Tsunami preparedness—feasibility study of vertical evacuation structures in Sumatra; 2014. Available via http://geohaz.org/projects/sumatra.html. Accessed March 28, 2014.

14] González F, Geist E, Jaffe B, Kanoglu U, Mofjeld H, Synolakis C, et al. Probabilistic tsunami hazard assessment at Seaside, Oregon, for near- and far-field seismic sources. J Geophys Res 2009;114:C11023.

15] Infogroup. Employer Database; 2011. Available via (http://www. infousagov.com/employer.asp>. Accessed October 1, 2011.

16] Jankowski P. lntegrating geographical information systems and multiple criteria decision making methods. Int J Geog Inf Sci 1995;9: 251-73.

17] Jankowski P, Nyerges T. GlS-supported collaborative decision making—results of an experiment. Ann AAG 2001;91(1):48-70.

18] Jonkmann S, Vrijling J, Vrouwenvelder A. Methods for the estimation of loss of life due to floods: a literature review and a proposal for a new method. Nat Hazard 2008;46:353-89.

19] Leonard, G, Evans, N, Prasetya, G, Saunders, W, Pearse, L, Monastra, D. et al. Scoping study for evaluating the tsunami vulnerability of New Zealand buildings for use as evacuation structures, GNS science report 2011/36 39 p.; 2011.

20] Malczewski J. GIS and multicriteria decision analysis. New York: John Wiley & Sons; 1999.

21] Malczewski J. GIS-based multicriteria decision analysis: a survey of the literature. Int J Geog Inf Sci 2006;20:703-26.

22] National Research Council. Tsunami warning and preparedness: an assessment of the U.S. Tsunami Program and the Nation's preparedness efforts, Committee on the Review of the Tsunami Warning and Forecast System and Overview of the Nation's Tsunami Preparedness. Washington, D.C: The National Academies Press; 2011.

23] National Tsunami Warning Center. Tsunami frequently asked questions, National Oceanic and Atmospheric Administration's National Weather Service; 2014. Available at <http://wcatwc.arh.noaa.gov/? page=tsunamiFAQ>. Accessed March 28, 2014.

24] Park S, van de Lindt J, Gupta R, Cox D. Method to determine the locations of tsunami vertical evacuation shelters. Nat Hazard 2012;63:891-908.

25] Soule R, Goldman R. Terrain coefficients for energy cost prediction. J Appl Physiol 1972;32:706-8.

26] State of Washington, Office of Superintendent of Public lnstruction. State level public school data; 2013. Available at <http://www.k12. wa.us/dataadmin/>. Accessed 5 Dec 2013.

27] Tobler, W. Three presentations on geographical analysis and model-ing—non-isotropic geographic modeling. Speculations on the geometry of geography; and global spatial analysis. UCSB. National Center for Geographic lnformation and analysis technical report 931; 1993. Available at <http://www.ncgia.ucsb.edu/Publications/Tech_ Reports/93/93-1.PDF>. Accessed 19 July 2010.

[28] Travel Weekly. Quinault Beach Resort and Casino; 2013. Available at <http://www.travelweekly.com/Hotels/Ocean-Shores-WA/Quinault-Beach-Resort-Casino-p4154521>. Accessed 5 Dec 2013.

[29] United States Census Bureau. American Factfinder; 2011. Available at http://factflnder2.census.gov/faces/nav/jsf/pages/index.xhtml. Accessed May 1, 2011.

[30] United States Department of Agriculture. Geospatial Data Gateway; 2009. Available at <http://datagateway.nrcs.usda.gov/>. Accessed February 1, 2011.

[31] United States Department of Transportation. Manual on uniform traffic control devices for streets and highways. Fed Highway Adm; 2009.

[32] Velotti L, Trainor J, Engel K, Torres M, Myamoto T. Beyond vertical evacuation—research considerations for a comprehensive vertical protection strategy. Int J Mass Emergencies Disasters 2013;31(1): 60-77.

[33] Walsh T, Caruthers C, Heinitz A, Myers III E, Baptista A, Erdakos G, et al. Tsunami hazard map of the southern Washington coast— modeled tsunami inundation from a Cascadia subduction zone earthquake. Washington department of natural resources division of geology and earth resources geologic Map GM 2000:49.

[34] Washington Department of Ecology (2009) Washington state marine shorelines. Available at <ftp://www.ecy.wa.gov/gis_a/shore/ shore.zip>. Accessed 2 Jan 2011.

[35] WatershedSciences. LIDAR remote sensing data collection; 2010. Available at <http://pugetsoundlidar.ess.washington.edu/lidardata/ restricted/nonpslc/swwash2009/swwash2009.html>. Accessed 7 Aug 2011.

[36] Wood, N, Soulard, C. Variations in community exposure to tsunami hazards on the open-ocean and Strait of Juan de Fuca coasts of Washington. USGS scientific investigations report 2008-5004; 2008.

[37] Wood N, Schmidtlein M. Anisotropic path modeling to assess pedestrian-evacuation potential from Cascadia-related tsunamis in the U.S. Pacific Northwest. Nat Hazard 2012;62(2):275-300.

[38] Wood N, Schmidtlein M. Community variations in population exposure to near-field tsunami hazards as a function of pedestrian travel time to safety. Nat Hazard 2013;65(3):1603-28.

[39] Yeh H, Fiez T, Karon J. A comprehensive tsunami simulator for Long Beach Peninsula phase-1—framework development final report. State of Washington Military Department Emergency Management Division; 2009.