Changes in population

THIS IS A PRE-PRINT VERSION. PLEASE CONTACT NATHAN WOOD ([email protected]) FOR THE PRINTED VERSION.

Title: Changes in population-evacuation potential for tsunami hazards in Seward, Alaska, since the 1964 Good Friday earthquake

Authors: • • •

Nathan Wood, PhD, U.S. Geological Survey Western Geographic Science Center Mathew Schmidtlein, PhD, California State University, Sacramento Jeff Peters, U.S. Geological Survey Western Geographic Science Center

Keywords: Good Friday earthquake, tsunami, evacuation, disaster, modeling

Abstract Pedestrian-evacuation modeling for tsunami hazards typically focuses on current landcover conditions and population distributions. To examine how post-disaster redevelopment may influence the evacuation potential of at-risk populations to future threats, we modeled pedestrian travel times to safety in Seward, Alaska, based on conditions before the 1964 Good Friday earthquake and tsunami disaster and on modern conditions. Anisotropic, path distance modeling is conducted to estimate travel times to safety during the 1964 event and in modern Seward, and results are merged with various population data, including the location and number of residents, employees, public venues, and dependent-care facilities. Results suggest that modeled travel-time estimates conform well to the fatality patterns of the 1964 event and that evacuation travel times have increased in modern Seward due to the relocation and expansion of port and harbor facilities after the disaster. The majority of individuals threatened by tsunamis today in Seward are employee, customer, and tourist populations, rather than residents in their homes. Modern evacuation travel times to safety for the majority of the region are less than wave arrival times for future tectonic tsunamis but greater than arrival times for landslide-related tsunamis. Evacuation travel times will likely be higher in the winter time, when the presence of snow may constrain evacuations to roads.

1 Introduction More than 244,000 people have lost their lives from tsunamis in the Indian and Pacific Ocean basins in the past decade (NOAA National Geophysical Data Center/World Data Center 2013). These staggering losses have heightened global awareness of the threat that tsunamis pose to coastal populations. Of particular concern are coastal communities threatened by tsunami waves that could arrive within minutes of the events that trigger them, such as local earthquakes or landslides. To support tsunami preparedness and evacuation planning, there has been considerable work in recent years to model pedestrian evacuation travel times out of tsunami hazard zones (e.g., Jonkmann et al. 2008; Yeh et al. 2009; Johnstone and Lence 2012; Wood and Schmidtlein 2012; Wood and Schmidtlein 2013). To date, tsunami

1

evacuation research has focused on current landscape conditions and population distributions. We are unaware of any efforts to compare evacuation potential based on pre-disaster conditions to evacuation potential based on conditions after communities have rebuilt or are in the process of reconstruction. This information could identify if and how population vulnerability to tsunamis would or has changed in a community due to post-disaster redevelopment decisions. The objective of this paper is to examine how post-disaster redevelopment after a catastrophic tsunami disaster may have changed the vulnerability of coastal populations to future tsunamis. One tsunami disaster that provides opportunities to reflect on this topic is the Mw 9.2 Good Friday earthquake and tsunami that occurred on March 27, 1964, in the Prince William Sound region of Alaska (U.S. Geological Survey 2012). Tsunami waves associated with the earthquake were measured throughout the Pacific Ocean basin and were responsible for 124 deaths and $119 million in losses ($USD, 1964 dollars), making it the costliest and second deadliest tsunami event in U.S. history (NOAA National Geophysical Data Center/World Data Center 2013). Our case study for the temporal analysis of population exposure to tsunamis is the coastal community of Seward, Alaska (fig. 1), which experienced significant damage and 12 deaths from the 1964 earthquake and tsunami disaster. Extensive documentation of impacts in Seward from the 1964 event (e.g., Lemke 1967; Anderson 1970; Eckel and Schaem 1970; Norton and Haas 1970; Rogers 1970a; Arno and McKinney 1973; Sturman 1973; Barry 1995; Lander 1996) and recent efforts to model future tsunami threats (e.g., Suleimani et al. 2010) present us with an opportunity to examine changes in population vulnerability to tsunamis over time. To do this, we model pedestrianevacuation travel times and estimate population exposure in Seward of residents, employees, and visitors based on conditions before the 1964 disaster and on current conditions. We also compare modeled evacuation times with post-disaster descriptions of the 1964 event to explore model validity. We demonstrate how modeled travel times to safety for various population groups differ across Seward due to changes in land use since 1964, assumptions of travel speed, and evacuation pathways. Our research provides insight on the implications of post-disaster redevelopment on the evacuation potential for future events and how population vulnerability varies within a community.

2 Seward and the 1964 Good Friday earthquake and tsunami Seward is a small port town (2010 population: 2,693, U.S. Census Bureau 2012) on the ice-free shores of Resurrection Bay, Alaska (fig. 1). It was founded in 1903 as the terminus for the Alaska Central Railway to provide access to the interior Alaskan coal fields (Seward Historic Preservation Commission 2012), and served as the primary port for Anchorage and the rest of the Alaskan interior until the early 1960s because of the attractiveness of its ice-free winters (Anderson 1970; Rogers 1970a). Seward was initially built on the alluvial fan of Lowell Creek on the northwestern end of Resurrection Bay and has since expanded into the northern and northeastern portions of the bay. Its location near the seismically active Alaska-Aleutian subduction zone has made it susceptible to past tsunamis (NOAA National Geophysical Data Center/World Data Center 2013). The 1964 Good Friday Mw 9.2 earthquake occurred on March 27 at approximately 5:35 PM local time at the eastern end of the Aleutian Megathrust. Violent shaking lasted three to four minutes and caused large ground ruptures that extended over 50 meters inland and were greater than five meters in depth in some areas (Lemke 1967). Tsunamis were generated both from the initial vertical displacement within the rupture area (referred subsequently as the tectonic tsunami) and by local landslides triggered by the earthquake ground shaking. The first waves to arrive were related to landslides that were generated when portions of the Lowell Creek alluvial fan, which were exposed due to low tide, began to slide into Resurrection Bay 30 to 45 seconds after the start of the initial ground shaking (fig. 2a and 2b; Lander 1996). The landslides led to a rapid drop in the water level along the edge of the fan (Lemke 1967) and generated a series of waves that moved towards the center of the bay (Lander 1996). The multiple waves merged several hundred meters off shore and then spread towards Seward and Lowell Point to the west and Fourth of July Point to the east (fig. 1). Tsunami waves were estimated to be 9 meters high and struck

2

Seward and Lowell Point within approximately 1.5 to 2 minutes (Lander 1996; Lemke 1967), which was only half way through the 3 to 4 minutes of heavy ground shaking from the earthquake. Smaller waves due to lesser submarine landslides, seiche, or subsequent landslide-wave cycles continued after the initial landslide-related tsunami (Lemke 1967; Lander 1996; Suleimani et al. 2010). Tectonic tsunami waves arrived in the northern portion of the bay approximately 25 minutes after the earthquake and reportedly reached wave heights of 12 meters (Lemke 1967; Lander 1996). Waves continued to arrive almost hourly until around 4:20 am the next morning. One of the most destructive waves associated with the tectonic tsunami occurred near high tide at approximately 10:30 pm, five hours after the initial earthquake (Lander 1996). Tectonic subsidence of approximately 1.1 meters also resulted in new areas in Seward being inundated during high tides (Lemke 1967). A full account of damages in Seward has been well documented elsewhere (e.g., Lemke 1967; Anderson 1970; Eckel and Schaem 1970; Norton and Haas 1970; Rogers 1970a; Arno and McKinney 1973; Sturman 1973; Barry 1995; Lander 1996). In summary, the combined effects of the earthquake, landslides, and tsunami waves destroyed approximately 95% of the industrial base and damaged or destroyed 15% of the residential structures. Dock facilities were destroyed, including those for the U.S. Army, Standard Oil, a newly opened barite plant, a halibut cannery, and a small-boat harbor. Railroad docks were almost completely destroyed, along with the loss of cranes, buildings, and railroad lines (fig. 2a and 2b). The city power plant sank into the bay (cutting power generation and delivery to the area) and sewer lines became clogged with sand from the repeated tsunami wave activity. The only radio tower was destroyed and all highway access to Seward was cut off (Lander 1996). Fires broke out due to ignited diesel fuel that had spilled from the Alaska Standard tanker ship, which was refueling at the Standard Oil Company docks on the southeastern shore of Seward at the time of the earthquake (fig. 2a). The sudden drop in water caused the ship to drop and roll towards the bay, causing hoses connected to the oil tanks on land to pull out and diesel fuel to spill. Resulting flames were estimated to reach over 60 meters in the air. Burning fuel was then carried ashore by the tsunami waves, where an eighty-car train was slowly moving from the Standard Oil tanks and dock towards the Texaco tanks to the north (fig. 2a). The burning diesel ignited the petroleum in the last 40 cars of the train, which began exploding sequentially towards the Texaco tanks, which then also caught fire within 3 to 4 minutes (Lemke 1967). Residents feared that the fires would spread and therefore fled north hoping to escape town via a causeway that crossed a lagoon. The causeway had only small amounts of debris from the initial landslide-related tsunamis but became completely blocked by larger debris (e.g., boats, houses) carried in by subsequent waves (Lemke 1967). North of downtown Seward, the bridges across the Resurrection River were destroyed, which effectively trapped Seward residents until an abandoned and overgrown road west of the lagoon was cleared the next day and the causeway cleared three days later (Frank and Haas 1970). Because the earthquake occurred late in the day on a religious holiday, most businesses were closed (Barry 1995, Lander 1996) and fewer people than normal may have been directly exposed to the tsunami. In the end, twelve people in Seward lost their lives due to the tsunamis. Eight of the fatalities were related to the landslide-generated tsunamis that arrived within 1.5 to 2 minutes -- six people were on boats or otherwise visiting the small boat harbor, one person was on the Alaska Standard, and another died of a heart attack as he fled the tsunami (Lander 1996). The remaining four fatalities were from the tectonic tsunami that arrived 25 minutes after the earthquake - two individuals died on the causeway and the other two were a pair of seal hunters who had been on the bay when the earthquake occurred (fig. 2b). In addition to the fatalities, the Red Cross listed 10 serious injuries and 190 minor injuries. A total of 85 were hospitalized, and hundreds suffered from exposure to the cold and snow (Lander 1996). After the 1964 disaster, the community of Seward took several steps to reduce its exposure to future earthquakes and tsunamis. The alluvial fan shoreline was set aside as an open space after an engineering study determined that the eastern edge of the fan had a high likelihood of future failure from earthquakes (Eckel and Schaem 1970). The new small-boat harbor, city dock, railroad docks, and deep-water port were built north of the fan

3

on ground that was considered less likely to fail during future events (Arno and McKinney 1973; fig. 2c). The move also provided the city with the ability to build a larger small-boat harbor, because the previous small harbor had limited the city’s economic development. Material removed from the new small boat harbor was also used to fill the salt water lagoon east of the causeway (Arno and McKinney 1973), which was then developed primarily for commercial means (fig. 2c). The Port of Seward and small boat harbor still operate today, but no longer represent the primary shipping location for the interior of Alaska. The destruction of the Seward harbor facilities and the disruption of the rail lines to Anchorage during the 1964 disaster forced ships to use the Port of Anchorage (Rogers 1970a), which officially commenced operations in 1961 (Port of Alaska 2013). By the time the new docks in Seward opened in 1966, petroleum handling and general shipping capacity in the Port of Anchorage had already been drastically increased and it replaced Seward as the primary shipping location for the interior of Alaska (Rogers 1970b; 1970a).

3 Pedestrian evacuation modeling We examine historic and current population vulnerability to tsunami hazards in Seward using a least-costdistance (LCD) approach to model pedestrian travel times to safety and then link results to various population data. This geospatial approach focuses on characteristics of the evacuation landscape, such as slope 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 in terms of increased travel time. We chose a LCD approach over an agent-based approach (e.g., Yeh at al. 2009) because the exact distribution of at-risk people in 1964 and today is not known and this is a critical element to modeling at-risk individuals as distinct agents on the hazard landscape. The following sections provide additional information on the various input data and geospatial analytical methods. The specific LCD method used in this evacuation analysis is based on the anisotropic, path distance approach described in Wood and Schmidtlein (2012; 2013) and implemented in ESRI’s ArcMap 10.1 geographic information system (GIS) software using its Path Distance tool. Three major data sets are required as input for the pedestrian-evacuation modeling, including hazard zones (or actual inundation areas in the case of the 1964 event), land cover, and elevation. Land cover and elevation derived slope data are transformed into speed conservation values (SCV) and represent the proportion of maximum travel speeds that are expected on areas with given conditions. Land cover SCVs are based on Soule and Goldman’s (1972) energy-cost terrain coefficients for certain land cover types and slope SCVs are based on Tobler’s (1993) hiking function. 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) and the path distance tool calculates distances and slopes between cells of varying elevations. Path-distance modeling estimates travel times based on optimal routes but not preferred routes by at-risk individuals. Actual travel times therefore may be longer than modeled travel times given the perceptions and preferences of evacuees. Additional factors that may increase travel times are environmental conditions at the time of an evacuation (e.g., inclement weather, nighttime) and impacts to evacuation routes due to other seismic hazards (e.g., ground shaking, ground rupture, lateral spread, liquefaction, rubble of damaged structures from ground shaking). Two sets of anisotropic LCD pedestrian evacuation models were developed for Seward. The first analysis is based on the populations and landcover conditions prior to the 1964 Good Friday earthquake and tsunami, and the second is based on modern populations and conditions. For the comparative analysis between pre-1964 and modern conditions (section 4), we focused on an abridged study of downtown Seward (fig. 1) because of limitations in historic imagery. We include all of Resurrection Bay when only discussing modern conditions (section 5) due to greater availability of imagery, elevation, and population data. Within the analysis of modern conditions, we developed one scenario assuming open travel across the landscape (modified by SCV) and a second scenario that confined people’s travel to roads, which would better approximate an evacuation during winter months when

4

substantial snow drifts would constrain travel. A roads-only evacuation was modeled by calculating travel times from population points to the nearest road and then along a road network to safe areas (discussed in greater detail in Wood and Schmidtlein 2012). Two sets of tsunami zones are used in this study (Suleimani et al. 2010). The first inundation zone summarizes the reported maximum inundation area for the landslide and tectonic tsunamis associated with the 1964 Good Friday earthquake (shown in figures 2a and 2b). The second set includes modeled inundation areas from a variety of sources (fig. 2c). Inundation modeling was based on a series of embedded digital elevation models, the finest being a 15-meter grid for the upper Resurrection Bay, and corresponding to Mean High Water (MHW). Maximum tsunami inundation in this area (yellow line in figure 2c) reflects a composite of four tectonic tsunami scenarios and three landslide tsunami scenarios. Tectonic tsunami scenarios include a source function of the 1964 tsunami, two modified 1964 events that include geologic asperities in Prince William Sound or Kodiak, and a hypothetical rupture of the Pamplona zone. Arrival times for the various tectonic tsunami sources vary and will depend on the distribution of slip in the rupture area. Because this slip distribution is difficult to predict, arrival times of tectonic tsunami waves are also difficult to predict. For discussion purposes only, we assume tectonic tsunami waves would likely arrive 25 to 30 minutes after a large earthquake (Suleimani et al. 2010). Landslide tsunami scenarios include (1) a repeat of the 1964 landslides of downtown Seward, Lowell Point, and Fourth of July Point, (2) hypothetical events associated with recent sediment accumulation from rerouted Lowell Creek and Fourth of July Creek, and (3) a similar hypothetical event but with added sediment volumes. We also include in Figure 2c the modeled inundation areas that approximate a repeat of landslide-related tsunamis during the 1964 event (scenario 5 in Suleimani et al. 2010). More information on each scenario and modeling assumptions can be found in Suleimani et al. (2010). We include this level of detail in the description of the modern, maximum-inundation zone for two reasons. First, the maximum zone represents a composite of multiple scenarios; therefore, all delineated areas may not be inundated by a single future tsunami. Second, Seward continues to be threatened by landslide-related tsunamis that could arrive in a matter of minutes, without sufficient time for official warnings. Onshore and submarine landslides can be expected as a result of large earthquakes in the future and their potential has been exacerbated by continuing sediment accumulation on underwater slopes of Resurrection Bay and by deposition in new areas due to rerouting of Lowell Creek and Fourth of July creek (Suleimani et al. 2010). The primary distinction in modeled inundation between the composite maximum zone and the landslide scenario in Figure 2c is that the commercial and industrial portions of Seward may evade inundation from landslide-related tsunamis immediately after an earthquake, but not later from the tectonic tsunamis (fig. 2c). In general, as we discuss population exposure to tsunamis in Seward, it is important to remember the potential for tectonic tsunami waves that may arrive in tens of minutes and landsliderelated tsunamis that could arrive in a matter of minutes. Elevation data for both the 1964 and modern evacuation models was based on a 2009 5-foot, LiDARderived digital elevation model of the area (Kenai Peninsula Borough GIS Division 2013). Additional processing was required to modify this data to represent the elevation for the portions of the 1964 shoreline that slid into the bay during the earthquake. Because we lacked elevation data prior to the earthquake, we simply assumed a flat area to the additional pre-1964 shoreline and an abrupt transition to the bay. Because the areas lost during the 1964 event were most likely graded for industrial uses, a flat slope for this area could be considered a good approximation. The approximate elevation of the lost shoreline was created by manually digitizing a series of points just inland from the modern shoreline and using zonal operations to identify their elevations. To extend these elevations outwards, an allocation surface was created, where the value for each cell represented the elevation of the sample point closest to it. This elevation allocation surface was then clipped to the edge of the shoreline identified from the 1964 pre-event imagery. These values were incorporated into the final elevation surface only for those areas between the modern and 1964 shorelines.

5

Land cover data for areas in the two tsunami-hazard zones were derived from the integration of several datasets (fig. 3). Current landcover in Seward (fig. 3b) was derived from a supervised and manual classification of 2005, 1-meter resolution, RGB-band orthorectified IKONOS imagery (Kenai Peninsula Borough GIS Division 2013) in ERDAS IMAGINE and then verified using Google Maps and Google Street View imagery. Additional imagery, including 2008 2-foot data and 2005 1-foot data (Kenai Peninsula Borough GIS Division, 2013), was used to deal with clouds and shadows in the northern part of Resurrection Bay in the 2005 IKONOS data. Land cover was classified into various categories and assigned SCV derived as the inverse of terrain energy coefficients discussed in Soule and Goldman’s (1972). Landcover classes (and associated SCV) include impervious surfaces (1.0), grass and dirt/gravel surfaces (0.9091), light brush (0.8333), heavy brush (0.6667), wetlands, sand, and shoreline (0.5556), and water and human obstacles, such as buildings and fences (0.0). Values fall on a spectrum from zero (meaning travel is not possible) to 1.0 (meaning pedestrian evacuation speed is 100% of the base travel rate). Fence lines that could impede pedestrian evacuation (i.e., a SCV of 0.0) around particular land-use areas, such as a prison, schools, baseball fields, etc., were identified using Google Street View and also incorporated into the 5-ft land cover dataset. The pre-1964 land cover data (fig. 3a) was created by manual interpreting a georectified black and white aerial image dated May 29, 1963 (Lemke 1967; U.S. Army Corps of Engineers, unpublished data) and comparisons to the classification based on 2005 multi-spectral imagery. Classified land cover data were then resampled to match the 5ft resolution elevation data. A comparison of land cover results for 1963 (fig. 3a) and 2005 (fig. 3b) demonstrate the increased development to the north of Seward that had previously been grass, brush, and gravel shores, as well as the absence of docks and other development to the east of Seward. Cost surfaces that integrate land cover and slope variability are converted to maps of pedestrian travel times using various travel-speed assumptions, including a slow walk (1.1 m/s; U.S. Department of Transportation 2009) and a slow run (1.79 m/s; MarathonGuide.com 2011). A slow-walk travel assumption is typically preferred given a mixed population with ranges in age and physical mobility (Wood and Schmidtlein 2012). However, for our comparative analysis of pre-1964 and modern conditions, we assume at-risk individuals could attain a slow running speed for several reasons. First, demographic data characterizing age profiles for Seward residents in 1960 and 2010 (U.S. Census Bureau 1963, 2012) suggest that the majority of residents in the study area were between 15 and 64 years in age (fig. 4) and therefore more likely to be capable of running. Second, the tsunami-hazard zones during both time periods largely impact industrial waterfronts (fig. 2), where maritime employees are likely to be in good physical shape and capable of greater travel speeds. Third, at-risk populations today would likely attain faster speeds given the 1964 experience, knowledge of other recent tsunami disasters throughout the world (e.g., 2011 Tohoku, 2010 Chilean), current tsunami-education efforts in Alaska, and the relatively short distance to safety. The final input data involved estimating the distribution of populations before the 1964 event and of populations that are at risk today from future tsunamis. Estimates of residential distributions in 1964 were generated by disaggregating population counts from the 1960 Census to residences manually interpreted from the 1963 imagery (Lemke 1967; U.S. Army Corps of Engineers, unpublished data). This dasymetric interpretation involved dividing the 1960 Seward city population of 1,891 residents (U.S. Census Bureau 1963) by the total number of residence points that fell within the 1964 city boundaries (personal communication, Donna Glenz, City of Seward planner, July 11, 2012). The integer portion of the resulting quotient was assigned to each residential point within the city, and one additional resident was assigned at random to a total number of points matching the remainder. The mode for population per residential unit from within the city was four and was assigned to all residential points in the study area that fell outside the city boundaries. This results in a total estimated pre-1964 residential population of 2,395 people within the abridged study area of the City of Seward and adjacent unincorporated land. As noted earlier, the area for which residential population distributions could be estimated in 1964 was limited due to imagery limitations to the portion of Seward on the alluvial fan and extending to the southern portion of the Clearview subdivision to the north of the lagoon (figures 1 and 2a).

6

Residential-population distributions in the current tsunami-hazard zone were estimated by using similar dasymetric mapping techniques but with higher-quality geospatial data, including 2010 U.S. Census Bureau blocklevel population counts (U.S. Census Bureau 2012), tax parcel data (Kenai Peninsula Borough GIS Division 2013), physical address points and roads (Kenai Peninsula Borough GIS Division 2013). We identified parcels in the tsunami-hazard zone with resident-related usage codes in the Kenai Peninsula Borough Parcel data, including residential dwellings (both single and 2-4 inhabitants), residential cabin (both single and 2-4 inhabitants), residential mobile home (both single and 2-4 inhabitants), condominium, commercial apartments and mobile-home parks, and senior apartments and housing. All physical address points that fell within these parcels were then selected and compared to Census blocks to ensure that there was at least one address within each populated block. For populated census blocks with no residential address locations, we placed points at the locations deemed most likely to have residential populations based on Google Street View imagery. For example, residential population points were placed on hotels, based on the assumption that the documented residents in census block may represent hotel owners or managers. In addition, an adjustment was made to properly locate the group-quarters population of the block containing Spring Creek Correctional Center (U.S. Census Bureau 2012) to the physical address point which represented it. Census block identification numbers and populations were associated with the physical address locations using a spatial join, and the total population for each block was divided by the number of physical address points that fell within it. Because these calculations resulted in population fractions at every point, we assigned the integer portion of the quotient for each point in a block and added populations selected at random from within the block up to the amount of the remainder portion of that quotient. Employee and customer/visitor distributions were estimated using a spring 2011 version of the Infogroup Employer Database (Infogroup 2011), a proprietary database that includes business locations (latitude and longitude coordinates), employee counts, and type based on the North American Industrial Classification System (NAICS). Street addresses for businesses with only U.S. Post Office boxes were identified using Internet searches. NAICS codes were used to identify businesses that are likely to have substantial customer or visitor populations, including community support businesses (banks or credit unions, civil or social organizations, gas stations, government offices, grocery stores, libraries, and religious organizations), dependent care facilities (child services, elderly services, medical centers, offices of physicians or other medical personnel, and K-12 schools), and public venues (museums, overnight accommodations, and parks or other outdoor venues). The various population data were merged with modeled pedestrian-evacuation times to assess population distributions as a function of travel time to safety. This was done by using a simple spatial join to assign evacuation times to the residential, employee, and business population locations. This data was then used to compare population-evacuation times between the various evacuation scenarios. Because of the highly dynamic nature of populations in terms of magnitudes and locations, we do not attempt to explicitly distinguish between night and day or day of the week. We simply report totals for the various population groups. Further site-specific work by local emergency managers would provide better insight on temporal trends and patterns for various populations types. 4 Historic and current evacuation times in downtown Seward Modeled evacuation times for pre-1964 conditions suggest that travel times to safety may have ranged from 0 to 9 minutes in the abridged study area of historic Seward, assuming a slow running speed (fig. 5a). The traveltime map reflecting 1964 conditions (fig. 5a) also includes estimated locations of fatalities based on historic accounts in Lemke (1967) and Lander (1996). Travel times in areas where deaths likely occurred are estimated to be on the order of 2 to 4 minutes and up to 5 to 9 minutes for individuals on the docks or breakwaters of the small-boat harbor on the northeastern shore of the community (fig. 5a). As noted earlier, landslide-related tsunami waves struck the Seward coastline approximately 1.5 to 2 minutes after slope failures of the Lowell Creek alluvial fan from the initial earthquake (Lander 1996; Lemke 1967). Therefore, people would not have had enough time to fully evacuate the inundation area prior to the arrival of the landslide-related tsunami waves. They also would have had to evacuate while ground shaking was still occurring, which was reported to last for four minutes, and across areas with ground

7

fractures and sand boils due to liquefaction (Lemke 1967). In general, individuals working in the rail yards may have had travel times of less than 2 minutes to high ground, but individuals on the docks prior to the earthquake would not have had enough to reach high ground before tsunami inundation. Although no detailed information describing actual evacuation times from various locations during the 1964 event were identified, two main pieces of information suggest that the model results are consistent with descriptions of post-disaster impacts. First, the position and activity of those who died supports the general finding that evacuation times were longer in the more industrial and commercial portions of Seward. All eight of the twelve fatalities that occurred as a result of the initial landslide tsunamis were located either at the docks, the small boat harbor, or the old rail yard near the shore. All of those who died were either employed in these areas, or otherwise engaged in maritime activities. Second, the location that saw the highest total number of deaths, the small boat harbor, was also the location where the longest evacuation times in the 1964 model area were observed. Although this information does not allow us to directly confirm the evacuation times, they are at least consistent with an understanding based on the evacuation modeling. Modeled evacuation times for the abridged study area based on current landscape conditions suggest that most residents in the maximum tsunami hazard zone likely would require less than 2 minutes to reach safety from their homes, again assuming a slow-running speed (fig. 5b). The tsunami exposure of residents in the abridged study area varies only slightly when one compares at-risk populations in 1960 and 2010 as a function of travel time to safety (fig. 6). The number of residents in the tsunami hazard zones of just the Seward community (fig. 5) has decreased from 256 to 198 between 1960 and 2010. For both time periods, 81% of the at-risk residents are in areas that may require less than 2 minutes running to high ground. The remaining 19% of the residents in the tsunamihazard zones (25 residents) may require 3 to 4 minutes, which may not be enough time for landslide-related tsunamis that arrive in less than two minutes. The primary shift in residential exposure between 1960 and 2010 is a decrease in the number of residents that would take less than 1 minute to run to safety and an increase in the number of people that would require 1 to 2 minutes. However, the number of individuals requiring more than two minutes of travel time to safety (the threshold for landslide-related tsunamis) only decreased by three residents (28 residents before 1964 and 25 residents based on 2010 estimates). Since the large area along the shoreline of the alluvial fan was set aside as open space after the 1964 event, we would expect to see a decrease in the total residential population for the abridged study area comparing 1964 and current conditions. Total estimated residential population within the hazard zone of the limited Seward study area (fig. 2) did indeed decrease from 256 in 1960 to 198 based on 2010 population data. But this was coupled with a much more dramatic decrease in the overall resident population within the 1964 study area from estimated 2,391 in 1964 to an estimated 1,426 in the modern scenario. This means that while the total resident population in the hazard zone decreased over the past 50 years, the percentage of study area population located within the hazard zone actually slightly increased, from 11% to 14%. Confidence in these results should be tempered by the fact that the comparison of the percent of populations in the hazard zone is limited by the area for which 1964 data was available. This means that the study area populations in both scenarios, but particularly in the modern scenario, are below the actual broader area populations. Residential population outside this area has expanded in more recent times, meaning that while residential populations may have decreased in the historic heart of Seward, they are much larger elsewhere. For example, in 1964, Seward residents include three subdivisions outside of city boundaries, including Clearview, Forest Acres, and Crawford (fig. 1). Both Clearview and Forest Acres are located to the west of the Seward Highway north of the lagoon, are outside of the tsunami-hazard zone. We conclude that the population has expanded in these areas based on the edges of the Clearview subdivision seen in the pre-1964 event imagery, photos take in Forest Acres following the 1964 event, and modern imagery. The Crawford subdivision, however, which was east of the southern end of the current airport runway, was completely destroyed during the 1964 tsunami and was never rebuilt. In addition, Lowell Point south of Seward (fig. 1) had little to no development prior to 1964 but

8

experienced significant residential development in the decades to follow. Therefore, the elimination of the Crawford subdivision, the increased population around Clearview and Forest Acres neighborhoods, the decreased population in the historic portion of Seward on the alluvial fan, and recent development in Lowell Point may have actually led to a decrease in the proportion of the population in the area in the tsunami-hazard zone. In other words, it is possible that if population expanded following the same distributions as existed in 1964, then we would have seen dramatically higher population exposure in the modern scenario had the changes following the 1964 event not been implemented. Although the residential populations likely have low travel times in modern Seward, commercial and industrial areas constructed after the 1964 tsunami have travel times of 2 to 4 minutes along the commercial strip that occupies the old causeway and lagoon, 5 to 9 minutes along the docks and cargo are of the port, and up to 10 to 18 minutes at the end of port docks (fig. 5b). Modeled inundation areas summarized in Suleimani et al. (2010) suggest that much of the commercial and industrial zones may not be inundated from landslide-related tsunamis (orange line in figure 2c) and tectonic tsunami waves would likely arrive in 25 to 30 minutes after a large earthquake (Suleimani et al. 2010). If this is true, then modeled travel times in our study suggest that most individuals in the commercial and industrial zones may be safe from landslide-related tsunamis (except for individuals on the docks on the small boat harbor) and may have sufficient time to evacuate before tectonic tsunami waves arrive. This, however, assumes that landslide-related tsunami inundation does not exceed the model scenarios and that at-risk individuals would recognize the natural cues of an imminent tectonic tsunami, realize that they are in a tsunamihazard zone, understand what to do to reach safer high ground, and would be able to move quickly and efficiently. Another unknown is the degree to which evacuations may be impaired due to liquefaction of the land under the new commercial and industrial areas, since some of it was created from materials excavated when the new small-boat harbor was developed (Arno and McKinney 1973). A third unknown is the ability of at-risk populations to immediately evacuate, given the potential for long-duration ground shaking (approximately four minutes during the 1964 event) and the possibility of individuals deciding to “duck, cover, and hold” until ground shaking ends. The lack of consistent employee or business location data prior to the 1964 event prevents a direct comparison of hazard exposure for these two population groups. However, the relocation of port and harbor facilities to the north of the alluvial fan in response to the 1964 event has increased population vulnerability if one looks simply at the landscape. As noted earlier, new facilities were moved north because it was determined that the new site contained more stable soils that were less likely to fail from future earthquakes and related landslides (Sturman 1973). While this area may be more stable to future events, tsunami inundation extends further inland here and creates longer evacuation times for the industrial zones if one compares pre-1964 and current evacuation travel times (fig. 5). By moving the docks and harbor to this area, employees, store customers, and people in the small-boat harbor now have to travel a greater distance to reach safety than they would have had these facilities been rebuilt near their initial locations. Due to greater land availability, the relocation of port and harbor facilities after the 1964 event also represented an opportunity to expand the size of what had already become an undersized small boat harbor, which results in greater numbers of people potentially in harm’s way from future tsunamis. Therefore, evacuation times at the new small-boat harbor and related commercial areas have increased because of increased distances from the waterfront to high ground. This means that while the re-built infrastructure may be more resilient to future earthquake and landslide damages, the workers and store customers in these areas may now require more time to successfully evacuate than they would have at the comparable locations of the pre-1964 harbor and dock facilities. Although evacuation times from these newer commercial and industrial areas are still on the order of minutes, they are as much as 5 times greater than the estimated 1.5 to 2 minute arrival time of a landslide tsunami. And this is in the small-boat harbor, the area that is the descendent of the location with the highest concentration of tsunami deaths during the 1964 event. In other areas of the airport runway and port facilities, evacuation times are greater than 20

9

minutes (fig. 7), suggesting individuals here may have difficultly escaping inundation from tsunami generated by both landslide and tectonic sources. Land-use decisions in the community of Seward in the aftermath of the 1964 Good Friday earthquake and tsunami reflect a desire to decrease potential losses or impacts to infrastructure and other development from future local earthquakes and resulting tsunamis. These decisions are consistent with modern concepts of sustainable development and community resilience to natural hazards. However, although the relocation of these key port facilities may result in reduced infrastructure losses from future events, they may have increased evacuation times for those who work there because of the size and location of the redevelopment. 5 Modern evacuation landscapes around Resurrection Bay Modeling results for all land in the modern tsunami-hazard zone of Resurrection Bay indicate that the longest travel times to safety are in the Port of Seward and just south of Fourth of July Point (fig. 7). Because of the mixed population across the entire study area, we show results assuming a slow walking speed (figures 7a and 7b) and a slow running speed (figures 7c and 7d). Within these travel-speed assumptions, we also modeled travel times based on open travel across the landscape (figures 7a and 7c) and on travel constrained to road networks (fig. 7b and 7d) to simulate winter conditions. For the majority of the study area, model travel times to safety are on the order of 0 to 9 minutes if we assume non-winter conditions and a slow-walking travel speed (yellow and orange zones in figures 7a and 7c). Travel times increase to 10 to 24 minutes in the Port of Seward and in the residential neighborhood north of the Resurrection River (fig. 7a). When we assume slow-running speeds in the models, these longer travel times are limited to only the southern reaches of the port docks and the airport runway (fig. 7c). Therefore, if we assume that at-risk individuals would be able to travel across the landscape (i.e., not constrained to roads), then travel times in the majority of the tsunami-hazard zones would be less than ten minutes, which is less than the likely 25 minutes for tectonic tsunami waves to arrive. However, if landslide-generated tsunamis were to occur again, then certain portions of the study area could be inundated before people would have sufficient time to evacuate. If travel is restricted to road networks, then modeled travel times increase substantially. The areas most impacted by this restriction are again the Port of Seward and the residential neighborhood to the north of the Resurrection River (figures 7b and 7d). This restriction causes modeled travel times (assuming a slow walking speed) to increase to greater than 25 minutes in several areas of the Port of Seward, north of Resurrection River, and south of Fourth of July Point (fig. 7b). If travel speeds are increased to a slow run, then evacuation times obviously decrease but are still high for portions of the commercial and industrial areas of the Port of Seward. In these areas, evacuations could still take over 20 minutes to reach safety. This suggests that evacuations across the landscape, rather than constrained to roads, may be more effective when conditions are feasible. It may also suggest that the population in Seward may be more vulnerable to tsunami inundation in the winter, when evacuation across the landscape may be more constrained due to the presence of snow. Modeled evacuation times are very similar, both in terms of magnitudes and spatial distributions, for evacuation models that assume open travel and a slow walking speed (fig. 7a) and constrained travel and a slow running speed (fig. 7d). In both cases, the majority of the study area is characterized by travel times of 0 to 9 minutes, except for the commercial and industrial segments of Seward where travel times increase to 10 to 20 minutes. This suggests that walking across yards and empty lots may produce the same end result as running along roads to reach safety. An overlay of residential points and modeled evacuation results suggests that there are approximately 333 residents in the modern tsunami-hazard zone, which represents 11% of the residents in the study area. Forty percent of the at-risk residents are outside of the downtown Seward area described in the previous section. All of the 333

10

residents are within 5 to 8 minutes of safety (assuming slow running and slow walking speeds, respectively) if they are able to cross the landscape and are not constrained to road travel (fig. 8a). If residents are constrained to roads, then there are residents that could require up to 15 minutes to reach safety (fig. 8b). It may be the case that certain evacuees will prefer evacuations along road ways under certain conditions, such as ease of travel, lack of familiarity with the area, or avoiding private property. Results suggest that at-risk populations may want to consider off-road evacuation strategies when appropriate in the event of a tsunami. As noted earlier with modeled evacuation times, the distribution of at-risk residents as a function of travel time to safety are very similar assuming slow walking speeds over an open landscape and slow running speeds constrained to roads. To assess the composition of the at-risk population in Seward, we merged modeled evacuation times with 2010 economic data that identifies the location, type, and employee counts of businesses. In addition to the 333 residents reported earlier based on U.S. Census Bureau decadal data, the 2010 business data indicates that the modern tsunami-hazard zone contains 1,146 employees, which represents 48% of the workforce in the study area. Because of the small size of the community, it is likely that there is overlap between these two populations and that they are not exclusive to each other. The tsunami-hazard zone also includes one bank, 15 public venues, seven government offices, nine overnight lodging businesses, 21 retail stores, one school, and two social-services organizations (fig. 9). The number of visitors at each site is not known. For the sake of discussion, we assume that there at least 20 customers at each of these locations. This is likely to be low for the school, but reasonable for the other sites. This assumption results in an additional 1,120 people in the tsunami-hazard zone. Again, the number of customers in the tsunami-hazard zone is not an exclusive group to the number of residents, as many of the customers at local stores are also residents that may live in the hazard zone. Although the three population categories (i.e., residents, employees, and customers) are not necessarily mutually exclusive, comparing the percentages of the total exposed population identified in each group helps to develop a clearer idea of the nature of the exposed population. If we assume the residents in the tsunami-hazard zone also work there, frequent stores, and attend the local school (a likely scenario), then we can subtract the 333 residents from both the 1,146 employees and the 1,120 customers. This results in an at-risk population of 333 residents, 813 other employees, 787 other customers, and a combined estimate of 1,933 people in the tsunamihazard zone. This translates to 17% are residents, 42% are other employees, and 41% are other customers or schoolchildren, suggesting that 83% of the at-risk population in Seward may not live in the tsunami-hazard zone, which is often where people root their hazard knowledge and plans for preparing for and responding to future events. If we assume the residential population is distinct from the other populations, then the total at-risk population rises to 2,599 people, of which 13% are residents, 44% are employees and 43% are schoolchildren or customers at local stores and hotels. And again, the majority of the at-risk population (87%) would not be at their homes in a future disaster. Based on an assumption of slow walking speeds, the majority of the at-risk population in our study area would require ten minutes or less to reach safety (fig. 9a and 9b). We chose to only portray these results assuming a slow walking speed because of the diverse nature of the at-risk populations (Wood and Schmidtlein 2013). As noted earlier, we would assume residents and employees along the waterfront could and would likely move much faster. However, a slow walking speed may be appropriate for school children, tourists staying at hotels and unfamiliar with their surroundings, and customers at stores who may know the area but have mentally prepared based on their home location. With these caveats in mind, the at-risk population may have sufficient time to evacuate in the event of future tectonic tsunamis that could arrive twenty-five minutes after generation. As noted earlier, all residents likely could evacuate in 8 minutes or less. The majority of employees have similar evacuation times, although there are 84 employees that may take up to 12 minutes to evacuate. All of the customer/visitor locations are in areas that would require ten minutes or less to evacuate. If landslide-generated tsunamis were to arrive within two minutes of generation (similar to those experienced during the 1964 event), then there would not be enough time to evacuate for

11

45% of employees, 41% of residents, 73% of public venues, 43% of government offices, 66% of hotels, and 71% of the retail stores. Only the bank, school, and social-service organizations may be able to evacuate in two minutes or less (fig. 9b). But it should be noted that this only reflects the time it would take to move from the location of the building to safety, and not the time it may take to evacuate the structure itself, which may result in much longer evacuation times. Estimates of population exposure as a function of travel time to safety assumes that everyone in the hazard zone will either be at the center of their home, their place of work, or at the store, because population data are georeferenced to point locations. However, model results suggest that many areas throughout the study area may require more than ten minutes (shown as red and black in figure 7). Approximately 86% of the study area has model evacuation results of ten minutes or less to safety if one assumes a slow-walking speed (fig. 9c). Fourteen percent of the study area, or 0.03 sq km, has evacuation times greater than 10 minutes. Areas with 25 to 30 minutes of travel time to safety (considered to be the amount of time before seismically-generated tsunamis arrive) comprise one percent of the total land in the tsunami-hazard zone and are located at the southern end of the airport runway at the Port of Seward. Therefore, although most at-risk populations would likely have enough time to evacuate future tsunamis from most locations in the study area, there are some remote areas where evacuations may be challenging. Studies to determine potential transient populations in these more-remote areas may be warranted, as would discussions of vertical-evacuation strategies if transient population magnitudes are high. One such area with relatively longer evacuation travel times and high transient populations is the Port of Seward section that accommodates cruise ships. In figure 5b, a cruise ship is clearly docked in the lower right corner of the 2005 image where modeled evacuation travel times are 17 minutes, assuming a slow walking speed. On many days (primarily during summer months), there can be several thousand tourists along the waterfront that are associated with cruise ships that either depart from Seward or include it on an itinerary. For example, on multiple days in 2013, one vessel departed from the Port of Seward with a capacity for 2,502 passengers and 859 crew members (Seward Chamber of Commerce 2013). Because of the highly dynamic spatial nature of these populations in the Port of Seward and nearby shopping areas, we did not explicitly include them in our population-exposure analysis. However, if a tsunami were to occur when this or other cruise ships are in port, then there could be thousands of tourists in areas that would need to evacuate but may be unfamiliar with their surroundings. Therefore, population vulnerability to tsunamis is heightened in summer months due to the high number and potentially high evacuation travel times for tourists boarding cruise ships or shopping nearby while on shore leave. A related topic for additional research for the entire region is to identify the preferred or likely routes of evacuees based on their perception of the landscape (which likely varies between residents and cruise ship tourists) and potential environmental conditions (e.g., inclement weather), as opposed to evacuation planning based solely on optimal routes.

Conclusions The goal of this paper was to determine if and how population vulnerability to tsunamis in Seward has changed in the aftermath of the 1964 Good Friday earthquake and tsunami disaster. In addition, we sought to compare estimated modeled pedestrian evacuation times to approximate the actual events of the 1964 Good Friday tsunami in Seward. With the 50th anniversary approaching for the 1964 disaster, we believe our analysis will help public officials understand the implications of post-disaster redevelopment on future evacuation potential, as well as how vulnerability to tsunamis varies among sub-populations within a coastal community. Based on our analysis, we reach several conclusions that bear on future tsunami-related population exposure and evacuation studies: •

Modeled travel-time estimates conform well to the fatality patterns of the 1964 event, in that the majority of fatalities were in the small boat harbor and industrial docks, which had longest modeled evacuation times;

12

•

•

•

•

Seward’s population vulnerability to tsunami has increased since 1964 due to the relocation and expansion of the industrial, commercial, and tourism zones, which led to an increase in exposed populations and travel times to high ground; Modern evacuation travel times to safety for the majority of the potential tsunami-inundation areas are less than wave arrival times for the tectonic tsunami scenarios considered in Suleimani et al. (2010) but not for landsliderelated tsunamis; There is a higher relative vulnerability to tsunamis in modern Seward among the employee, customer, and tourist populations than that among the residential population, in terms of higher numbers and longer travel times to safety, citing a need for tsunami outreach focused on businesses; and Seward will likely have longer evacuation travel times during winter months, when the presence of snow may constrain evacuations to roads.

This study demonstrates the utility of modeling pedestrian-evacuation potential based on landscape conditions before and after actual tsunami disasters. Evacuation modeling based on hazard zones before an event can identify areas where targeted preparedness or vertical-evacuation strategies may be warranted. If done immediately after a disaster, pedestrian-evacuation modeling could provide insight on the implications of various land-use or mitigation strategies being considered during the recovery phase. First, linking the locations of fatalities with evacuationmodeling results may provide context for the number and locations of casualties, as well as the potential roles of additional evacuation training in some areas and vertical-evacuation strategies in other areas. Second, evacuation modeling may help identify additional areas for considering land-use changes. Areas that had loss of life are immediately apparent to decision makers and discussions of potential land-use changes may naturally gravitate to these places. Modeling may, however, indicate other areas where evacuations also would not have been successful, but luckily lacked a population presence during the last disaster. These are the areas that may not attract as much attention in the aftermath of an event where decision makers also may wish to make post-disaster adjustments to their communities. Third, outside of the impacted area, this kind of information provides lessons learned that are transferable to other coastal communities that may be threatened by local tsunamis. Given the number of catastrophic tsunami disasters in recent years and the substantial number of communities that are threatened by tsunami hazards throughout the world, there is much to be learned by analyzing past events and sharing experiences. Acknowledgements This study was supported by the United States Geological Survey (USGS) National Geospatial Program and the USGS Land Change Science Program. Mara Tongue, Susan Benjamin, and Keith Kirk of the USGS and two anonymous reviewers gave insightful reviews of the manuscript. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the US Government. References Anderson W (1970) Disaster and organizational change in Anchorage. In: The Great Alaska Earthquake of 1964— Human Ecology. National Academy of Sciences, Washington, D.C. 96-115 Arno N, McKinney L (1973) Harbor and waterfront facilities. In: The Great Alaska Earthquake of 1964— Engineering. National Academy of Sciences, Washington, D.C. 526-643 Barry M (1995) Seward, Alaska—a history of the gateway city. Vol. 3: growth, tragedy, recovery, adaptation, 19241993. Anchorage, Alaska: M.J.P. Barry Eckel E, Schaem W (1970) The work of the scientific and engineering task force. In: The Great Alaska Earthquake of 1964—Human Ecology. National Academy of Sciences, Washington, D.C. 168-182

13

Norton F, Haas J (1970) The cities and towns—Anchorage, Kodiak, Whittier, Seward, Seldovia, Cordova, Valdez. In The Great Alaska Earthquake of 1964—Human Ecology. National Academy of Sciences, Washington, D.C. 248356. Johnstone W, Lence B (2012) Use of flood, loss, and evacuation models to assess exposure and improve a community tsunami response plan—Vancouver Island. Natural Hazards Review 13(2): 162–171 Jonkmann S, Vrijling J, Vrouwenvelder, A (2008) Methods for the estimation of loss of life due to floods: a literature review and a proposal for a new method. Natural Hazards 46: 353-389 Kenai Peninsula Borough GIS Division (2013) Geographic information systems downloads. Available via http://www2.borough.kenai.ak.us/GISDept/Downloads.html. Accessed 7 Jan 2013 Lander J (1996) Tsunamis affecting Alaska 1737 – 1996. NCGC Key to Geophysical Research Documentation No 31. National Geophysical Data Center, Boulder, CO Lemke R (1967) Effects of the earthquake of March 27, 1964, at Seward, Alaska. Geological Survey Professional Paper 542-E. United States Government Printing Office: Washington, D.C. MarathonGuide.com (2011) Boston marathon race results 2010, Available at http://www.marathonguide.com/results/browse.cfm?MIDD=15100419. Accessed 8 Mar 2011 NOAA National Geophysical Data Center/World Data Center (2013) Global Historical Tsunami Database. Available at http://www.ngdc.noaa.gov/hazard/tsu_db.shtml. Accessed 28 Jan 2013 Norton F, Haas J (1970) The cities and towns. In: The Great Alaska Earthquake of 1964—Human Ecology. National Academy of Sciences, Washington, D.C., 248-356 Port of Alaska (2013) History. Available via http://www.portofalaska.com/about/history.html. Accessed 10 May 2013 Rogers G (1970a) Economic effects of the earthquake. In: The Great Alaska Earthquake of 1964—human ecology, National Academy of Sciences, Washington, D.C. pp. 58-76 Rogers G (1970b) Appendix C—basic population and employment statistics, South Central Alaska, 1960-1967. In: The Great Alaska Earthquake of 1964—human ecology, National Academy of Sciences, Washington, D.C. 441-448 Seward Chamber of Commerce (2013) 2013 Cruise ship schedule. Available at http://www.sewardchamber.org/wpcontent/uploads/2013-Cruise-Ship-Schedule.pdf. Accessed 14 Aug 2013. Seward Historic Preservation Commission (2012) Community history and character. Available via http://www.cityofseward.net/hpc/seward_history/index.html. Accessed 18 Aug 2012 Sturman G (1973) The Alaska railroad. In: The Great Alaska Earthquake of 1964—engineering. National Academy of Sciences, Washington, D.C. 958-986 Soule R, Goldman R (1972) Terrain coefficients for energy cost prediction. Journal of Applied Physiology 32:706– 708 Suleimani E, Nicolsky D, West D, Combellick R,Hansen R (2010) Tsunami inundation maps of Seward and Northern Resurrection Bay, Alaska. Report of Investigations 2010-1, Alaska Division of Geological & Geophysical Surveys

14

Tobler W (1993) Three presentations on geographical analysis and modeling—non-isotropic geographic modeling. Speculations on the geometry of geography; and global spatial analysis. UCSB. National Center for Geographic Information and Analysis Technical Report 93-1. Available at http://www.ncgia.ucsb.edu/Publications/Tech_Reports/93/93-1.PDF. Accessed 19 July 2010 United States Census Bureau (1963) U.S. Census of Population—1960. In: Characteristics of the Population, vol 1, part 3, Alaska. U.S. Government Printing Office, Washington, D.C. United States Census Bureau (2012) American FactFinder. Available at http://factfinder2.census.gov/faces/nav/jsf/pages/index.xhtml. Accessed 27 Oct 2012 United States Department of Transportation (2009) Manual on uniform traffic control devices for streets and highways. Federal Highway Administration United States Geological Survey (2012) Earthquake “Top 10” Lists & Maps. Available via http://earthquake.usgs.gov/earthquakes/eqarchives/. Accessed 26 Oct 2012 Wood N, Schmidtlein M (2012) Anisotropic path modeling to assess pedestrian-evacuation potential from Cascadiarelated tsunamis in the U.S. Pacific Northwest. Natural Hazards 62(2): 275-300 Wood N, Schmidtlein (2013) Community variations in population exposure to near-field tsunami hazards as a function of pedestrian travel time to safety. Natural Hazards 65(3): 1603-1628 Yeh H, Fiez T, Karon J (2009) A comprehensive tsunami simulator for Long Beach Peninsula phase-1—framework development final report. State of Washington Military Department Emergency Management Division.

15

Figure 1. Maps of a the maximum tsunami-inundation zone in Resurrection Bay, Alaska (Suleimani et al. 2010) and b the study area relative to the State of Alaska, United States of America (USA). Map is based on geospatial data acquired from Kenai Peninsula Borough GIS Division (2013).

16

Figure 2. Georectified imagery of a downtown Seward on May 29, 1963 (Lemke 1967; U.S. Army Corps of Engineers, unpublished data) showing the 1964 inundation zone, b downtown Seward one day after the earthquake (Lemke 1967; U.S. Army Corps of Engineers, unpublished data) showing the 1964 inundation area, the pre-disaster shoreline and estimated fatality locations, and c downtown Seward in 2005 showing multiple tsunami-hazard zones and the pre-1964 shoreline (Kenai Peninsula Borough GIS Division 2013).

17

Figure 3. Maps of landcover a manually interpreted from 1963 imagery and b manually interpreted from 2005 imagery.

18

Figure 4. Demographic age profiles for residents based on 1960 and 2010 U.S. Census Bureau estimates.

19

Figure 5. Maps of modeled evacuation travel times in downtown Seward based on a 1963 land cover and 1960 population distributions over a 1963 image of Seward and b modern day conditions over a 2005 image for Seward. Figure 5a also includes estimates of fatality locations as described in Lemke (1967) and Lander (1996).

20

Figure 6. Graph of the distribution of residents as a function of modeled pedestrian travel time to safety for 1960 and 2010 estimates and evacuation model runs.

21

Figure 7. Maps of modeled evacuation travel times across Resurrection Bay, Alaska assuming a slowwalking speeds and open travel, b slow-walking speeds and travel constrained to road networks, c slowrunning speeds and open travel, and d slow-running speeds and travel constrained to road networks.

22

Figure 8. Graphs of the distribution of residents as a function of modeled pedestrian travel time to safety assuming slow-walking and slow-running speeds for a open travel to high ground and b travel constrained to road networks.

23

Figure 9. Graphs of the distribution of a residents and employees, b businesses and organizations with substantial customers, and c cumulative percentage of land in the maximum tsunami-hazard zone, as a function of travel time to safety.

24