Florida State University Libraries

Transcription

1 Florida State University Libraries Electronic Theses, Treatises and Dissertations The Graduate School 2015 Tsunami Evacuation: Using GIS to Integrate Behavioral and Vulnerability Data with Transportation Modeling Khameis Alabdouli Follow this and additional works at the FSU Digital Library. For more information, please contact

2 FLORIDA STATE UNIVERSITY COLLEGE OF SOCIAL SCIENCES AND PUBLIC POLICY TSUNAMI EVACUATION: USING GIS TO INTEGRATE BEHAVIORAL AND VULNERABILITY DATA WITH TRANSPORTATION MODELING By KHAMEIS ALABDOULI A Dissertation submitted to the Department of Geography in partial fulfillment of the requirements for the degree of Doctor of Philosophy Degree Awarded: Spring Semester, 2015

3 Khameis Alabdouli defended this dissertation on March 3, The members of the supervisory committee were: Mark Horner Professor Directing Dissertation Timothy Chapin University Representative Earl J Baker Committee Member Tingting Zhao Committee Member Eren Ozguven Committee Member The Graduate School has verified and approved the above-named committee members, and certifies that the dissertation has been approved in accordance with university requirements. ii

4 To my wife, my parents, and my lovely children (Mohamed and Maryam) iii

5 ACKNOWLEDGMENTS It is hard to imagine that eight years have passed since I first came to the United States to begin my graduate studies in Geography. It has been an interesting academic journey filled with opportunity to gain in knowledge and experience. As I began to approach the last stage of my graduate studies, I found that I was lucky to be surrounded with such an outstanding group of faculty members and colleagues here in the Geography Department of Florida State University. This dissertation could only have been completed with the guidance and the support from those who both in academia and in my personal life have helped me, in innumerable ways, throughout my life and academic journey. I would like to give my thanks and gratitude to those who contributed to each aspect of my academic achievement. I owe my deepest gratitude to my advisors, Dr. Mark Horner and Dr. Jay Baker, who have guided and assisted me throughout my study and drew the successful path for the accomplishment of my research and this dissertation. This research would have been next to impossible without their support and guidance. It is my pleasure to thank my outside dissertation committee member, Dr. Eren Ozguven, for his advice and suggestions for improving my work; his contribution to the quality I have tried to achieve is immense. I would like to thank Dr. Tingting Zhao for her participation on my dissertation committee and her contributions in strengthening my document. Also, I would like to express my appreciation to Dr. Timothy Chapin for serving as a second outside committee member: I am thankful for your feedback and for the interest that you have shown in my research. To Ms. Marry Anne Sennet, our family friend, thank you for the support and efforts that you have given throughout the writing of my dissertation. I also would like to thank my friend, Kaveh Shahabi, for his cooperation throughout the period of time that I have worked on my research and dissertation. Many thanks are owed to my family for their support and encouragement. I hope I have made you proud and will always continue to do so. It has been a long journey filled with obstacles and challenges. However, with your untiring support and efforts, I was able to make it to this point. Lastly, for my dear wife and my children (Mohamed and Maryam), you were among the best in giving me the motivation and support to reach my iv

6 goal. It is hard to believe the number and the difficulty of the challenges we overcame over the past few years, especially when I was away from you. I am so happy for the accomplishments that we have achieved, and I am really proud of you and the future I imagine before us. Thank you. v

7 TABLE OF CONTENTS List of Tables... ix List of Figures... xiv Abstract...xv 1. INTRODUCTION Research Background Importance and Contribution of Research Broader Impacts Limitations of Research Organization of Research LITERATURE REVIEW History of Evacuation Evacuation Research Evacuation Planning Components of Evacuation Studies Hazard Analysis Vulnerability Analysis Behavioral Analysis Transportation Analysis Tsunami Evacuation Studies GEOGRAPHIC DATA MOST Model Product Demographic and Transportation Data Limitation of the Data MOST Model Limitations Census and Road Data Limitations Study Area Tsunami History in Orange County, California Tsunami Evacuation Planning in Orange County, California TSUNAMI EVACUATION AND HAZARD ANAALYSIS Delineate Evacuation and Shadow Evacuation Zones Convert MOST Model Maps to GIS Format (Evacuation Zone)...43 vi

8 4.1.2 Define Shadow Evacuation Zone Number of Households in Evacuation and Shadow Evacuation Zones Number of Vehicles per Household in Evacuation and Shadow Evacuation Zones BEHAVIORAL ANALYSIS AND SURVEY QUESTIONNAIRE Survey Questionnaire Information Sources and Residents Awareness in Orange County, California: Tsunami Knowledge, Beliefs, and Perceived Vulnerability General Knowledge Tsunami Perception and Beliefs Information Sources Tsunami Scenarios Intention for Evacuation Refuge Type Main Evacuation Roads Vehicle Usage for Evacuation Demographic Characteristics Physical Characteristics Demographic Characteristics of Orange County, CA EVACUATION PARTICIPATION RATE AND TRIP GENERATION Evacuation Participation Rate Logistic Regression Models Data Logistic Regression Models for Each of the Three Tsunami Scenarios (10-, 20-, 30- foot) Enter Method Backward Stepwise Method Predicting Number of Evacuees Estimating Number of Evacuating Vehicles (Trip Generation) EVACUATION MODELING AND CLEARANCE TIME Overview of CASPER Tool Assumptions The CASPER Construct Data in the CASPER Tool vii

9 7.5 CASPER Tool Processing Phases Phase Phase Phase Comparing the Power Model to the Bureau of Public Roads Function Tsunami Evacuation Modeling Approaches Modeling Tsunami Evacuation Scenarios Results One-Second, Initial-Delay Cost per Evacuee Approach Seven-Seconds, Initial-Delay Cost per Evacuee Approach CASPER Limitations CONCLUSION AND FUTURE RESEARCH Chapter Summary Importance of Research and Its Broader Impacts Future Research APPENDICES A. TELEPHONE SURVEY B. HUMAN SUBJECT COMMITTEE APPROVAL LETTERS C. PERMISSION TO USE MOST MODEL MAPS REFERENCES BIOGRAPHICAL SKETCH viii

10 LIST OF TABLES Table 2.1 Current tsunami evacuation studies that consider evacuation process components...22 Table 4.1 Total number of household per block group for evacuation and shadow evacuation zones...50 Table 4.2 Total number of vehicles per household for evacuation and shadow evacuation zones...52 Table 5.1 Q1) What is the most frequent cause of tsunamis?...59 Table 5.2 Q2) When did the last tsunami cause damage to coastal areas in California?...59 Table 5.3 Q12) Do you live in a tsunami inundation hazard zone?...60 Table 5.4 Q13) How high is the first floor of your house from sea level?...60 Table 5.5 Q14) Which of the following signs might warn you about possible tsunami?..61 Table 5.6 Q16) Which of these locations is/are considered to be a significant source of earthquake that might generate a local tsunami along the coast of Orange County?...61 Table 5.7 Q17) Which of these locations is/are considered to be a significant source of earthquake that might generate a distant tsunami along the coast of Orange County?...62 Table 5.8 Q18) How long does it take a tsunami waves generated from local earthquake or landslide to arrive to Orange County?...62 Table 5.9 Q19) How long does it take a tsunami from a distant earthquake to arrive to Orange County?...62 Table 5.10 Q20) Would you leave your home to go someplace safer from a possible tsunami if you felt an earthquake?...63 Table 5.11 Q21) Would you evacuate your home in case of a tsunami evacuation warning because of..?...64 Table 5.12 Q22) What would be the main reason/s you wouldn t evacuate your home in case of a tsunami affecting the coastal areas of Orange County?...66 Table 5.13 Q66) Do you know what vertical evacuation means?...67 Table 5.14 Q67) Are you planning to use it?...67 ix

11 Table 5.15 Q3) Are you familiar with the Orange County Alert system, sometimes called OCAlert?...67 Table 5.16 Q4) Are you registered in it?...67 Table 5.17 Q5) Do you have access to the internet at home?...67 Table 5.18 Q6) Do you have internet access on your cell phone?...68 Table 5.19 Q7) Does anyone else in your household have internet on their cell phone?..68 Table 5.20 Q8) Do you have a social media account such as Facebook or Twitter?...68 Table 5.21 Q9) How often do you use social media?...68 Table 5.22 Q10) Do you have a NOAA Weather Radio in your home?...68 Table 5.23 Q11) When deciding whether to evacuate from future tsunami, to what extent would you rely on each of the following for information?...69 Table 5.24 Q23, 38 and 52) If the government officials suggested an evacuation for a 10, 20, and 30 ft tsunami affecting the Orange County coast, would you evacuate your home?...71 Table 5.25 Q24, 40, and 54) How soon would you evacuate your home if the government officials suggested an immediate evacuation for a 10, 20, and 30 ft tsunami affecting the Orange County coast?...72 Table 5.26 Q26, 42, and 56) What type of refuge you would seek if you decided to evacuate your home based on the evacuation suggestion issued by the government?...73 Table 5.27 Q27, 43, and 57) Where would that refuge be located?...73 Table 5.28 Q29, 45, and 59) How far do you think you would need to evacuate inland based on your home location if government officials suggested an evacuation from a 10, 20, and 30 ft tsunami?...74 Table 5.29 Q31, 46, 60) What main roads would you use if the government officials suggested an evacuation for 10, 20, and 30 ft tsunami impacting the Orange County?...75 Table 5.30 Q34, 49, and 63) How long do you think it would take you to get to your planned safe evacuation location if government officials suggested an evacuation of a possible 10, 20, and 30 ft tsunami?...75 Table 5.31 Q36, 50, and 64) How many vehicles would be available in your household that you could use to evacuate?...76 x

12 Table 5.32 Q37, 51, and 65) How many vehicles would your household take if you evacuated?...76 Table 5.33 Q39, and 53) Would you evacuate to the same location in the same way as you would for a 10 or 20 ft tsunami?...77 Table 5.34 Q68) How old were you on your last birthday?...77 Table 5.35 Q69) Which of the following best describes your race?...78 Table 5.36 Q70) Which of these best describes your ethnic background?...78 Table 5.37 Q71) What is your marital status?...78 Table 5.38 Q72) What is your highest level of education?...79 Table 5.39 Q81) Was the respondent male or female?...79 Table 5.40 Q73) What is your yearly household income in 2013?...80 Table 5.41 Q75) Is there anyone in your household with a disability?...80 Table 5.42 Q76) What is the disability?...80 Table 5.43 Physical characteristics of the surveyed household in Orange County, CA for the (10, 20, and 30 ft) tsunami scenarios...81 Table 5.44 The characteristics of age, gender, and race for the population of Orange County, CA...83 Table 5.45 The percentage of the educated people in Orange County, CA...84 Table 5.46 The income level for the population of Orange County, CA...84 Table 5.47 Different disability types for the population of Orange County, CA...84 Table 6.1 Demographic variables codes in SPSS for logistic regression...92 Table 6.2 Physical variables codes in SPSS for logistic regression...92 Table 6.3 The output of the logistic regression model using the enter method for 10-ft tsunami...95 Table 6.4 The output of the logistic regression model using the enter method for 20-ft tsunami...96 xi

13 Table 6.5 The output of the logistic regression model using the enter method for 30-ft tsunami...97 Table 6.6 The output of the logistic regression model using the Backward Stepwise LR method for a 10-foot tsunami scenario Table 6.7 The output of the logistic regression model using the Backward Stepwise LR method for a 20-foot tsunami scenario Table 6.8 The output of the logistic regression model suing the Backward Stepwise LR method for a 30-foot tsunami scenario Table 6.9 Equivalent R Square statistical measures for 10-ft tsunami scenario using Backward Stepwise method Table 6.10 Percentage of correct predictions using only the constant in the 10-ft tsunami model Table 6.11 Percentage of correct predictions after including the variable zone in the 10-ft tsunami model Table 6.12 Equivalent R Square statistical measures for 20-ft tsunami scenario using Backward Stepwise method Table 6.13 Percentage of correct predictions using only the constant in the 20-ft tsunami model Table 6.14 Percentage of correct predictions after including the variable zone and gender in the 20-ft tsunami model Table 6.15 Equivalent R Square statistical measures for 30-ft tsunami scenario using Backward Stepwise method Table 6.16 Percentage of correct predictions using only the constant in the 30-ft tsunami model Table 6.17 Percentage of correct predictions after including the variable zone and gender in the 30-ft tsunami model Table 6.18 Evacuation participation rate using the prediction models Table 6.19 Tsunami evacuation participation rate based on the survey results Table 7.1 Tsunami evacuation time with and without congestion for 10-ft tsunami scenario using 1 second Initial Delay Cost approach xii

14 Table 7.2 Tsunami evacuation time with and without congestion for 20-ft tsunami scenario using 1 second Initial Delay Cost approach Table 7.3 Tsunami evacuation time with and without congestion for 30-ft tsunami scenario using 1 second Initial Delay Cost approach Table 7.4 Tsunami evacuation time with and without congestion for 10-ft tsunami scenario using 7 second Initial Delay Cost approach Table 7.5 Tsunami evacuation time with and without congestion for 20-ft tsunami scenario using 7 second Initial Delay Cost approach Table 7.6 Tsunami evacuation time with and without congestion for 30-ft tsunami scenario using 7 second Initial Delay Cost approach xiii

15 LIST OF FIGURES Figure 2.1 The interactive relationship between human behaviors and transportation system in making the evacuation decision for a disaster...13 Figure 3.1 The inundation line produced from the MOST model (Department of Conservation, 2014)...33 Figure 3.2 Location of Orange County and its subdivisions...38 Figure 4.1 MOST model inundation with the USGS 24K Quads for Orange County, CA...44 Figure 4.2 Orange County tsunami inundation maps produced using MOST model...45 Figure 4.3 Sample of MOST model map of Orange County, CA...46 Figure 4.4 The distance of the farthest inundation point (7,254 m), evacuation zone...47 Figure 4.5 Tsunami evacuation and shadow evacuation zone for Orange County, CA...49 Figure 4.6 Centroids of the US Census block group polygons within evacuation zone...51 Figure 4.7 The concentration of the number of households within the evacuation zone..51 Figure 4.8 The concentration of the number of households within the shadow evacuation zone...52 Figure 4.9 The total number of vehicles per household in the evacuation zone...53 Figure 4.10 The total number of vehicles per household in the shadow evacuation zone 53 Figure ft tsunami evacuation routes using one second for the initial delay cost per evacuee and 500 saturation density per unit capacity Figure ft tsunami evacuation routes using one second for the initial delay cost per evacuee and 200 saturation density per unit capacity Figure ft tsunami evacuation routes using seven seconds for the initial delay cost per evacuee and 500 saturation density per unit capacity Figure ft tsunami evacuation routes using seven seconds for the initial delay cost per evacuee and 200 saturation density per unit capacity xiv

16 ABSTRACT This dissertation adds to ongoing research efforts that seek a better understanding of the complex relationships between human beings and their environment. Specifically, the impact of natural disasters on coastal communities from the sudden occurrences of such disasters and their potentially devastating consequences e g., earthquakes and tsunamis requires communities to be prepared, by planning specific land-use polices and by developing evacuation plans. Through identifying the locations of populations at risk of a tsunami, determining their behavioral responses during an evacuation, and utilizing a GIS evacuation tool Capacity-Aware Shortest Path Evacuation Routing (CASPER), the research reported in this dissertation proposes a means to predict the number of evacuating vehicles from various tsunami scenarios and calculate evacuation clearance time for these scenarios. Motivated by the need to provide new ways to use the results of the behavioral analysis in evacuation modeling, this study developed a methodology to predict the number of evacuating vehicles based on the result of the behavioral analysis to estimate the evacuation clearance time using evacuation modeling. In order to estimate the evacuation clearance time, a framework was utilized and is described in four analytical chapters of this dissertation chapters 4 7: Chapter 4 identifies the population at risk of tsunami. Chapter 5 explores the characteristics of the sample population; Chapter 6 offers predictions on the number of evacuating vehicles, and, Chapter 7 models the evacuation process to estimate the evacuation clearance time with and without congestion. Each of these chapters interconnects with the succeeding chapter as they together complete the framework used to model evacuation clearance time. The ability to assess the preparedness for a natural hazard will broaden the understanding of the relationship between behavioral responses and evacuation transportation options, which in turn will provide planners, government officials, and geographers the required knowledge to address related planning issues such as evacuation planning and sustainable development. xv

17 CHAPTER ONE INTRODUCTION One of the traditional lines of research in the study of geography is that of the interaction of the complex relationships between humans and their environment. The impact of natural disasters on human activities has been among the topics of greatest interest to researchers in human-environment studies. The sudden occurrence of natural disasters that have the potential to result in devastating consequences, such as earthquakes and tsunamis, requires communities to be prepared by planning specific land-use polices and by developing evacuation plans. The variations in socioeconomic and demographic characteristics of a society -- in addition to the variety of physical settings that can affect the character of a disaster -- necessitate developing models that better explain this interactive relationship. By identifying the locations of populations at risk of a tsunami and determining their evacuation participation rate, the research reported in this dissertation proposes a means to calculate evacuation clearance time in face of a tsunami threat. The clearance-time calculation is based on an assessment of both behavioral responses and road-network capacity. The research reported here models the evacuation decisions of residents within the population study group and produces a methodology that evaluates the preparedness of coastal communities in the context of different tsunami scenarios. The ability to assess the preparedness for a natural hazard will broaden the understanding of the relationship between behavioral responses and evacuation transportation options. Such understanding will, in turn, provide planners, government officials, and geographers the required knowledge to address related planning issues, such as evacuation planning and sustainable development. 1.1 Research Background In 2010, 39 percent of the population of the United States, roughly more than 123 million people, lived in coastal shoreline counties (NOAA s state of the coast, 2014). Furthermore, the density of the United States coastal shoreline population is expected to 1

18 increase to 37 person/mi 2 between 2010 and 2020 (Woods and Poole, 2011). This increase in the density of the coastal population will be accompanied by shifts in demographic structures, an increase in urban sprawl, and variations in such demographic characteristics as an increase in the elderly population (Strategic Foresight Initiative, 2014). All of these changes are expected to affect the emergency management activities, including evacuation planning to protect the coastal communities from coastal hazards (NOAA s state of the coast, 2014; Strategic Foresight Initiative, 2014). Research shows that behavioral responses of evacuees during an evacuation are generally affected by a number of socioeconomic factors such as age, race/ethnicity, education, family size, and social/physical cues (Trainor et al, 2013). Therefore, variation in the demographic and socioeconomic characteristics from one coastal population to another will result in different evacuation responses of the evacuees, affecting in turn the evacuation process. There has been considerable evacuation behavioral research developed side-by-side with evacuation transportation modeling to understand the evacuation process and to plan for it according (Lewis, 1985; ORNL, 1995; PBS&J, 2000; Franzese and Han, 2001; Urbina, 2002; Lindell and Prater, 2007; Yazici and Ozbay, 2008; Trainor et al, 2013). The research setting for evacuation from a transportation perspective differs from that of a social science perspective. Transportation evacuation modeling usually starts with a problem that needs to be solved and/or a system that needs to be designed or redesigned to solve the problem; whereas social scientists are interested in creating theories or empirical tests while studying evacuation to explain how people act in case of an evacuation order in the event of a specific hazard (Trainor et al, 2013). Having differing research settings and focuses for behavioral and evacuation transportation modelers has led to the creation of a problem: behavioral findings cannot be directly transformed into simulation models (Lindell and Prater, 2007). Although a number of studies have aimed to improve evacuation research by integrating the behavioral characteristics of evacuees and various traffic evacuation models, there has remained a miscommunication between the two. Thus, it is the intention of this research to continue to build on the body of research that has worked to fill this need. More 2

19 specifically, this dissertation aims to contribute to the field of evacuation research through integrating the demographic characteristics of a population with transportation modeling in the context of evacuation in tsunami events. 1.2 Importance and Contribution of Research The conceptual setting of this research, in addition to the nature of the data used in this dissertation, suggests that this dissertation would be classified as interdisciplinary research. This dissertation will contribute to the tsunami evacuation literature, specifically in the United States since there is a lack of tsunami evacuation studies in this part of the world. This dissertation contributes to the general knowledge through introducing a methodology to communicate/convert the findings from a behavioral analysis to a transportation analysis in order to estimate evacuation clearance time. This is done through using statistical models to identify the most significant variables that contribute to the evacuation intention in order to predict the number of evacuees a calculation necessary to model the evacuation process in transportation modeling. This dissertation also will contribute to the field of geography by utilizing the notation of expectation, a statistical method to address scale issues related to the prediction models that were built based on data collected at the individual level, then used to make predictions on data at the aggregated level (census block group data). For purposes of emergency management, this research contributes to the ability to identify vulnerable populations, those at risk of tsunami, through delineating the extent of the shadow evacuation zone in the event of an impending tsunami. Furthermore, identifying the evacuation intention rates for different tsunami scenarios and predicting the expected number of evacuating vehicles, based on vehicle occupancy, will benefit emergency management officials by offering them a means to estimate the expected number of evacuating vehicles in the event of a tsunami. Emergency management officials will benefit, in addition, through an understanding of how the evacuation process can be enhanced by the use of CASPER tool: e.g., giving them the ability to produce simulated evacuation routes and to estimate evacuation time. In broader application, it is hoped that as this study furthers understandings of the tsunami evacuation process, it will, as well, lead to the 3

20 support of the sustainability of coastal communities through producing planning policies derived from the output of such research. 1.3 Broader Impacts This research is expected to benefit emergency managers and public officials in various ways. First, utilizing GIS to delineate the shadow evacuation zone will assist in identifying the spatial distribution of vulnerable populations, potentially at risk of a tsunami. Hypothesizing the magnitude of such a tsunami (i.e., the wave height) allows planners to view more than one possible scenario and the possible impacts of each on the public response and the evacuation process. Evacuation planners can estimate evacuation demand by using the predicted number of evacuating vehicles and the available clearance time that was simulated by CASPER tool with and without congestion. Estimating the evacuation demand and the clearance time may lead to enhance the current evacuation plans, or, may lead to redesign the evacuation routes in a way that better meets the needs of a revised, expected number of evacuees in order to reduce evacuation clearance time. Further, this research is expected to benefit academia, especially those professionals interested in evacuation research, by introducing the method of using prediction models built on data at the individual level to make predictions using data at the aggregated level. Such a method can help bridge the communication gap between the findings of behavioral analysts and those who work with transportation modeling, not only for tsunami evacuation, but also for any type of hazard event that requires an evacuation. 1.4 Limitations of Research Although this dissertation is expected to contribute to various disciplines including geography, transportation and urban planning, it has limitations. First, the results of this research are restricted by the data quality and availability. The MOST model inundation line that was used to delineate the evacuation zone in this research represents tsunami inundation of the worst-case scenario that was produced by several tsunamis (See Chapter 3 for greater detail). This indicates that the evacuation zone used in this research should not necessary be 4

21 inundated entirely from one single tsunami event. Also, the accuracy of the MOST model depends on the accuracy of the DEM that was used in the modeling process. The behavioral data analysis that was used to predict the evacuation participation rate was based on only 235 samples due to a limitation in the financial resource to collect more samples. Increasing the sample size might produce different evacuation trends in various tsunami scenarios. Increasing the sample size would likely produce different evacuation trends in various tsunami scenarios. The distribution of the sample size, with the majority of the sample being in the evacuation zone, affected the results of the prediction models: more samples collected in the shadow evacuation zone would have allowed greater consistency with the census data used in the prediction models. CASPER tool was able to locate evacuation routes and estimate congested and uncongested evacuation times, but with large amounts of travel time(s) due to the impact of the global parameters which treat all the roads the same way; conversely, traffic assignment models, which do not treat all roads the same, are able to reroute the traffic to an alternate route to reduce the overall travel time. The clearance time that was calculated using the behavioral responses was not based on day/or night tsunami scenarios. A daytime tsunami evacuation order versus a nighttime order might affect the evacuation participation rate since evacuee activities or occupations may vary greatly depending on time of day the evacuation order is received. This research focused only on private vehicle evacuation, whereas there are other possible means of evacuation that could be used in a tsunami event, such as motorcycle, public transportation, or by foot. 1.5 Organization of Research Chapter 1 has introduced the issues explored throughout this dissertation. This research develops a methodological framework to link the behavioral responses of evacuees to the evacuation transportation modeling; the purpose of the research is to use socioeconomic and demographic characteristics of the evacuees to predict the number of evacuating vehicles and simulate the evacuation process using GIS to calculate the evacuation clearance time. It is hoped the contributions of this research will affect the ongoing development policies and plans of the coastal communities. 5

22 Chapter 2 reviews a broad body of literature related to the conceptual and analytical chapters found later in the dissertation. The first two sections explore the history of evacuation in order to track the development of evacuation research originating from within various disciplines. The third section, which focuses on evacuation planning, reviews a number of studies in which evacuation practices and strategies were the focus. The forth section reviews the evacuation components of varying hazard types. Finally, several tsunami evacuation studies, each utilizing differing methods and evacuation components are discussed. Chapter 3 summarizes both the study area and data used throughout the conceptual and analytical chapters of this dissertation. A description of the MOST (Method of Splitting Tsunami) model maps that are used in this research is presented in terms of the simulation process and the characteristics of its final product. The limitation of the data that are used in this research is explained. This chapter also provides an overview of tsunami history and tsunami evacuation planning in Orange County, California. Chapter 4 is the first of the analytical chapters in this dissertation. The evacuation and shadow evacuation zones used throughout this research is developed using the inundation line of the MOST model in this chapter. These zones are used to identify the populations at risk of tsunami and their distribution. In addition, the distribution of the number of households and the number of vehicles per block group are identified based on these two zones. Identifying the evacuation and shadow evacuation zones is necessary for collecting the behavioral data in Chapter 5, predicting the number of evacuees in Chapters 6 and simulating the evacuation process in Chapter 7. Chapter 5 is another methodological chapter that describes the behavioral characteristics of the sampled population. It begins with the importance of the behavioral analysis in predicting the evacuation participation rate and the factors that affect the behavioral analysis. A description of the survey design is then given in addition to a general description of the survey sections, which are: 1) information sources and resident awareness in Orange County, California, 2) evacuation scenario (10-, 20-, 30-foot), 3) demographic 6

23 information. These descriptions are followed by details of the responses for each question in the three sections. The physical characteristics -- including distance, elevation, and location -- are described for the whole sampled population. Chapter 6 begins with an explanation of the logistic regression model as a mean of linking the output of the behavioral data with the transportation network to estimate the evacuation clearance time (Chapter 7). The logistic regression model has the capability of predicting the evacuation participation rate based on the demographic and physical characteristics of each respondent. Two methods of logistic regression model are tested, Enter and Backward Stepwise. The output of the Backward Stepwise method showed higher statistically significant results when compared with the results of the Enter method. The output of the logistic regression model, using the Backward Stepwise method, was used to predict the evacuation participation rate for the population at risk in all three tsunami scenarios. This research provides not only the number of evacuees for each tsunami scenario, but also the number of evacuating vehicles, which was used to model the evacuation process and estimate the clearance time reported in Chapter 7. Chapter 7 introduces a new evacuation-modeling tool to simulate the evacuation process and calculate the evacuation clearance time. The CASPER tool (Capacity-Aware Shortest Path Evacuation Routing) produces evacuation routes based on road capacity and number of evacuees. The tool goes through three data preparation phases: building the road network, specifying the evacuation and safe zones, and selecting the Power model for evacuation. This chapter demonstrates the use of the CASPER tool to model the evacuation process and estimate the evacuation clearance time with and without congestion. Different clearance times, based on vehicle occupancy, for the three tsunami scenarios are then reported to conclude the chapter. Chapter 8 is the final chapter of this dissertation. It summarizes the major findings reported in the analytical chapters (4-7) and presents these findings in terms that suggest broader conclusions may be drawn. Also, this chapter provides suggestions for the implementation of this research as well as possible directions for future research. 7

24 CHAPTER TWO LITERATURE REVIEW Natural disasters by their very nature vary widely. In addition to their uniqueness of origin whether by the movement of plates within the depths of the earth, a volcanic eruption, force of wind or water, on land or sea there are also the variables within each of these forces that can make them more or less destructive. These are based largely on the factors of location and intensity or size. People have for centuries and beyond tried to reduce the impact of such disasters as they have at the same time struggled to deal with these disasters within the context of the ever-changing elements of their natural and man-made environments. Such efforts, in more recent times, have included measures that range from legalistic in character to those that utilize such fields as social, geological and technological sciences. Numerous examples exist: building codes have been instituted to safeguard structures; evacuation plans that consider human behavioral patterns maximize the secure exit of a population in danger from the effects of a disaster. And, complicating the success of all efforts to secure populations at risk are the ever-changing, evolving challenges posed by such phenomena as rapid increases in population and the resulting explosions of urban development both of which increase the complexity of a successful evacuation of a given population in the event of a natural disaster (Pel et al, 2012). Lim et al (2013) defined evacuation as a critical part of disaster management as it entails moving people at risk to safety. Another detailed definition of evacuation, as cited by Abed El- Hameis et al (2012) is a crisis management activity in which all or part of the population is temporary relocated, whether in an organized or unorganized manner, from the location that has been struck, or is about to be struck by a disaster, to a place not considered to be dangerous in term of the health and safety. Generally, there are several factors that make the evacuation effective: warning and response time, information and instruction and the dissemination procedure, evacuation routes, traffic flow conditions, and dynamic traffic control measures (Lim et al, 2013). However, not all evacuations can follow the same procedures since each evacuation is somewhat unique, depending on such variables as: socioeconomic factors, infrastructure 8

25 features, the type of hazard and its intensity. The evacuation can be classified either as small or large-scale, immediate (no-notice), or one that is pre-warned (short-notice), with the latter being further defined and classified as mandatory, recommended, and voluntarily (Lim et al, 2013). Depending on the disaster type, its intensity and population at risk, different evacuation types are implemented. Usually in voluntarily evacuation, in which most people will not leave their homes, no specific traffic control or transportation measures are taken. In the case of a recommended evacuation, the disaster is viewed as having a high probability of causing a threat to people living in risk areas (Urbina and Wolshon, 2003). 2.1 History of Evacuation Research that focused on evacuation as a means of moving populations from potentially dangerous or life-threatening areas and situations began to develop in the 1950s, with studies directed toward evacuee behavior in hurricane events (Baker, 1991). There has been a significant amount of behavioral evacuation survey research in this area since that time, especially in the aftermath of many hurricanes. The resulting literature surrounding such research is now sufficient enough in size and geographical extent that researchers can now form conclusions about coastal populations in terms of their behaviors towards approaching storm threats. There has been a consistency in the findings of hurricane evacuation studies explaining how and why people evacuate. The level of warning and the perception of risk affect the evacuation participation rates within ranges of approximately 30% to 100%. For example, the evacuation participation rate is approximately 90 to 95% in populations living in the low lying coastal areas and barrier islands -- areas considered as high risk; these rates are in contrast to those (25 to 35%) of populations in low risk areas (Urbina, 2002). These findings and others were related to hurricane evacuation research that continued up to the 1970 s (USACE, 1970; Urbanik, 1978; USACE, 1979). However, later research began to focus more on evacuation in the context of nuclear power plant failure, particularly in the aftermath of the meltdown accident at Three Mile Island in Pennsylvania on 8 March This event changed the focus of evacuation research from evacuees behaviors to estimating the number of vehicles that would be used. Researchers developed approaches to vehicle availability by determining the total number of people or households expected to evacuate and by estimating the number of people who would use a 9

26 vehicle to evacuate. The estimated number of vehicles intended to evacuate was analyzed with the highway network traffic. At the time, this research was considered to be quite different from previous evacuation research (Yazici and Ozbay, 2008). Transportation planning analysis, as an element of a successful evacuation, advanced during the 1980s because of behavioral research: measurement of public response to both perceived hazards and evacuation orders formed the basis for the development of conceptual and computer modeling of travel demand forecast patterns (Urbina, 2002). Lewis (1985) was the first to describe the general process for travel demand forecasting for hurricane evacuation, by using the traditional forecasting methodology for urban travel demand (Yazici and Ozbay, 2008). Although most transportation evacuation models were developed in case of nuclear power plant emergencies, in the mid 1980 s, a study was conducted for the city of Virginia Beach that developed such a model for hurricane evacuation, by using the MASS Evacuation (MASSVAC) model, a simulation model developed for hurricane evacuation. The MASSVAC model was run using different hurricane intensity levels and transport operational strategies, e.g., contra flow, the use of shoulders, and change-signal operation to flashing mode. The results showed that the size of the population to be evacuated, the location and the number of shelters, the capacity of evacuation routes and the time available for evacuation affected the evacuation process (Urbina, 2002). Since that study, there have been several hurricane evacuation models developed, such as the Oak Ridge Evacuation Modeling System (OREMS), originally designed for nuclear attacks, but utilizable for hurricane evacuation. The OREMS model is used to arrive at the following: an estimate of clearance time, operational traffic characteristics, exit routes, and time needed for evacuation calculations and determinations necessary to develop an evacuation plan (ORNL, 1995). Incident Management Decision Aid System (IMDAS) is a tool modified from the OREMS model; it is able to model hurricane evacuation activities in a more timely and accurate manner (Franzese and Han, 2001). Another demand forecast model, Evacuation Traffic Information System (ETIS), was developed as a result of the unanticipated large volume of traffic generated during the hurricane Floyd evacuation. The ETIS is a GIS web-based travel demand forecast model that anticipates evacuation traffic congestion and cross-state travel flows for every coastal state between Delaware and Texas (PBS&J, 2000). All of these models represent development trends and needs in evacuation modeling. Evacuation research has 10

27 continued to improve by integrating the behavioral characteristics of evacuees and various traffic evacuation models a significant need for the field of evacuation research. Thus, it is the intention of this researcher to continue to build on the body of research that has worked to fill this need. More specifically, this dissertation aims to contribute to the field of evacuation research through integrating the demographic characteristics of a population with transportation modeling in the context of evacuation in tsunami events. 2.2 Evacuation Research Introducing varying social theories and practices regarding evacuee behaviors in addition to the development of computer modeling has led to improving the field of evacuation research. As stated before, behavioral studies had been conducted since the 1950 s, specifically in the aftermath of many hurricanes (Urbina, 2002). The main focus of these studies was on how and when the evacuees chose to evacuate, in addition to how they chose their routes and destinations (Yazici and Ozbay, 2008). There are several socioeconomic factors such as age, race/ethnicity, education, family size, social/physical cues, etc., that affect the behavioral responses of evacuees during an evacuation (Trainor et al, 2013). Such evacuee characteristics are generally used to predict evacuation patterns. For instance, according to Yazici and Ozbay (2008), the first study conducted to model the effect of intensity of a hurricane on evacuation behavior (in addition to destination patterns) was conducted by Whitehead et al (2000) through applying a logistic model in order to estimate the evacuation destination. Having knowledge about the evacuee behavior is critical to defining the evacuation situation since each person or household in the risk area may react differently based on his or her perception and past experience towards the same disaster (Lim et al, 2013). As stated above, evacuation behavioral research developed side by side with evacuation transportation modeling. Trainor et al (2013) stated that there are generally three main steps in evacuation transportation modeling: the pre-modeling decision, supply modeling and demand modeling. The pre-modeling decision usually deals with identifying the problem that the model is intended to solve. It defines the elements of the evacuation the model must represent: i.e., the evacuation zone, bottleneck locations, clearance time, and an evacuation scenario. The supply-modeling step focuses on the capacity of the road network to move people out of harm s way. Demand modeling deals with the extent to which people might use the roads 11

28 in case of an evacuation (Yazici and Ozbay, 2008). Having transportation modeling divided into these three phases led to the development of specific models that could focuses on each step separately. Evacuation research is considered to be an interdisciplinary field since it incorporates both behavioral and transportation foci (Lindell and Prater, 2007). Whereas transportation engineering research focuses on estimating the time needed for evacuation using traffic generation, trip distribution, vehicle routing, and supporting transportation infrastructure, behavioral research focuses on people s behaviors towards various warnings and hazard conditions (Cova et al, 2005). Social scientists are interested in creating theories or empirical tests while studying evacuation to explain how people act in case of an evacuation order from a specific hazard. However, the research setting for evacuation from a transportation perspective is different since it starts with a problem that needs to be solved and/or a system that needs to be designed or redesigned to solve the problem (Trainor et al, 2013). Having differing research settings and focuses for behavioral and evacuation transportation modelers has led to the creation of a problem: behavioral findings cannot be directly transformed into simulation models (Lindell and Prater, 2007). Furthermore, most evacuation studies have failed to capture the fact that human behavior and transportation systems have an interactive relationship: not only does the disaster itself affect the evacuation decision; equally affecting are transportation conditions and options, demographic variables, people s experiences and their risk perceptions (see Figure 2.1). For instance, according to behavioral research, using pre-and post-event surveys, there is a lack of perception and preference on the part of the public for specific evacuation strategies (Trainer et al, 2013). Miscommunication between behavioral and transportation modelers represents a significant challenge that needs to be addressed for evacuation planning. 2.3 Evacuation Planning Evacuation planning is considered to be one of the most important aspects of today s resilient communities (ADPC, 2008). This is due to the role evacuation planning plays in reducing the loss of lives and properties from various consequences that may result from a specific disaster. Very commonly, evacuation plans are used to create the needed evacuation policies and practices for individuals and/or households within the risk areas (Lim et al, 2013). Modeling alternative evacuation scenarios contributes to the creation of optimal evacuation 12

29 policies and strategies, which in turn leads to the improvement in evacuation planning (Lumbroso et al, 2008). Models of evacuation planning differ in their geographical scale, the density or size of an affected population, and in the time spans needed for evacuation. These differences will lead to varying evacuation policies and strategies (Xie et al, 2010). Therefore, planning for an evacuation should not be limited to the offices of emergency managers and law enforcement officials, but should also involve transportation officials -- due to their experience in traffic control, design, and planning. Calculating the clearance time is one of the critical issues related to evacuation planning; it is one in which transportation planners have a significant role, both in forecasting the travel demand in an evacuation and -- by using various transportation modeling techniques quantifying the travel patterns (Urbina, 2002). Disaster Human Behaviors: 1- Demographic variables such as (age, gender, income, race, etc). 2- Experience 3- Perception of risk Evacuation Decision Transportation System: 1-Transportation condition such as traffic or highway work zone 2- Transportation options such as public or privet Figure 2.1 The interactive relationship between human behaviors and transportation system in making the evacuation decision for a disaster. Although transportation engineers had been involved in evacuation modeling for a variety of disasters (natural and/or man-made) for many years previous, it is only since 1998 that they have been engaged in the actual development of evacuation plans, i.e., more particularly, 13

30 after hurricanes Georges and Floyd, which produced the two largest evacuations in the history of the United States and, probably the largest two traffic jams. The engagement of transportation professionals in evacuation planning led to improving the evacuation issues specifically related to transportation such as forecasting evacuation travel demand, evacuation traffic analysis and modeling, and the application of intelligent transportation system technologies (Urbina and Wolshon, 2003). Different evacuation policies and strategies have been created to enhance evacuation performance and reduce clearance time from various disasters. Urbina and Wolshon (2003) conducted a study to confine the latest policies and practices related to evacuation and compare them from one location to another. Their study focused on transportation evacuation issues where departments of transportation and emergency management officials in coastal states, threatened by hurricanes, participated in a survey to measure the effectiveness of the evacuation plans and their preparedness. The sudden occurrence of a disaster requires emergency management officials and transportation planners to be prepared to use a variety of disaster scenarios in advance of an actual evacuation event. However, experience has shown that there are several conditions that need to be met in order to improve the evacuation process. In addition to better planning for and coordination of regional and cross-state evacuations, these conditions include, but are not limited to increasing the capacity of the evacuation route, limiting the evacuation travel demand, and improving the transportation systems in a way that insure better, faster, and more reliable exchange of traffic flow and travel information (Urbina and Wolshon, 2003). There have been a number of evacuation practices and strategies proposed and executed that have improved the process and reduced the evacuation time. One of these practices is the contra-flow strategy, which is known also as reverse laning. This strategy requires traffic to reverse the driving direction in one or more of the inbound lanes (or shoulders) for use in the outbound direction in order to increase road capacity during an evacuation. There are four types of contra-flow strategies: all lanes reversed, one lane reversed and one lane with normal inbound flow for emergency/service vehicle entry, one lane reversed and one lane normal flow for inbound traffic entry, and one lane reversed with the use of the shoulder of the outbound lanes (Urbina and Wolshon, 2003). A study conducted by FEMA (2000) concluded that almost 70% of the road capacity increased when using a full-reversal lane strategy. Despite the fact that the 14

31 contra-flow strategy represents a significant improvement in road capacity, the pros and cons regarding its use and benefits to capacity improvements, safety, and manpower requirements are still unknown. Another evacuation transportation strategy is the use of the right shoulder of the outbound lanes. Almost 8% of the road capacity increases due to the use of this strategy (FEMA, 2000). There are two negatives to using shoulders in the evacuation process: pavement suitability and bridge widths. The shoulders of the roads usually have thin pavement that cannot resist the high volume of traffic during an evacuation, a situation, which in turn may lead to the creation of bottlenecks as the vehicles merge back into the evacuation lanes. Moreover, some of the old bridges are not designed with wide shoulders, potentially presenting another challenge by causing congestion close to these bridges. One of the most critical issues related to evacuation during an emergency is the real time status of the traffic. Thus, an Intelligent Transportation System (ITS) for evacuation is another strategy that has been used by transportation planners during an evacuation to monitor traffic flow rates and speeds -- along with lane closures, weather conditions, travel time, traffic-related incidents, and the availability of alternative routes toward which to guide evacuees (Urbina and Wolshon, 2003). Another issue that has been taken into consideration while planning for an evacuation is that of the need to identify the highway work zone. Having an active highway work zone during an evacuation may negatively affect the evacuation process. Thus, some of the Departments of Transportation (DOT) offices restrict the contracts for highway improvement to insure that such modifications do not conflict with the evacuation traffic through the work zone. Another issue transportation planners must take into consideration, in the context of an evacuation, is how to facilitate the movement of groups of people who do not have a means of transportation out of the area at risk. The low mobility group includes people who do not own a vehicle, the indigent and elderly or infirm, prisoners, and tourists. The responsibility of moving out the low-mobility group is generally that of the chief administrator of a given facility (e.g., a prison, hospital, nursing or assisted-living home). Therefore, it is recommended that this administrator have an overview of the local and state evacuation plans and communication procedures to insure the effectiveness of the evacuation (Urbina and Wolshon, 2003). Addressing these evacuation issues 15

32 while planning for an evacuation is a critical need and transportation planners play a significant role in not only creating and implementing the evacuation plans, but also improving the effectiveness of the evacuation. 2.4 Components of Evacuation Studies Planning an evacuation in the event of a natural disaster is not an easy task due to the variations in the nature and the development of the disaster, spatially and temporally (NCHRP, 2009). Moreover, there are various dynamic interactions between different physical and socioeconomic variables that may differ from one disaster to another. Therefore, it is important to identify specific evacuation components with regard to a specific hazard: in this immediate study, it is a tsunami. Most of the hurricane evacuation studies were initiated during the 1980 s by the Federal Management Agency (FEMA) to improve hurricane evacuation planning and assist in disaster preparedness (Urbina, 2002). These studies focused on storm hazard and vulnerability analysis, behavioral analysis of evacuees, and shelter and transportation analyses. Baker (2000) suggested a list of evacuation components, specifically for hurricane evacuations, which can be used for any hazard type. These components include the following forms of analysis: hazard, vulnerability, behavioral, transportation, shelter, decision-making, and development management (NCHRP, 2009). Since the focus of this research project is to analyze the evacuation clearance time of a possible tsunami affecting Orange County, California, only the first four components of the evacuation will be discussed below. The organization of the discussion will start with an overview of each component in the context of different forms of a disaster. Then, the researcher will discuss the components specifically for tsunami studies to identify the research needs this project will cover Hazard Analysis The first component of the evacuation plan is the hazard analysis, which examines the population and area at risk based on specific conditions of the hazard. The hazard analysis aims to highlight the possible affected area of a specific hazard event, such as a hurricane (Baker, 2000). Identifying the spatial extent of the potential risk area is not the only objective of the 16

33 hazard analysis; it also helps in directing attention toward other hazards associated with the specific hazard event itself. In case of a hurricane event for instance, a hazard analysis helps to identify the areas at risk of damage that results from a storm surge, hurricane-force winds, tornadoes, or fresh water flooding. This could be done through utilizing the Sea, Lake, and Overland Surge from Hurricanes (SLOSH) model, considered to be the basis for the hazard analysis component of the hurricane evacuation plan and is typically used to identify different evacuation zones based on various hurricane conditions (CFRPC, 2010a; Baker, 2000). There are a number of studies that have used hazard analysis as part of evacuation planning. In the hazard analysis of the technical data report of the Connecticut hurricane evacuation study, conducted by the U. S. Army Corps of Engineers (1994), the SLOSH model was used as a tool for hazard analysis in determining that the intensity of the storm was the most significant metrological factor that might affect storm surge generation, based on a worst case hurricane surge. Also, evacuation planning improves due to the ability of a hazard analysis to highlight the factors that might amplify the storm surge. A hazard analysis takes various forms to identify the area that would need to be evacuated. These forms might include a hazard profile of the area in which general information about each hazard is specified. Another form of the hazard analysis is the history of hazard activity in the region. Additionally, a geo-spatial analysis of a hazard, e.g., inundation areas, wind fields, dam location, etc., is considered to be effective within the overall hazard analysis. Trople (2004) conducted a hazard analysis to assess volcanic hazard vulnerability. In his study, Trople analyzed the possible effects of the hazards associated with volcanic eruption through examining potential hazards in relation to critical facilities and social, economic, and environmental variables. Hazard analysis serves as the basis for the rest of the evacuation analyses since it provides the spatial extent of the risk area -- in addition to the other factors that might impact the evacuation process Vulnerability Analysis The second element of the components in the evacuation planning process is the vulnerability analysis in which different socioeconomic and physical factors are used to measure the vulnerability of a population toward different hazards. The vulnerability analysis follows the 17

34 hazard analysis of the planning process and seeks to specify the number of households and people at risk of the hazard (CFRPC, 2010a). In case of a hurricane, this is done through using the evacuation zone maps, produced through the SLOSH model as part of the hazard analysis, to identify the population and properties at risk within each zone. Usually, census data are used in this step to classify population characteristics within each hazard zone. It is important to understand the social components of the communities at risk since different populations may respond differently to the same disaster. For instance, variations in population characteristics, e.g., size and distribution -- especially in cases involving special needs populations -- may affect speed of travel behaviors during an evacuation (NCHRP, 2009). The vulnerability analysis conducted by the U. S. Army Corps of Engineers (1994) as part of the hurricane evacuation study for Connecticut showed that even if the spatial extent of the inundation of a hurricane is small, more people are vulnerable to a hurricane surge due to their concentration along the hazard zones. Several social vulnerability indices have been created to explain how a social system is vulnerable to a specific hazard. Cutter et al (2003) created a social vulnerability index to understand the role of the social variables in shaping the patterns of social vulnerability to natural hazards. Different socioeconomic variables such as income, gender, age, race and ethnicity were used to create social vulnerability indices by applying Pareto ranking to a complex, developed, socioeconomic landscape exposed to storm surge associated with hurricanes (Rygel et al, 2006). Curtis et al (2007) used income, age, gender, race and ethnicity to highlight the locations of the affected population in case of a disaster. Another set of socioeconomic variables including age, gender, income, race and ethnicity were used to assess the social vulnerability of population of Oregon coast to Cascadian tsunamis (Wood et al, 2010). Although there is an agreement that there is no specific set of socioeconomic and demographic variables that can explain social vulnerability of groups and/or individuals for all disasters, there are specific indicators that represent the least requirements to measure the social vulnerability such as age, gender, race, house value, income and special needs populations (Cutter et al, 1997; Kuhlicke et al, 2011). Using these indicators to measure social vulnerability will allow the researcher to specify the number of households and people in need of evacuation (Baker, 2000). Variations in the characteristics of the population may lead to diverse size, distribution, and 18

35 speed of travel behaviors during an evacuation, especially with the existence of special needs populations (NCHRP, 2009). Chakraborty et al (2005) conducted an assessment of a spatial pattern of hurricane evacuation assistance needs in Hillsborough County, Florida. Two quantitative indexes were used -- geophysical and social vulnerability -- to analyze four evacuation dimensions: 1) population traits and building structures, 2) differential access to resources, 3) special need evacuation, and, 4) a combination of variables. The results showed that geophysical and social vulnerability could lead to various spatial patterns that complicate the evacuation process. The vulnerability analysis represents a critical element of the evacuation analysis because it provides the analyst with specific social and physical measures that impact the vulnerability towards a specific hazard Behavioral Analysis One of the critical components of the evacuation plan is the behavioral analysis, which predicts the way a threatened population will respond to a disaster (Baker, 2000). Variations in the demographic characteristics will be reflected in different behavioral responses in different disaster events. Usually the population evacuation behaviors vary from one place to another for the same hazard, i.e., a hurricane, and, from one specific hurricane to another in the same place (Baker, 1991; Lindell and Prater, 2010). This understanding indicates that there will be variations in behavioral evacuation responses to different disasters, e.g., hurricanes and tsunamis. There are several factors that need to be considered in the behavioral analysis: the evacuation participation rate, evacuation time, public shelter use, evacuation destination, and vehicle use (Baker, 2000; CFRPC, 2010b). Identifying and understanding the specific variable characteristics of each of these factors would enhance evacuation planning. Baker (1991) stated that usually few people evacuate before evacuation orders are issued, and they evacuate as quickly as the situation requires. Furthermore, mobile home residents are more likely to evacuate when compared with built-site-home residents due to the perceived vulnerability of the hurricane risk, the strength of the house itself and the storm severity. 19

36 The behavioral analysis component of the U.S. Army Corps of Engineers hurricane evacuation study (1994) measured the evacuation participation rate, the timing of response, the number of vehicles to be used for evacuation, and the percentage of population that would seek shelters. The study showed that not all of the population within evacuation zones would participate in the evacuation process. Estimates of evacuation participation rates produced from a behavioral analysis are used to establish assumptions for transportation and shelter analyses. Luathep et al (2013) conducted a flood evacuation behavior analysis in urban areas before, during, and after a flood event. They used a questionnaire survey to interview the affected population and then applied logistic regression to develop two flood evacuation models: the evacuation decision model and the evacuation mode choice model. The results showed that gender, number of adults, and levels of ability or disability of persons are the significant factors that affect decision-making. Ricchetti-Masterson and Horney (2013) studied whether such factors as social control, social cohesion, and social capital could have modified the relationship between demographic groups and the failure to evacuate during Hurricane Irene. This was done through conducting a cross-sectional stratified two-stage cluster sample among residents of Beaufort County, NC. The results showed no significant association between demographic or social factors and evacuation in bivariate analysis. A behavioral analysis provides the analyst with significant findings relative to the behavioral characteristics of the targeted population in case of evacuation. The evacuation process can be modeled based on the output of the behavioral analysis, which gives an overview of the possible evacuation obstacles before the occurrence of an actual evacuation Transportation Analysis A transportation evacuation analysis is one of the most critical steps of evacuation planning. It links people s behavioral responses to the nature of the evacuation process through integrating the hazard analysis, an existing transportation network, shelters, and the population to be evacuated (CFRPC, 2010c). Various modeling techniques are used to model the evacuation process, which, in turn, produce different results depending on the input data and the model capability. Andem (2003) compared the traffic evacuation flow using static traffic assignment, dynamic traffic assignment, and observed traffic counts, based on survey data for South Carolina 20

37 after hurricane Floyd in Results showed that static models underestimate congestion levels while dynamic models account for non-uniform demand. Observed traffic counts results showed that traffic volumes vary during evacuation. The dynamic model was able to detect variations of observed traffic counts and capture the delay. Another study, conducted by Lindell et al (2011) of hurricane Lili, investigated evacuation logistics issues such as departing timing, vehicle use, evacuation routes, travel distance, shelter type, evacuation duration and evacuation costs. The results showed there was little correlation among demographic and geographic variables with the evacuation logistic variables. Murray-Tuite and Wolshon (2013) reviewed evacuation modeling and simulation research for roadway transportation and operation. In their review, they discussed various issues related to evacuation modeling such as the forecast of evacuation demand, the distribution of evacuation demand with different travel patterns, and the assignment of evacuation demand with various transport modes to regional/local road networks to reach safe destinations. In addition, they discussed evacuation modeling in various disciplines to fill the gap between behavioral and engineering research while utilizing emerging techniques for the verification, validation and model calibration. It is clear that behavioral analysis plays a critical role in transportation analysis as part of evacuation planning. There are some populations outside of the evacuation zone that participate in the evacuation process known as a shadow evacuation population. The estimation of the shadow evacuation population could be done through using the results of the behavioral analysis, which helps in specifying the possible evacuation time for most of the population, their participation rates and other characteristics that lead to calculating the clearance time (Lindell and Prater, 2007; CFRPC, 2010b; CFRPC, 2010c). The final output of the transportation analysis is the clearance time that is required to get the threatened population to a safer place. 2.5 Tsunami Evacuation Studies As stated above, different disasters may possess the same components that together contribute to understanding the disaster and, therefore, to planning for it. The natural setting of the disaster, the physical characteristics of the area at risk, in addition to the variations within the 21

38 socioeconomics involved affect evacuation clearance time -- one of the major points of focus in evacuation planning. Since this dissertation project is about tsunami evacuation, it is necessary to review the current literature in tsunami evacuation to identify some of the research gaps it proposes to fill. Table 2.1 lists some of the recent tsunami evacuation studies that specify components of the evacuation process. Table 2.1 Current tsunami evacuation studies that consider evacuation process components Article Hazard Vulnerability Behavioral Transportati Analysis Analysis Analysis on Analysis Lammel et al DEM and Socioeconomic 1000 Multi-agent (2010) building variables household Traffic shape and (population about their Simulation heights density, sex daily (MATSim) and age) activities Taubenbock TsunAWI Digital Surface agent traffic et al (2009) model and Model (DSM) household simulation Australian and number of about their model National people per daily (MATSim) University square meter activities and using sample Geoscience of 500 Australia geocoded (ANUGA) building tool Charnkol and 907 samples logistic Tanaboriboon to measure regression (2006) response patterns and response times 22

39 Table 2.1 continued Article Hazard Vulnerability Behavioral Transportation Analysis Analysis Analysis Analysis Post et al Germen- Cost Distance (2009) Indonesian Weighting Tsunami Early Warning project (GITEWS) Muhari et al Constant (2011) Roughness Model (CRM), Topographic Model (TM), and Digital Elevation Model (DEM) Imamura et al Tsunami Measure Static and (2012) hazard maps population dynamic based on vulnerability evacuation earthquake during models probability evacuation 23

40 Table 2.1 continued Article Hazard Vulnerability Behavioral Transportation Analysis Analysis Analysis Analysis Dall Osso Papathoma Sample size and Tsunami 894 for short Dominey- Vulnerability questionnaire Howes Assessment (6 questions) (2010) (PTA) model for vertical evacuation Di Mauro et Agent based al (2013) model (vehicular and pedestrian evacuation models). Mas et al Tohoku Netlogo used to (2012) University s create agent Numerical based model Analysis Model for Investigation of Near-field Tsunamis (TUNAMI model) 24

41 Table 2.1 continued Article Hazard Vulnerability Behavioral Transportation Analysis Analysis Analysis Analysis Kim et al Create Multi-agent (2013) tsunami Traffic hazard maps Simulation through (MATSim) analyzing the seismic zones Freire et al Tsunami High Modeling the (2013) inundation resolution elevation speed maps (High data set of susceptibility nighttime and and daytime moderate population susceptibility density tsunami) zones Gonzalez- Tsunami Exposure of Measure the Riancho et modeling of population, clearance time al (2013) historical road network from tsunami and potential and its arrival and tsunami characteristics response time threats Lammel et al (2010) conducted a tsunami evacuation study in the urban coastal areas for the city of Padang, in western Sumatra, Indonesia. Their research aimed to estimate the evacuation time, identify bottlenecks, and detect highly endangered areas of the study area. They used GIS to extract information from remotely sensed data, such as street data and building 25

42 shape in order to classify them, based to their vulnerability. The researchers then combined the Digital Elevation Model (DEM) with their data to form a hazard analysis for the study area. From the hazard analysis, they produced tsunami inundation zone maps and found that the flow velocity of water from a tsunami is higher between buildings. They performed a vulnerability analysis using census data, specifically for population density, sex and age at the residential district level to measure the population distribution and the inhabitants' physical capability to evacuate. Further to their study, they (2010) conducted a behavioral analysis by using a questionnaire survey of 1000 household; they asked the residents about their daily activities in order to estimate their distribution as a function of time in case of a tsunami. The Multi-agent Traffic Simulation (MATSim) model was used to model the evacuation by each agent s decision separately within the below-10 m. zone. The results were invalid due to limited inundation scenario characteristics and the distribution of population. Taubenbock et al (2009) studied tsunami evacuation and the early warning system for the city of Padang by integrating data and methodologies from multiple disciplines such as engineering, remote sensing, and the social sciences. Their focus was on high geometric and thematic analysis to meet the needs of small-scale, heterogeneous and complex urban systems. Hazard analysis was conducted through combining the TsunAWI model with the Australian National University and Geoscience Australia (ANUGA) tool to measure run-up on land. Four hazard zones were identified from different scenarios that measure the time between the warning of an earthquake and the arrival of a tsunami wave front. In their vulnerability analysis, Taubenbock et al (2009) assessed the height of individual structures, measured building size, specified roof types, and determined built-up density by using a digital surface model in combination with two different house masks derived from remotely sensed data. Resulting data were then used with fieldwork experience to identify residential, mixed, commercial and industrial usages on a basic level. Additionally, the fieldwork experience was used to localize the critical infrastructures such as hospitals and schools. The integration of hazard maps with the 3-D city model and time-dependent population distribution led to produce a probabilistic and quantitative assessment of the affected buildings and people -- depending on different scenarios. In addition, they calculated the average number of people per square meter using a sample of 500 geocoded buildings, distributed around Padang. Further, they conducted a behavioral analysis by 26

43 using the results of a questionnaire survey of a sample population of 1000 inhabitants to provide them with the necessary socioeconomic data that described the daily activities and mobility data of their research population. A regression model was applied to identify the socioeconomic variables that would most likely affect evacuation intentions. The researchers applied a transportation analysis of tsunami evacuation by using the multi-agent traffic simulation model (MATSim) to estimate bottleneck occurrence and evacuation time. Another study conducted by Charnkol and Tanaboriboon (2006) sought to analyze the behaviors and the backgrounds of permanent and transient populations of Phuket and Phang-nga, Thailand during a hypothetical tsunami evacuation. They collected behavioral data of 907 samples to measure the reaction of the sampled population to a future tsunami warning, specifically to determine the evacuees response patterns as fast, medium, and slow under different preparation and response times (60, 45, 30, and 15 minutes). The outputs of the preparation and response curves of permanent and transit residents were compared. The researchers then used logistic regression to estimate the probability of evacuees participating in quick or slow evacuation as part of the transportation analysis. The results concluded that the model outputs were natural in term of response times and evacuation behaviors according to their different backgrounds. Post et al (2009) assessed the human response to tsunami threats in Indonesia at a subnational scale. They grouped the human immediate response into various time categories that were compared with evacuation time from tsunami. In order to perform the hazard analysis, the researchers used a database of tsunami modeling results produced from the Germen-Indonesian Tsunami Early Warning project (GITEWS) to create the hazard zones (warning level zone and major warning level zone). These hazard zones were used along with environmental factors (e.g., land cover, topography) and demographic factors (e.g., age, gender distribution, and population density) to analyze the evacuation time people needed to rescue themselves using the Cost Distance Weighting GIS method. A study done by Muhari et al (2011) focuses mainly on a hazard analysis that examines three practical tsunami run-up models which can be used to evaluate the tsunami impact in 27

44 populated areas. The first model is the Constant Roughness Model (CRM) that applies a uniform roughness coefficient throughout the study area. The second model is the Topographic Model (TM) that uses a very detailed topographic dataset that includes height information integrated on a Digital Elevation Model (DEM). The third model uses a varying bottom roughness coefficient depending on type of land use and on the percentage of building occupancy on each grid cell. The study concluded that if no building height information exists on the topographic data, modeling the effect of building on tsunami flow could be done through an equivalent roughness coefficient. If the goal of the analysis is to measure the inundation of tsunami characteristics of an area where topography is less varied, using a cell size larger than the building size is recommended. However, in cases in which CRM is used, a smaller cell than average building size is recommended. Imamura et al (2012) analyzed the tsunami disaster in Padang City, Indonesia using several analysis techniques. First, they created hazard maps through using the output of prediction of comprehensive geodetic, paleo-geodatic and micro-atoll that accommodate the energy reduction caused by the 2007 potential tsunami impact on residents. In order to identify the time that the evacuees need to leave the evacuation zone, two evacuation models were used as part of the transportation analysis: static and dynamic. Results indicate that not all residents might have enough time to evacuate from the inundation zone before the arrival of the first wave. Dall Osso and Dominey-Howes (2010) conducted a survey analysis to evaluate the importance of tsunami evacuation maps for Sydney, Australia. They produced these maps using the Papathoma Tsunami Vulnerability Assessment (PTVA) model, which is a GIS-based model, developed using information about tsunami impact and results from post-tsunami surveys and building damage assessments. The attribute of the model allows for assessing building vulnerability. The outputs of the model were used to classify the study area into tsunami evacuation maps that consist of safe evacuation areas and safe evacuation buildings for the purpose of vertical evacuation. These maps then were introduced to the public of the study area (Sample size 894) along with a short questionnaire. The results show that these maps represent a 28

45 significant resource for tsunami evacuation and may be used for planning and emergency response purposes. Di Mauro et al (2013) conducted a study to measure the evacuation time for Padang, Indonesia using an agent based model (vehicular and pedestrian). The data input for these two models was provided from the literature and included: road capacity, intersection capacity, vehicle size, person size, and free flow speed. The results showed that pedestrian evacuation is strongly preferable to vehicular evacuation due to high population density and limited road capacity. Results also showed that most of Padang s population could not evacuate within 30 minutes by foot and further indicated that evacuation rates in the first 30 minutes strongly related to the availability of evacuation shelters in which effectiveness is limited by the capacity of the structure. Mas et al (2012) introduced the integration of an evacuation model with a numerical simulation of a tsunami and a causality estimation evaluation for the village of Arahama in the Sendai plain area of Miyagi Prefecture in Japan. They used the Tohoku University s Numerical Analysis Model for Investigation of Near-field Tsunamis (TUNAMI) to simulate the tsunami inundation. Netlogo was used to create an agent-based model to simulate the evacuation process in which human behavior and individual characteristics of evacuees were considered in addition to using road and shelter data derived from GIS. Tsunami departure curves were used to set up the departure time for agents to evacuate. The model allows identifying the bottleneck, shelter demand, and causality estimation, in addition to testing the ability of the model to allow vehicular and pedestrian agents to find their own way in evacuation routes. The results indicated that 90% of the population could be evacuated and that, in addition, 520 evacuees could be sheltered in the event. Kim et al (2013) conducted a hazard analysis for the eastern coast of Korea. They created tsunami hazard maps through analyzing the seismic zones along the coastal area. Researchers developed hypothetical tsunami scenarios to examine inundation characteristics in order to create tsunami hazard maps. The results showed that due to the bathymetric characteristics of the East 29

46 Sea, tsunamis tend to land on the eastern coast of Korea. Also, the highest tsunami wave produced from the tsunami scenarios modeling was 12 meters above sea level. Freire et al (2013) studied the impact of the spatio-temporal population distribution on different tsunami hazard zones to measure the evacuation susceptibility in Lisbon, Portugal. The study used tsunami inundation maps produced by modeling an event similar to the 1755 tsunami event to represent a worst-case scenario. These tsunami maps represent the hazard analysis for the study area where two susceptibility classes are identified: a high susceptibility tsunami zone, representing low lying coastal areas, and a moderate susceptibility tsunami zone where adjacent low-elevation areas lie further from the coast, with decreasing likelihood of inundation by tsunami. A vulnerability analysis was conducted through using an improved high-resolution data set of nighttime and daytime population densities; the results were combined with tsunami susceptibility zones in order to estimate population exposure. Also, a transportation analysis was performed by modeling the elevation speed of both scenarios in the study area. The results showed that the exposure to a tsunami is time dependent due to the strong variations from daytime to nighttime population distributions. The analysis indicated that the population at risk increases at nighttime when compared to daytime, specifically in the zone of high susceptibility. Further, the evacuation modeling results indicate that full horizontal evacuation might be problematic during the daytime, even if started after an early warning of a major tsunamitriggering earthquake event. Gonzalez-Riancho et al (2013) presented a framework for tsunami evacuation planning of coastal areas of El Salvador. A hazard analysis was conducted using tsunami numerical modeling to simulate historical and potential tsunamis originating from local and distant sources. The resulting hazard maps were further used to identify the exposed population and their characteristics and to analyze the characteristics of the road network. Tsunami arrival time and response time were calculated to measure evacuation clearance time. The results showed that the warning systems would need to be improved since a tsunami had the potential to reach most of the coastal communities before a warning could be issued, using the system then in place. 30

47 Each of the above noted evacuation components play a significant role in planning for an evacuation for any hazard type. However, reviewing the evacuation literature for different forms of hazards, including tsunamis, revealed that not all evacuation studies utilize the major evacuation components that were considered by Baker (2000) when calculating clearance time, the ultimate concern -- along with maximization of numbers of evacuees -- of all evacuation studies. In terms of tsunami evacuation, most of the tsunami studies focus on the physical settings of tsunami phases and their impact on the physical environment. This is classified under hazard analysis where the physical characteristics of a place contribute to increasing or reducing the impact of a tsunami. Also, identifying the right method of conducting the hazard analysis is critical in developing the remainder of the evacuation analysis. For instance, in Imamura et al (2012), the authors produced hazard maps based on a single earthquake event a sample less representational than one in which several historical earthquakes were included. Estimating a realistic evacuation clearance time depends not only on the results of a hazard analysis, but also on the results of vulnerability and behavioral analyses; both of these, together, provide details of the demographic variables that significantly affect evacuation clearance time (Charnkol and Tanaboriboon, 2006; Post et al, 2009). 31

48 CHAPTER THREE GEOGRAPHIC DATA This chapter describes the nature of the data used in this research; data characteristics and limitations of the data sets used will be discussed. In addition, this chapter will introduce the geographic areas of study explored in forthcoming chapters. Lastly, a discussion about tsunami history and evacuation planning in the study area, Orange County, California, will be covered. 3.1 MOST Model Product The MOST (Method of Splitting Tsunami) model is a collection of numerical simulation codes that are able to simulate three processes of tsunami development: earthquake generation, propagation, and inundation (Burwell et al, 2007; NOAA Center, 2014). It was named as a method of splitting tsunami since it includes the three separate processes. The MOST model was selected to represent the hazard analysis for this research for several reasons. First of all, the MOST model uses high bathymetric and topographic resolution data, which has ensured the accuracy of its final outputs, i.e., the inundation zone (Department of Conservation, 2014). Second, the MOST model is the standard model for the NOAA Center for Tsunami Research (NCTR) where it assists the Tsunami Warning Center with forecasting operations (NOAA Center, 2014a). This is done through running sequences of MOST model simulations where the information about a new earthquake is used as input for the model simulation. The MOST model has also been used extensively in both laboratory experiments and in simulating historical tsunamis (NOAA Center, 2014). In addition many research groups have chosen the MOST model for tsunami mitigation research due to its capabilities in modeling tsunami inundation (Burwell et al, 2007; NOAA Center, 2014a). The MOST model works through producing several tsunami development processes including generation, propagation and inundation (Titov and Gonzalez, 1997). Thus, it is critical to highlight how the MOST model simulates these tsunami phases to get the final tsunami inundation output. The first phase, generation, is based on the disturbance of surface water caused by an earthquake or sub-marine landslide. The MOST model simulates this process 32

49 through utilizing the location and the magnitude of local and distant historical earthquakes to estimate the vertical displacement of the ocean floor, which initiate the tsunami wave (Burwell et al, 2007; Wilson et al, 2008). For the tsunami propagation phase, a near-shore bathymetric nested grid in which resolution varies from three arc-seconds (90m) to one arc-seconds (30m) is usually used, depending on the availability, to simulate this process (Department of Conservation, 2014). The third phase of the MOST model simulation involves inundation modeling where various inundation run-ups of different earthquake sources are combined into one inundation area (Wilson et al, 2008). To improve the result of using a 90m inundation grid, a higher resolution DEM (10m) is employed to define the tsunami inundation line of the combined inundation scenarios. Digital imagery and terrain data are usually used to enhance the location of the inundation line using the GIS as a platform to digitize the inundation line. As a final verification of the MOST model output, the inundation line is usually checked through fieldwork and is adjusted if needed (Wilson et al, 2008; Department of Conservation, 2014). The inundation line produced from the MOST model represents the hazard analysis for a tsunami and is generally represented as a map (Figure 3.1). Figure 3.1 The inundation line produced from the MOST model (Department of Conservation, 2014) 33

50 3.2 Demographic and Transportation Data The U.S. Census data is publicly available data that provides various spatial, temporal, and demographic datasets organized at multiple scales (i.e. block, block group, census tracks, county and state levels). Mainly, all of the U.S Census data used in this dissertation will be for year 2010 and the block group level will be used as the unit of analysis. The 2010 U.S. census was selected because it contains both the most current and reliable data, applicable to the needs of this study. Almost all of the U.S. Census data can be downloaded in a Microsoft Excel spreadsheet format with a field called (GEO.ID), which allows the tabulated census details of each block group to be spatially joined to the same block group in order to be mapped. The Geographic Information System (GIS) has the capability of representing the census data spatially. Once the data was configured, it was then related to other phenomena for further analyses. There are several varieties of census data used in this dissertation and discussed below, e.g., the total number of households and vehicles for each block group; and, demographic data such as age, gender, race, and income. In this research, the total number of households for each block group in Orange County, California was downloaded from the U.S. Census website to identify the spatial distribution of households within the evacuation and shadow evacuation zones. The data for the number of households represents the total number of households with specific details about the occupied and vacant households for each block group. A second dataset used in this research is the vehicle ownership data, which represents the total number of vehicles per household in each block group. This data was downloaded from the NHGIS (National Historical Geographic Information System) website ( Assessing the total number of vehicles was necessary since this information was needed to predict the total number of evacuating vehicles (Chapter 6) when modeling the different evacuation scenarios (see Chapter 7 for greater detail). Demographic data was also used in this research. There are various demographic variables at the block group level available through the U.S. Census website. However, the selection of those variables used in this research (age and gender) was based on the results of what would become statistically significant among the variables in the logistic regression analysis (Chapter 6). The age and gender data are available in one Excel spreadsheet where the 34

51 total population for each block group is classified by age and gender. Not all of the age groups were included in the analysis. As stated in Chapter 6, the age group over 65 was included in the analysis since this age group was statistically significant in the logistic regression model output. Road data was used in this research to model the evacuation scenarios. The Orange County road dataset was bought from Korem Company, a company specializing in geospatial technologies. Originally, the Orange County road data was collected by HERE, a company that specializes in mapping and location intelligence. HERE produces NAVSTREETS street data with the most accurate geometry, the highest number of attributes, and the most complete, detailed coverage. The NAVSTREETS data represents the ideal dataset for the high accuracy and optimization required in route planning and GIS application (Street data manual, 2014). Initially, the reason for using the U.S. Census Topologically Integrated Geographic Encoding and Referencing (TIGER) line road data in this research was to model the evacuation. However, since the TIGER line road data was missing a number of attributes, which were necessary to model the evacuation direction of travel, speed, and number of lanes etc. the NAVSTREET data was used to build up the road network and model the evacuation scenarios. Having a complete dataset in terms of geographic coverage and attribute detail was a requirement for model evacuation. 3.3 Limitations of the Data MOST Model Limitations Despite the promising potential that the MOST model provides, there are some limitations to the inundation maps produced by this model. One of the major drawbacks of the MOST model is related to the accuracy of the inundation line which, in turn, must rely on the validity of the topographic and bathymetric data, the availability of terrain data, and the tsunami source information (Department of Conservation, 2014; Mahmood et al, 2012). Having accurate input data such as high resolution DEM would insure the accuracy of the MOST model outputs. Model initialization parameters, deformation of the tsunami source, and the DEM may provide the potential for error in the MOST model maps (Venturato et al, 2007; Mahmood et al, 2012). 35

52 Critical errors in the amount of the inundation might occur due to the set up of the initial conditions that were used to simulate the inundation process. The slip rate, the location of the potential source of the tsunami, and the number of potential earthquakes are important factors in determining the initial conditions of the inundation simulation (Legg et al, 2003). An example of an error originating from initial conditions is that of the simulated grid not covering the entire extent of the initial deformation. An error of this nature would render the inundation scenario misrepresentative in modeling against a real tsunami event since it would inaccurately estimate tsunami energy produced from an earthquake (Venturato et al, 2007). The inundation line produced from MOST model might not be accurate. Since the MOST model uses the DEM in its simulation to represent the inundation line, the resulting calculations for a given scenario may not be accurate. This is depending on the DEM that been used in the simulation process where the DEM might have quantitative errors that may derive from datum conversion and unknown inherent errors produced by combining multiple data sources (Venturato et al, 2007). In addition, the MOST model map that depicts the inundation line in this study represents not only one tsunami event; instead, it represents a combination of several inundation events, from various tsunami sources. Therefore, not all of the inundation area defined in this study will likely be affected from a single tsunami event: the inundation line represents the worst case scenario if a tsunami occurred in the same location (Department of Conservation, 2014). Highlighting the limitations of the MOST model is necessary, especially when, ultimately, MOST model maps are intended to be used for further analyses -- as in this research for tsunami evacuation modeling Census and Road Data Limitations Research that utilizes census data should insure that all values of such data used in the analysis are available to avoid any error that might appear due to missing values. Almost all of the U. S. Census data, including the number of households, the number of vehicles, and age and gender data, that were used in this research did not have any missing values. The NAVSTREETS data that was used in this research also had all the required attributes needed to build up the road network and simulate the evacuation scenarios. This section reviews the major data sources used in this study except for the questionnaire survey data, discussed in Chapter 5. The limitations of different data types have been addressed and issues related to this research have been covered. 36

53 The next section introduces the study area, the history and tsunami evacuation planning in Orange County, California. 3.4 Study Area Orange County, California serves as the study area for this project. It was chosen due to its location in the southern part of California where most of the area s active seismic faults exist. Barberopoulou et al (2011) stated that the Catalina fault, west of Santa Catalina Island, in addition to the Lasuen Knoll, that lies submerged to the southeast of San Pedro Bay, may produce future tsunamis affecting the coastal areas of Orange County and northern coastal areas of San Diego County. Contributing to tsunami risk factors in this area are the topographical characteristics of Orange County, specifically its low elevated areas along the coast. The flat coastline of Orange County can expose it to critical damage and flooding from -- at a minimum - - a 4 meter tsunami (Borrero et al, 2003). Also, Lin et al (2013) have estimated that almost 750,000 people of Long Beach and Orange County will need to be evacuated from the coastal areas if a large tsunami hits California. There are several local and distant tsunami sources such as Chile, Alaska, Japan, and the Cascadian sub-duction zone; the Newport-Inglewood Fault might as well affect the coastal areas of Orange County (CEMA, 2009). This county, with a total population of 3,010,232, represents the third largest populated county (12.4%) in California according to the 2010 U.S. Census. Furthermore, the population of the Orange County is considered to be diverse demographically: 74.5% of its population is white, 2% black or African American, 1.1% Asian, and 34.1% is Hispanic or Latino (U.S. Census, 2013). This demographic diversity represents a somewhat ideal situation for studying the role of social vulnerability in tsunami evacuation. The county is divided into seven sub-divisions: North Coast, Central Coast, South Coast, Mission Viejo, Silverado, Irvine-Lake Forest, and Anaheim-Santa Ana-Garden Grove (Figure 3.2). All of the three coastal sub-divisions of Orange County are at risk of tsunami inundation from both local and distant tsunami sources. Another reason for selecting Orange County as a study area for this research project is the availability of the necessary data in different forms for various types 37

54 of analyses to plan for tsunami evacuation. These data forms include the tsunami inundation zones produced from the MOST model, census data, and road network data. Figure 3.2 Location of Orange County and its subdivisions 38

55 3.4.1 Tsunami History in Orange County, California Historically, California has had 14 tsunamis since 1812 where the heights of the waves were higher than three feet. One of California s major tsunamis followed the 1964 Alaskan earthquake and caused 12 deaths, with damages totaling to at least 17 million dollars. And, because the distance from Cascadia to Southern California is nearly equivalent to the distance between Alaska and Crescent City, the impact of an earthquake event from Cascadia may cause significant consequences in Southern California, similar to the impacts that resulted from the 1964 Alaskan earthquake. For Orange County specifically, four-to-five-foot tidal surges hit the Huntington Beach area from that event causing moderate damage. The entire coastline of Orange County is at risk of tsunami run-ups since large destructive waves of unknown source have been reported to have hit the area from Malibu to Laguna Beach. During that event, a run-up of 260 meters occurred inland of Newport Beach, with waves three meters above mean high tide level (Paradise Lost, 2008; Tsunami Annex, 2009). Dr. Jose Borrero, University of Southern California, validated the probabilistic assessment of predicting wave highest done by A. W. Gracia and J. R. Houston (1975) where the highest tsunami wave could occur from a100-year tsunami event and reach a 9.2-foot maximum, while for a 500-year event, the maximum wave height may reach the maximum of 16 feet. According to Jose Borrero and Costas Synolakis in their report, Tsunami Inundation Mapping, Field Survey Report and Final Recommendations for Orange County, the worst case Tsunami scenario in Orange County, California, is a wave of approximately 10 meters or feet (Tsunami Annex, 2009). Having evidence of historical tsunami events, even if not major ones, in addition to the probabilities of tsunami occurrences in the future requires the attention of both responsible administrative officials and the foundation of sound research directed toward planning for such events Tsunami Evacuation Planning in Orange County, California Following the great 2004 tsunami, countries around the world, including the United States, started working on improving tsunami warning systems and enhancing tsunami evacuation plans. Several warning systems were developed by NOAA and USGS to reduce the 39

56 impact of an unexpected tsunami that could result from an earthquake or sub-marine landslide. As already noted, given the similarity in distances between Alaska and Crescent City and between Cascadia and southern California, it is expected that any strong earthquake that occurs in Cascadia may result in serious consequences along the southern California coastal region -- similar to those that occurred in Crescent City after the 1964 Alaskan earthquake (Tsunami Annex, 2009). It will take between five and nine hours to warn the public in Orange County of distant tsunami events; whereas it would likely take not more than 15 minutes to warn the public of the coastal areas of Orange County in case of a local tsunami event. All of the coastal residents of Orange County would have to be evacuated if a large tsunami were to hit the area. Emergency Alert System (EAS) sirens, route alerting (vehicle mounted public address systems), Alert OC (Orange County), and media releases would be used to broadcast emergency information, along with warning-and-protective-action instructions for a tsunami event to the general public ( Tsunami Annex, 2009). In most states, almost all emergency planning and response activities are developed at the local level (e. g. county or city). However, emergency management agencies at the state level are usually responsible for the coordination among local emergency management agencies, law enforcement agencies, and transportation agencies. In Texas, for instance, specific hurricane evacuation planning has been left to local jurisdictions, which must, nevertheless, follow the state emergency management evacuation plan (Urbina and Wolshon, 2003). Each city of Orange County will be responsible for analyzing the amount of time needed for evacuation and decision-making based on the data from different sources (Paradise Lost, 2008). The emergency response command structure differs from one state to another. However, by law, the ultimate authority to order an evacuation belongs to the governor of a state. Some governors entrust this authority to officials at the local levels -- i.e., the mayor, city council, city sheriff, county judge, or county president -- due to their knowledge of the local characteristics and their information on current local conditions (Urbina and Wolshon, 2003). In Orange County, it is the responsibility of the local government to develop a plan that meets the county, state and federal standards for tsunami evacuation in case of a local or distant tsunami source. The concept and the procedure of the evacuation plan should follow the guidelines described by NOAA, the Tsunami Hazard Mitigation Program, NIMS, SEMS, CalEMA, OC EOP, and the 40

57 County EOP. Also, in Orange County, local jurisdictions are responsible for determining the number of residents who must be evacuated. There are various factors affecting evacuation orders, such as time of day, time of the year, road conditions, the tourist population, sporting and social events (Tsunami Annex, 2009). The Orange County Sheriff Department, Emergency Communication Bureau, and Control One are the centralized points for warning both the County of Orange and the Orange County Operational Area. Notifying appropriate county agencies, city warning points, and the operation area emergency management staff for the county is the responsibility of personnel at Control One. Alerting and warning the public are the responsibilities of the law enforcement agencies throughout Orange County. Selecting the evacuation routes and areas for the various communities is the responsibility of local law enforcement officials within each jurisdiction. As soon as the evacuation status has been determined, the population will be asked to leave the tsunami inundation hazard zone using the designated evacuation routes (Tsunami Annex, 2009). Measuring the clearance time for tsunami evacuation is both challenging and critical due to the complex interactions of a variety of factors that feed into the evacuation process. 41

58 CHAPTER FOUR TSUNAMI EVACUATION AND HAZARD ANALYSIS One of the major challenges of tsunami hazard analysis research -- in addition to both the limited accuracy of Digital Elevation Models (DEM) and the bathymetric data used to model tsunami inundation -- is the reality that tsunami events are rare. These challenges have prompted studies that have employed the use of simulation models in which different data inputs and modeling approaches, such as (Post et al, 2009; Taubenbock et al, 2009; Lammel et al, 2010; Muhari et al, 2011; Imamura et al, 2012; Mas et al, 2012; and Kim et al, 2013), were used to study various tsunami inundation scenarios. While this dissertation doesn t focus on producing a new tsunami modeling approach, it mainly utilizes the product of an existing tsunami model, one created for the purpose of simulating tsunami inundation from different tsunami events: the Method of Splitting Tsunami (MOST) Model. This chapter describes the MOST Model results (maps) produced by the cooperative work of the Tsunami Research Center of the University of Southern California and the California Emergency Management Agency for the purpose of identifying tsunami evacuation and shadow evacuation zones in Orange County, CA. For the purposes of this researcher, the inundation area specified on these maps serves as the basis for further delineating evacuation zone and shadow evacuation, areas based on the distance of the farthest inundation point, as determined by the MOST Model, to the coast of the Orange County region. Defining these zones provides the necessary context for identifying the number of households and vehicles per household in each. The evacuation and shadow evacuation zones defined in this chapter are foundational to subsequent components of the study; and, as this chapter serves to describe the hazard analysis portion of the tsunami evacuation analysis, it also contributes to the chapters of this dissertation that follow by specifying the spatial extent of different variables within the zones. Utilizing accurate inundation maps that also show evacuation zones will increase the accuracy of the hazard analysis and evacuation modeling. 42

59 The objectives of Chapter 4 are: (1) delineate the tsunami evacuation zone using the MOST Model tsunami inundation maps for Orange County, California; (2) specify the extent of the shadow evacuation for a tsunami hazard in Orange County, California; (3) identify the number of households and their spatial distribution for each block group within the evacuation and shadow evacuation zones; (4) identify the total number of vehicles per household for each block group within each evacuation zone. 4.1 Delineate Evacuation and Shadow Evacuation Zones This chapter will specify the spatial extent of the evacuation and shadow evacuation zones for Orange County, California by converting the MOST model maps to a GIS format for further analyses. One of the major components of the evacuation planning analysis is conducting a hazard analysis in which the possible affected area from a hazard can be highlighted. The Method of Splitting Tsunami (MOST) model was selected to identify the possible affected areas in the event of a tsunami inundation (See Chapter 3 for a discussion of the MOST Model). It is important to note that the original hazard analysis, in terms of developing the model to define the final tsunami inundation zone, was conducted by the Tsunami Research Center of the University of Southern California in cooperation with the California Emergency Management Agency, through applying the MOST model on the coastal areas of California. However, since the tsunami evacuation planning process extends to vulnerability, behavioral and transportation analyses, and, given that the output of the MOST model will be further analyzed in GIS for this research project, it was necessary to identify the capabilities of the MOST model and how they work (See chapter 3 for details) Convert MOST Model Maps to GIS Format (Evacuation Zone) The MOST model maps represent the extent of the tsunami inundation results from multiple scenarios. This means that the full extent of the MOST model result covers the maximum inundation area that a tsunami might cause. Therefore, the extent of the inundation area in the maps produced from the MOST model for Orange Country will be used as the evacuation zone (Figure 4.1). The MOST model maps are available in (PDF) format on the 43

60 website of the Department of Conservation ( There are several steps in the conversion of the MOST model inundation area from (PDF) to a single shapfile in order to perform a tsunami evacuation analysis. First, a total of six (PDF) files, representing the extent of the tsunami inundation along the coastal area of Orange County, were downloaded (Figure 4.2). These (PDF) files were converted into (JPEG) format using Paint Software from Windows Accessories in order to be displayed in ArcMap. These six (JPEG) files were then added into ArcMap to digitize the inundation zone (evacuation zone). However, since these (JPEG) files do not have spatial reference, they were geo-referenced using Orange County shapefiles. The Mosaic tool was used to merge all six geo-referenced (JPEG) files into one in preparation for the digitizing process. The Orange County boundaries shapefile was used as a reference to select random control points over the entire county. The digitizing tool was used to digitize the inundation area, the pink area as shown in Figure 4.3. The digitized inundation area represents the evacuation zone for Orange County from various local and distance tsunami sources. This digitized area, i.e., evacuation zone will be used later to identify various socioeconomic variables for further analysis. Figure 4.1 MOST model inundation with USGS 24K Quads for Orange County, CA Source: /Orange/Pages/Orange.aspx. 44

61 Figure 4.2 Orange County tsunami inundation maps produced using MOST model Source: Orange/Pages/Orange.aspx. 45

62 Figure 4.3 Sample of MOST model map of Orange County, CA. The pink area represents the maximum inundation, calculated by combining the run-up of several tsunamis from various local and distant sources. The accuracy of the inundation area in this map is subject to limitations in the accuracy and the completeness of available terrain and tsunami sources information in addition to the current understanding of tsunami generation and propagation phenomena as expressed in MOST Model. Source: Orange/Documents/Tsunami_Inundation_LosAlamitosSealBeach_Quad_Orange.pdf 46

63 4.1.2 Define Shadow Evacuation Zone The evacuation zone was identified based on the extent of the inundation area produced from MOST Model maps. However, in terms of the shadow evacuation zone, there is no existing data or product that represents the shadow evacuation zone in the event of a tsunami run-up in Orange County, California. Since the evacuation zone covers almost the entire Orange County coast, the shadow evacuation will cover the same area to a different extent. Defining how far the shadow evacuation zone extends inland was determined based on the distance of the farthest inundation point of the MOST model output (Evacuation Zone) to the coast: a distance of 7,254 m (See Figure 4.4). The buffer tool was used to create a buffer of 7,254 m around the digitized inundation zone (Evacuation Zone); it (the buffer) was created to insure that the area of the shadow evacuation zone would equal double that of the evacuation zone. The use of the buffer tool around the evacuation zone also insured the greatest similarity possible between the shape of the shadow evacuation zone and that of the evacuation zone (See Figure 4.5). Figure 4.4 The distance of the farthest inundation point (7,254 m), evacuation zone 47

64 4.2 Number of Households in Evacuation and Shadow Evacuation Zones Identifying the evacuation and shadow evacuation zones is a necessary component of the hazard analysis for calculating a tsunami hazard in Orange County. As mentioned above, the original set up of the hazard analysis was done using the MOST model through the collaborative efforts of the Tsunami Research Center of the University of Southern California and the California Emergency Management Agency. Extending the results of the MOST model through further analysis -- by integrating the hazard analysis results with the number of households and number of vehicles per household -- will assist in specifying the vulnerable population and its distribution. Such specification will lead to creating a general overview of the evacuation needs in the county; locating the vulnerable population is necessary in order to conduct the behavioral and transportation analyses as part of the evacuation planning a process to be done through selecting the sample population, evaluating responses in different tsunami scenarios, and then generalizing their responses to the whole population at risk of tsunami inundation. The 2010 US Census number-of-household data, at the block group level, was used to define the number of households in each block group. This Census data consists of three main attributes: total number of households, total occupied households, and total number of vacant households for each block group. The household census data table was joined to the Orange County, CA block group shapefile through GEOID field to define the spatial distribution of the households within the county. Joining the US census household-data table to the Orange County block group shapefile led to the ability to identify spatially the number of household for both the evacuation and shadow evacuation zones, by using the total number of household field from the joined household block group shape file. This was done by using the joined shapefile to select those polygons with centroids located in either the evacuation or shadow evacuation zone. (See Figure 4.6). Horner and Murray (2004) estimated the size of the population within the service coverage for bus transit using centroid of census block, block group, census tract and areal interpolation. Using these approaches revealed there is a tradeoff when deciding on the method to be used for analysis. The areal interpolation method is not stable across aggregation levels, and the polygon centroid method is sensitive to the scale changes in terms of the census geography used to represent the demand. However, since the scale of analysis in this research is 48

65 stable (block group level) among various evacuation analyses, the polygon centroid method was selected to identify the block group polygons regardless of their zone of location, whether evacuation or shadow evacuation. Figure 4.5 Tsunami evacuation and shadow evacuation zone for Orange County, CA 49

66 Table 4.1 Total number of household per block group for evacuation and shadow evacuation zones Zones Evacuation Zone Shadow Evacuation Zone Number of Polygons Total Number of Household Per Block Group Total Number of Occupied Households Total Number of Vacant Households 71 42,490 36,192 6, , , ,140 Furthermore, maps were produced from the combined data to locate the concentrations of the number of households within each zone (See Figures 4.7 and 4.8). The map shows that the block group polygons in the middle area of Orange County consist of the highest number of households in both the evacuation and shadow evacuation zones. However, when comparing the number of households in the evacuation zone to the number of households in the shadow evacuation zone, it is clear that the shadow evacuation zone has a larger quantity of total number of households in each block group. Identifying the number of households for each block group in both the evacuation and shadow evacuation zones will be used later in calculating the evacuation participation rate in Chapter Number of Vehicles per Households in Evacuation and Shadow Evacuation Zones The same method that was used to join the number of households to the Orange County block group shapefile was used again to join the number of vehicles per household. The 2010 US Census number-of-vehicles, per-household, data consists of both the total number of vehicles per household and the number of households that have one-to-five vehicles, separately, as shown in Table 4.2. Joining the number of vehicles per household to the block group in both the evacuation and shadow evacuation zones led to identifying the total number of vehicles per household in both zones as shown in Figures 4.9 and Specifying the total number of vehicles per household for each block group in both the evacuation and shadow evacuation zones was a necessary step in undertaking the research described in Chapter 6: to calculate the evacuation participation rate for each block group in both zones. 50

67 Figure 4.6 Centroids of the US Census block group polygons within evacuation zone Figure 4.7 The concentration of the number of households within the evacuation zone. 51

68 4.8 The concentration of the number of households within the shadow evacuation zone Table 4.2 Total number of vehicles per household for evacuation and shadow evacuation zones Zones Number of Polygons Total Number of Vehicles Evacuation Zone Shadow Evacuation Zone

69 Figure 4.9 The total number of vehicles per household in the evacuation zone Figure 4.10 The total number of vehicles per household in the shadow evacuation zone 53

70 The hazard analysis described in this chapter set up the basic spatial extent of the study area at risk of tsunami, and, as such, provides the context for analyses described in the chapters that follow. It included delineating the evacuation and shadow evacuation zones, in addition to identifying both the total number of households and vehicles per household in each block group. 54

71 CHAPTER FIVE BEHAVIORAL ANALYSIS AND SURVEY QUESTIONNAIRE Behavioral analyses are a necessary component in designing new and improving on existing tsunami evacuation plans. Without such analyses, there are no bases for predicting population behaviors during an evacuation. There are several factors that should be included in a behavioral analysis; these include: the evacuation participation rate, evacuation times, public shelter use rates, evacuation destinations, and vehicle use (Baker, 2000; CFRPC, 2010b). Surveys are considered the main data source for a behavioral analysis; they measure the population s responses based on a hypothetical hazard event or their experience with a hazard. In this study, the goal of the behavioral analysis is to measure how the surveyed population would respond in different, hypothetical tsunami scenarios should Orange County Officials advise the population to evacuate. This study goes beyond the simple decision of whether to evacuate. Through a survey, crucial variables are determined that characterize an at-risk population: the number of cars intended to be used in the evacuation, their chosen destination location, and the evacuation route. Previous studies of tsunami evacuation focused on the daily activities of the surveyed population, or, on their intention to evacuate or not (Charnkol and Tanaboriboon, 2006; Taubenbock et al, 2009; Lammel et al, 2010); but, they did not include information on the type of refuge that they would seek, the location of the refuge, number of vehicles available at their households, number of vehicles planned for use during an evacuation, and information about the evacuation routes, as this research, reported here, has done. For this dissertation, a telephone survey was created to measure the behavioral responses of the Orange County population to different hypothetical tsunami scenarios. The telephone survey method was selected for several reasons. When compared with the interview survey for instance, it provides more honest responses, gives the researcher greater control, saves time and money, and ensures fewer incomplete questionnaires (Babbie, 1998). The survey will be divided into three sections: 1. Information Sources and Resident Awareness in Orange County, California, which focuses on the subject s knowledge of the tsunami phenomenon in his/her geographic area and their preparedness, 2. Evacuation scenarios, which focus on how the 55

72 population will behave in 10-, 20-, and 30-foot tsunami inundation scenarios, and, 3. Demographic information, which focuses on specifying the demographic characteristics of the sampled population. (See Survey in the appendix). The objectives of Chapter 5 are: (1) to measure the tsunami knowledge of the sampled population and their individual perceptions, (2) to measure the evacuation intentions of the sampled population in evacuation and shadow evacuation zones for 10-, 20-, 30-foot hypothetical tsunami scenarios, (3) to find out the number of vehicles, the destinations, and the routes evacuees would use in case of 10-, 20-, 30-foot tsunami scenarios, (4) to define the demographic characteristics of the sampled population. 5.1 Survey Questionnaire The questions in the survey were designed to capture the knowledge, perception and behaviors of the population regarding a tsunami hazard. The selection of the survey questions was based on a combination of an existing hurricane and tsunami research that studied the behaviors of the population at risk of tsunami and hurricane (Horan and Meinhold, 2008; Lindell and Prater, 2012). The reason for combining the hurricane research with the tsunami survey design was based on the larger number of hurricane studies done to capture the behaviors of the evacuees, compared to the tsunami studies, which focused mainly on the intention of evacuation (Charnkol and Tanaboriboon, 2006; Taubenbock et al, 2009; Lammel et al, 2010). The Information Sources and Resident Awareness in Orange County, California section of the survey focuses on measuring the population s perception and knowledge of tsunamis in general and the tsunami potential in Orange County, California. Tsunami knowledge questions include the population s knowledge of historical tsunamis, their causes, tsunami indicators, sources of a tsunami (i.e., location of the original earthquake, considered the source) and the estimated arrival time of a possible tsunami from the time tsunami waves are generated due to an earthquake or landslide. Also, this section of the survey asks questions related to access to information that contains tsunami alerts information that would affect the evacuation decision. 56

73 The tsunami scenario and demographic information sections provide the needed data to estimate the evacuation participation rate in this research using the logistic regression model (see Chapter 6 for more details). For the tsunami scenario section of the survey, specifically, the researcher is concerned with knowing if the sampled population would evacuate in case of an evacuation order by government officials of Orange County in the events of tsunamis of 10, 20, and 30 feet (e.g., questions 23, 38, and 52 of the scenario section of the survey). According to Osborn (2014) and Tsunami Annex (2009) the worst-case scenario for such an event in Orange County would be one involving a 32-foot inundation; therefore, the researcher chose to measure the resident behaviors based on 10-, 20-, and 30-foot scenarios. Survey questions in this regard are important to get a sense of the extent to which people are willing to participate in different tsunami evacuation events. The responses will then be used to predict the participation rate for the entire inundation population. Moreover, the tsunami scenarios section of the survey assists in identifying the final destination, vehicle availability within each household, and the number of vehicles available for use by each respondent in the evacuation process. Identifying this data was important for evacuation modeling (Chapter 7) where it would be used to identify the final destination for most evacuees, and the number of vehicles out of those available -- that would be used for evacuation. This last piece of information, i.e., vehicle use, will be used to estimate the percentage of vehicles (Chapter 6) that will be used by the population at risk of inundation during the specified scenarios. Such data is necessary for transportation modeling since the number of vehicles used in evacuation will affect the evacuation clearance time. Responses to the remainder of the questions within this section of the survey explain the sampled population s reasons for their answers. These questions aim at revealing the following types of information: type of refuge (hotel/motel, house of friend, or public shelter), distance the refugee is from the safe zone, the location of the refuge (if it is in the respondent s neighborhood or outside, if it is within Orange County or outside), the major road(s) likely to be taken for evacuation. In addition to these survey elements, there are questions requiring responses that indicate knowledge of vertical evacuation, and whether the respondent is planning to use such evacuation in case of a tsunami event. All of these questions assist in explaining the evacuee behaviors during different evacuation scenarios. The third section of the survey, which focuses on the demographic 57

74 information of the sampled population, will help in identifying personal characteristics of the sampled population who have responded by saying they would or would not evacuate in case of a tsunami event. It may, in addition lead to identifying specific trends in the evacuation decision. This information includes age, gender, race, marital status, education, income, number of people within the household, and type of household. The survey was conducted by the Kerr and Down research staff, to whom the researcher provided the GIS layers that represent both the evacuation and shadow evacuation zones for sampling purposes produced from data contained in the hazard analysis chapter. The population sample size was 235 with a 90% confidence level and a ±5.36 error rate. The selection of the sample size of 235 was due to the limited funding for this research, i.e., collecting more samples would require more financial resources. A total of 185 samples were collected within the evacuation zone and 50 samples were collected within the shadow evacuation zone. The reason for selecting the sample quantities of 185 in the evacuation zone and 50 in the shadow evacuation zone was based on the understanding that the majority of the sample should be in the evacuation zone and that a portion of the sample must include respondents from the shadow evacuation zone. A simple random sampling was used, taken from published landline telephone listings. The Kerr and Down research staff recorded the answers of the sampled population along with the (x,y) coordinates of every respondent. The final output of the telephone survey was received in an Excel sheet that included all the answers along with the coordinates of each respondent. Statistical analysis will be performed to measure the variations within each variable. Generally, the output of the survey will be used to identify the demographic variables that may contribute to the evacuation decision-making process. Identifying these variables will contribute to understanding the factors that affect the participation rate, which in turn controls the evacuation clearance time. 5.2 Information Sources and Residents Awareness in Orange County, California: Tsunami Knowledge, Beliefs and Perceived Vulnerability General Knowledge Tsunamis are considered rare, natural disaster events; they occur suddenly and have highly destructive potential. Knowledge of their causes, sources, and their history is a necessary 58

75 context for developing a means to measure a population s understanding of tsunami events. According to the NOAA Center for Tsunami Research (2014), earthquakes are the major cause of tsunamis. One of the most destructive tsunamis to hit the California coast resulted from the 1964 Price William Sound event (California Geological Survey, 2014). Most of the respondents in households surveyed in this study recognize that earthquakes are the major cause of tsunami events; however, somewhat more than half of the respondents did not know when the last destructive tsunami had hit the coasts of California (See Tables 5.1 and 5.2). Table 5.1 Q1) What is the most frequent cause of tsunamis? Evacuation Zone % Shadow Evacuation Zone % (n=185) (n=50) Landslide 3% 0% Earthquake 81% 82% Hurricanes/storms 4% 4% High tides 6% 8% Don't know 6% 6% Table 5.2 Q2) When did the last tsunami cause damage to coastal areas in California? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) % 2% % 2% % 12% % 30% Don't know 50% 54% Tsunami Perception and Beliefs Individual perceptions and beliefs are affected by experience, educational level, and a variety of other factors; people s reactions to natural disasters, though these reactions may vary 59

76 in intensity when compared to occurrences of everyday life, are based as well on knowledge of, and how they interpret such events (Shaw et al, 1992). In order to measure the perceptions of the residents of Orange County, each survey respondent was asked about the location of his/her household relative to the evacuation zone, the height of the first floor of his/her household, indicators of a tsunami, the source location of both a local and distant tsunami, the estimated arrival times of different tsunami types, the reasons for his/her evacuation decision if an earthquake were felt, reasons for a decision not to evacuate. The resulting responses were used to evaluate tsunami perceptions of the Orange County population. Table 5.3 summarizes the results by stating that more than half of the sampled population in the evacuation zone thought they were living in a tsunami inundation hazard zone, and only around 5 of the respondents in the shadow evacuation zone thought they were living in a tsunami inundation hazard zone. Table 5.3 Q12) Do you live in a tsunami inundation hazard zone? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 69% 10% No 9% 62% Don t Know 22% 28% Table 5.4 Q13) How high is the first floor of your house from sea level? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) 1-5 ft 39% 12% 5-10 ft 23% 6% ft 13% 6% More than 15 ft 12% 52% Don't know 12% 24% 60

77 Table 5.5 Q14) Which of the following signs might warn you about possible tsunami? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Strong earthquake 61% 70% Rapid rise/fall coastal water 23% 22% Storm surge 8% 2% Strong wind 5% 2% None of the Above 3% 4% The surveyed households of Orange County were able to recognize the signs of an impending tsunami -- an earthquake and a rapid rise and fall of coastal water -- in addition to identifying the sources of local and distant tsunamis as shown in Table 5.5. Respondents were introduced to the definitions of local and distant tsunami sources: i.e., respectively, a tsunami where the source is within 1000 kilometers (about 621 miles) of the area of interest; and, a tsunami where the source is greater than 1000 kilometers from the area of interest (NTHMP, 2014). Only a small proportion of the survey respondents were able to predict the estimated arrival times of local and distant tsunami waves in a manner close to reality: a local tsunami may require only a few minutes to reach an area at risk; a distant tsunami can take several hours to reach an area such as that of coastal Orange County, CA (see Tables for more details). Table 5.6 Q16) Which of these locations is/are considered to be a significant source of earthquake that might generate a local tsunami along the coast of Orange County? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Chile 5% 4% Alaska 15% 6% Catalina Island 45% 42% All of the above 19% 34% None of the above 2% 0% 61

78 Table 5.7 Q17) Which of these locations is/are considered to be a significant source of earthquake that might generate a distant tsunami along the coast of Orange County? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Chile 10% 12% Alaska 29% 16% All of the above 37% 38% None of the above 13% 18% Don't know 11% 16% Table 5.8 Q18) How long does it take a tsunami waves generated from local earthquake or landslide to arrive to Orange County? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Couple of minutes 23% 14% Half an hour 23% 20% One hour 15% 22% More than 1 hour 19% 16% Don't know 21% 28% Table 5.9 Q19) How long does it take a tsunami wave generated from a distant earthquake to arrive to Orange County? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) One hour 17% 6% 6 hours 36% 30% 12 hours 16% 22% 24 hours 11% 12% Don't know 19% 30% 62

79 Table 5.10 shows that not all of the sampled population would evacuate if they felt an earthquake (sign of a potential tsunami). However, they also reported that if they were to evacuate, it would be in response to an official order (See Table 5.11). Most of the population also did not specify the reason for not deciding to evacuate if there were a tsunami threatening their area as stated in Table Also, the respondents were asked about their thoughts regarding vertical evacuation as a means to reaching a safe location. The term vertical evacuation is used to define any earthen mound or building that can provide evacuees an elevation high enough above the level of a predicted tsunami inundation to secure their safety (Applied Technology Council, 2009). Almost half of the sampled population knew about vertical evacuation as a means to evacuate from an area at risk of a tsunami, and they specified they would use such a means in case of a tsunami threat (See tables 5.13 and 5.14). Such information provides an initial overview of the surveyed population s general knowledge and perceptions of tsunamis. An opportunity exists for future researchers to build on the findings of this study to determine if, and to what degree, a correlation may exist between individual decisions to evacuate, given varying tsunami scenarios, and his/her general knowledge and perception of tsunamis. Table 5.10 Q20) Would you leave your home to go someplace safer from a possible tsunami if you felt an earthquake? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 43% 24% No 44% 64% Don't know 1% 4% Depends 12% 8% Information Sources There are various sources of information available to people in the event of a natural hazard in Orange County. One of these is the Orange County Alert System (OCAlert), which notifies County residents and businesses of emergencies (Orange County, 2014). While slightly over half of the sampled population knew about this system, not all of them are registered (See Table 5.15 and 5.16). Although most of the respondents have access to the Internet in their 63

80 Table 5.11 Q21) Would you evacuation your home in case of a tsunami evacuation warning because of? Evacuation Zone % (n=185) Concern about water inundation Shadow Evacuation Zone % (n=50) Yes 74% 44% No 20% 48% Don't know 6% 8% Home not well built/home at low elevation Yes 72% 40% No 25% 48% Don't know 3% 12% Official said evacuate Yes 91% 76% No 6% 20% Don't know 3% 4% Military said personnel/families should evacuate Yes 88% 72% No 8% 24% Don't know 4% 4% Concern about loss of electricity or water Yes 57% 48% No 40% 52% Don't know 3% 0% Concern about isolation after storm Yes 52% 42% No 42% 52% Don't know 6% 6% Something else Yes 33% 24% No 55% 58% Don't know 12% 18% 64

81 Table 5.11 continued Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Home not well built/home at low elevation Yes 72% 40% No 25% 48% Don't know 3% 12% Official said evacuate Yes 91% 76% No 6% 20% Don't know 3% 4% Military said personnel/families should evacuate Yes 88% 72% No 8% 24% Don't know 4% 4% Concern about loss of electricity or water Yes 57% 48% No 40% 52% Don't know 3% 0% Concern about isolation after storm Yes 52% 42% No 42% 52% Don t know 6% 6% Something else Yes 33% 24% No 55% 58% Don t know 12% 18% households or through use of their cell phones, they do not use the Internet as a major source of information to warn them about possible tsunami events (See Tables ). Furthermore, social media accounts such as Facebook and Twitter, according to respondents, would not be a well-used source of information for warming people in case of a tsunami event in Orange County (Tables 5.20 and 5.21). NOAA radio represents another way of getting information, yet only a 65

82 small portion of the Orange County population uses it to be informed of tsunami events (See Tables 5.22 and 5.23). Instead, most of the Orange County sampled population relies on local television channels for alerts about possible tsunamis. In other words, decisions as to whether to evacuate or not would be made based on information given on local television stations (See Table 5.23). This is probably because in general most hazard warnings, such as those relating to tsunamis accompanied with evacuation advice -- are broadcast on local television channels (Tsunami Annex, 2009). Identifying the most popular or most used sources of information for threats of a possible tsunami is important in order to insure the best means of communicating the warning message. Table 5.12 Q22) What would be the main reason/s you wouldn't evacuate your home in case of a tsunami affecting the coastal areas of Orange County? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Home wouldn't flood 4% 10% Home is well built 5% 6% Not located in area told to evacuate 6% 10% Concern about looting 4% 2% Concern about traffic 4% 6% No transportation 2% 2% No place to go 3% 6% Can't afford it 0% 0% Job/Military obligation 1% 2% Medical conditions 5% 4% Pet 2% 0% Shelter conditions (crowding, etc.) 0% 4% Other 17% 18% Don't know 25% 14% Depends on tsunami magnitude 22% 16% 66

83 Table 5.13 Q66) Do you know what vertical evacuation means? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 48% 50% No 52% 50% Table 5.14 Q67) Are you planning to use it? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 55% 48% No 34% 44% Don't know 11% 8% Table 5.15 Q3) Are you familiar with the Orange County Alert system, sometimes called OCAlert? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 83% 52% No 17% 48% Table 5.16 Q4) Are you registered in it? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 58% 35% No 42% 65% Table 5.17 Q5) Do you have access to the internet at home? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 84% 88% Yes, but not in home 3% 2% No 12% 10% Don't know 0% 0% 67

84 Table 5.18 Q6) Do you have internet access on your cell phone? Evacuation Zone % Shadow Evacuation Zone % (n=185) (n=50) Yes 55% 54% No 43% 46% Don't know 2% 0% Table 5.19 Q7) Does anyone else in your household have internet on their cell phone? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 48% 52% No 50% 44% Don't know 2% 4% Table 5.20 Q8) Do you have a social media account such as Facebook or Twitter? Evacuation Zone % Shadow Evacuation Zone % (n=185) (n=50) Yes 49% 40% No 51% 60% Don't know 0% 0% Table 5.21 Q9) How often do you use social media? Evacuation Zone % (n=91) Shadow Evacuation Zone % (n=20) Several times a day 20% 35% Once or twice a day 30% 25% 3-6 days a week 13% 15% 1-2 days a week 18% 5% Less than once a week 18% 20% Never 2% 0% Table 5.22 Q10) Do you have a NOAA Weather Radio in your home? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 21% 18% No 77% 82% Don't know 2% 0% 68

85 Table 5.23 Q11) When deciding whether to evacuate from a future tsunami, to what extent would you rely on each of the following for information? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) National TV networks (e.g., CNN, FOX News, Weather Channel) Not at all 12% 20% May be 6% 10% Medium 15% 6% Mostly 16% 14% Great deal 50% 48% Don t know/refused 1% 2% Local TV Stations Not at all 6% 10% May be 3% 4% Medium 15% 20% Mostly 19% 10% Great deal 57% 56% Don t know/refused 0% 0% Local radio stations Not at all 19% 14% May be 7% 6% Medium 18% 14% Mostly 16% 14% Great deal 39% 52% Don t know/refused 0% 0% Local Newspapers Not at all 52% 52% May be 8% 8% Medium 10% 20% Mostly 8% 6% Great deal 21% 14% Don t know/refused 1% 0% Websites Not at all 44% 22% May be 10% 6% Medium 19% 30% Mostly 11% 14% Great deal 15% 26% Don t know/refused 0% 2% 69

86 Table 5.23 continued Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Social Media (e.g., Facebook, Twitter) Not at all 63% 56% May be 13% 14% Medium 12% 10% Mostly 4% 4% Great deal 7% 12% Don t know/refused 1% 4% Local Officials (e.g., county judge, mayor) Not at all 32% 36% May be 7% 10% Medium 16% 12% Mostly 13% 12% Great deal 30% 28% Don t know/refused 2% 2% Friends, Relatives, Neighbors, or Coworkers Not at all 17% 20% May be 11% 12% Medium 18% 18% Mostly 18% 18% Great deal 35% 32% Don t know/refused 1% 0% NOAA Watches and Warnings Not at all 35% 40% May be 6% 8% Medium 6% 6% Mostly 5% 10% Great deal 42% 36% Don t know/refused 5% 0% Notification App on Phone Not at all 37% 34% May be 4% 6% Medium 4% 12% Mostly 13% 12% Great deal 40% 32% Don t know/refused 2% 4% 70

87 5.3 Tsunami Scenarios Intention for Evacuation The second part of the survey questionnaire focused on the respondents behaviors regarding three different tsunami scenarios -- wave heights of 10, 20, 30 feet. Respondents were asked about their intention to evacuate in each of the three different scenarios. For each of these scenarios, they were asked questions relating to the following: 1) the refuge they are planning to seek, 2) main roads they would be using for evacuation, and, 3) vehicle availability for evacuation. Table 5.24 indicates that more than half of the population would plan to evacuate for each of the three tsunami scenarios, and that the evacuation participation rate would increase as the tsunami wave increased, from 10 to 20, then 30 feet in both of the evacuation and shadow evacuation zone. Table 5.24 Q 23, 38,and 52) If the government officials suggested an evacuation for a 10, 20, and30 ft tsunami affecting the Orange County coast, would you evacuate your home? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) 10 ft 20 ft 30 ft 10 ft 20 ft 30 ft Yes 72% 86% 90% 46% 52% 56% No 20% 12% 7% 50% 44% 38% Depends 5% 2% 2% 4% 4% 6% Don't know 3% 1% 1% 0% 0% 0% Most of the population would evacuate their homes immediately if an official announcement were made that suggested an immediate evacuation for a tsunami affecting the Orange County coast regardless of the tsunami scenario (See Table 5.25).The respondents who said they would evacuate for each of the three tsunami scenarios were asked about their refuge destination, the road that they envisioned using for evacuation, and the number of vehicles they would use for evacuation. When comparing respondent intentions to evacuate, given the two imagined situations -- one in which respondents would feel an earthquake (Table 5.10) and the other in which there would only be an official evacuation suggestion (Table 5.24) it appears that 71

88 the Orange County population mostly relies on official suggestions rather than on their own perception and experience. Table 5.25 Q24, 40, and 54) How soon would you evacuate your home if the government officials suggested an immediate evacuation for a 10, 20, and 30 ft tsunami affecting the Orange County coast? Evacuation Zone % Shadow Evacuation Zone % 10 ft (n=143) 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) Immediately 39% 40% 43% 40% 39% 42% 5-10 min 22% 20% 19% 12% 14% 13% min 18% 21% 20% 8% 7% 6% More than 15 min 12% 10% 11% 24% 21% 19% Don't know 3% 4% 3% 8% 11% 10% Depends 6% 5% 4% 8% 7% 10% There is a small portion of the sampled population that is uncertain as to their evacuation intentions since they believe finalizing their decision would be based on the following variables: the proximity of the tsunami at the time they receive information of its approach, whether other people need their help, the availability of transportation, the traffic conditions, their experience and their health or medical conditions. All of these factors would affect the timeliness of the evacuation decision of the portion of the sampled population that determined, in all three scenarios, that their decision would be based on an official announcement of a tsunami threat along with a recommendation to evacuate. This indicates effectiveness of the government officials evacuation announcement in Orange County, California Refuge Type As stated above, the respondents who determined they would evacuate for each tsunami scenario were asked about the type of refuge they would seek. Most of the respondents in this category selected the house of friend or relative as a shelter, as shown in Table Most of the respondents within the evacuation zone stated that the refuge will be located within Orange 72

89 County; however, for the respondents within the shadow evacuation zone, they stated that the refuge will be located outside Orange County (See Table 5.27). The evacuating population was then asked about how far they would evacuate inland based on the location of their home in the event officials suggested evacuation, in each of the three-tsunami scenarios. Most of the respondents in the evacuation zone said they would plan to evacuate between one and five miles for all tsunami scenarios whereas in the shadow evacuation zone, the respondents don t know how far they would evacuate inland (See Table 5.28). Table 5.26 Q26, 42, and 56) What type of refuge you would seek if you decided to evacuate your home based on the evacuation suggestion issued by the government? Evacuation Zone % Shadow Evacuation Zone % 10 ft (n=143) 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) Hotel/Motel 14% 14% 16% 12% 11% 10% House of a friend or relative 48% 48% 47% 40% 36% 39% Public shelter 12% 12% 12% 4% 4% 3% Other 15% 15% 14% 16% 18% 16% Don't know 11% 10% 11% 28% 32% 32% Table 5.27 Q27, 43, and 57) Where would that refuge be located? 10 ft (n=143) Evacuation Zone % Shadow Evacuation Zone % 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) Within the neighborhood 3% 4% 3% 4% 4% 3% In your community 23% 21% 20% 20% 18% 19% In Orange County 43% 41% 40% 24% 21% 19% Outside Orange County 20% 23% 25% 36% 36% 39% Don't know 10% 10% 11% 16% 21% 19% 73

90 Table 5.28 Q29, 45, and 59) How far do you think you would need to evacuate inland based on your home location if government officials suggested an evacuation from a 10, 20, and 30 ft tsunami? Evacuation Zone % Shadow Evacuation Zone % 10 ft (n=143) 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) 1-5 miles 34% 33% 33% 16% 14% 16% 5-10 miles 24% 23% 20% 16% 14% 13% miles 12% 13% 13% 16% 14% 16% More than 15 miles 22% 22% 26% 24% 25% 23% Don't know 8% 9% 8% 28% 32% 32% Depends 0% 33% 33% 0% 14% 16% Main Evacuation Roads The respondents who determined they would evacuate in all three scenarios were then asked about the main evacuation road they planned to use. Most of these respondents did not select the evacuation route listed in the questionnaire (Table 5.29) or they do not know which route they would chose to evacuate to their planned safe location. All of the evacuation routes listed in the survey questionnaire are Interstate Highways, main arteries, State Highway, and/or County Highway. According to the survey respondents, the selection of the evacuation route would be mostly based on the distance between the route and the surveyed household, the location of the road relative to the ocean, the respondents knowledge of the route, traffic conditions, and whether the road was the only option to get out of the risk area. Table 5.30 indicates that most of the respondents who decided to evacuate to a safe location, in the event of an official warning, would need more than 20 minutes to get to their planned safe location for both of the evacuation and shadow evacuation zone in the three tsunami evacuation scenarios. Estimating the needed time for evacuees to arrive at their locations of choice is critical to evaluating the capacity of current road networks and to planning for future emergencies. 74

91 Table 5.29 Q31, 46, 60) What main roads would you use if the government officials suggested an evacuation for 10, 20, and 30 ft tsunami impacting the Orange County? Evacuation Zone % Shadow Evacuation Zone % 10 ft (n=143) 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) Highway 5 9% 9% 10% 12% 11% 10% Highway % 15% 14% 20% 18% 16% State Route 1 3% 2% 2% 0% 0% 0% State Route 39 5% 6% 6% 12% 11% 10% State Route 73 3% 4% 4% 0% 0% 0% State Route 133 3% 3% 3% 8% 7% 6% Don't know 19% 18% 18% 16% 25% 26% Other 43% 44% 43% 32% 29% 32% Table 5.30 Q34, 49, and 63) How long do you think it would take you to get to your planned safe evacuation location if government officials suggested an evacuation of a possible 10, 20, and 30 ft tsunami? 10 ft (n=143) Evacuation Zone % Shadow Evacuation Zone % 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) Less than 10 minutes 17% 18% 18% 8% 7% 6% minutes 17% 17% 16% 24% 21% 19% minutes 15% 15% 13% 12% 11% 10% More than 20 minutes 43% 41% 44% 44% 39% 42% Don't know 6% 7% 8% 8% 21% 23% Other 2% 1% 18% 4% 0% 6% Vehicle Usage for Evacuation The respondents who selected to evacuate in the event of any tsunami scenario, regardless of size, were also asked about the number of vehicles available in their households. For the 10-foot tsunami scenario, the respondents in the evacuation zone would have only one vehicle that is available for evacuation whereas, for the 20-, and 30-foot scenarios, most 75

92 respondents would have at least two vehicles in their households that could be used in an evacuation event. For the respondents in the shadow evacuation zone, most respondents would have at least two vehicles in their households that could be used in an evacuation event (See Table 5.31 for details). The respondents then were asked about how many vehicles the household would actually use if they decided to evacuate. Table 5.31 Q36, 50, and 64 How many vehicles would be available in your household that you could use to evacuate? Evacuation Zone % Shadow Evacuation Zone % 10 ft (n=143) 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) 0 31% 5% 4% 4% 4% 6% 1 45% 29% 30% 20% 18% 16% 2 16% 45% 45% 44% 39% 35% 3 1% 17% 17% 28% 32% 35% 4 2% 1% 2% 4% 7% 6% 5 31% 2% 2% 0% 0% 0% Table 5.32 Q37, 51, and 65) How many vehicles would your household take if you evacuated? Evacuation Zone % Shadow Evacuation Zone % 10 ft (n=143) 20 ft (n=163) 30 ft (n=171) 10 ft (n=25) 20 ft (n=28) 30 ft (n=31) 0 9% 10% 9% 8% 7% 10% 1 60% 57% 57% 60% 54% 48% 2 27% 29% 29% 28% 36% 39% 3 3% 3% 3% 4% 4% 3% 4 1% 1% 2% 0% 0% 0% 5 1% 1% 1% 0% 0% 0% Results indicate most of the population within both zones would take only one vehicle for all tsunami scenarios, as shown in Table Finally, for the 20- and 30-foot scenarios, respondents were asked if they would evacuate to the location specified in the previous, 10-foot, scenario. Results showed the respondents would select the same location regardless of the size of the tsunami (See Table 5.33). 76

93 Table 5.33 Q39, and 53) Would you evacuate to the same location in the same way as you would for a 10 or 20 ft tsunami? Evacuation Zone % 20 ft (n=163) 30 ft (n=171) Shadow Evacuation Zone % 20 ft (n=28) 30 ft (n=31) Yes - same as 10 ft tsunami scenario 85% 46% 86% 58% Yes - same as 20 ft tsunami scenario 0% 43% 0% 32% No 15% 11% 14% 10% 5.4 Demographic Characteristics Demographic information identifies characteristics of the surveyed population that are critical in designing an effective evacuation model. Age represents a descriptor that provides an overview of how a population may behave towards an evacuation order. The importance of age stems from the association between the age of the respondents and specific illness and mobility problems, e.g., the elderly. Also, the age of the respondents might be linked to previous experience of the same disaster, a factor that may affect their evacuation decision (Bytheway, 2007). The age of more than half of the sampled population in this study ranged from 45 to 79, as stated in Table Race and ethnic backgrounds also provide significant information about the sampled population and how they might react to a specific hazard (Van Willigen et al, 2005; Riad et al, 1999). Most of the sampled population is classified as white and non-hispanic and the majority of respondents are married in both zones (See Tables 5.35, 5.36, and 5.37). Table 5.34 Q68) How old were you on your last birthday? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) % 12% % 32% % 30% 80 or more 15% 16% Didn t Answer 5% 10% 77

94 Table 5.35 Q69) Which of the following best describes your race? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) White 90% 98% Black 2% 0% Asian 2% 0% Native Hawaiian and other pacific Islander 1% 0% American Indian and Alaskan Native 0% 0% Other 5% 2% Table 5.36 Q70) Which of these best describes your ethnic background? Evacuation Zone % Shadow Evacuation Zone % (n=185) (n=50) Hespanic Origin 9% 10% Non-Hispanic Origin 91% 90% Table 5.37 Q71) What is your marital status? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Married 62% 68% Single 10% 16% Divorced 12% 6% Widowed 16% 10% Gender and education level are other important demographic variables that can lend insight as to how a population may behave in the event of an evacuation order (Whitehead, 2001; Bateman and Edwards, 2002; Lindell, Lu and Prater, 2005). The majority of the sampled population is female in both of the evacuation and shadow evacuation zone and the most of the respondents hold a bachelor degree in the evacuation zone and some college in the shadow zone (See Tables 5.38 and 5.39). 78

95 Table 5.38 Q72) What is your highest level of education? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Less than high school 1% 4% High school graduate 5% 14% Some college 22% 28% Associate degree 7% 6% professional degree 3% 2% Bachelor degree 32% 20% Master's degree 24% 18% Doctorate degree 4% 8% Table 5.39 Q81) Was the respondent male or female? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Male 48% 48% Female 52% 52% Income has also proved a significant demographic factor in predicting behavior towards a hazard event; it is known that higher income groups, when compared to those of lower incomes, tend to evacuate areas under disaster warnings (Zamore, 2014). The majority of the respondents had annual incomes ranging from $50,000 to $99,999 for the year 2013 in both zones (Table 5.40). Identifying respondent s income might indicate the location of the household, i.e., it may be in a poorer neighborhood, or the number of vehicles per household may be limited; income can also indicate type of household structure inhabited by the respondents. A small portion of the sampled population has a disabled person in the home, with the majority of these classified as having a physical disability (See Tables 5.41 and 5.42 for greater detail). 79

96 Table 5.40 Q73) What is your yearly household income in 2013? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Less than $29,999 15% 22% $30,000-$49,999 15% 14% $50,000-$99,999 30% 38% $100,000-$199,999 29% 20% $200,000 or more 11% 6% Table 5.41 Q75) Is there anyone in your household with a disability? Evacuation Zone % (n=185) Shadow Evacuation Zone % (n=50) Yes 14% 22% No 86% 78% Table 5.42 Q76) What is the disability? Evacuation Zone % (n=26) Shadow Evacuation Zone % (n=11) Loss of sight, hearing, smell 15% 0% Physically disabled 62% 73% Mentally disabled 15% 9% Difficulty in dressing, bathing, or getting around inside the home 4% 9% Difficulty in going outside the home alone to shop or visit a doctor's office 4% 9% Difficulty in performing job duties due to disability 0% 0% 5.5 Physical Characteristics Information on the physical location of a household (evacuation and shadow evacuation zones), its distance from the coast, and the elevation of each of the surveyed households was derived from various sources using GIS. The location of each respondent household was provided from the Kerr and Downs Company in (x, y) coordinates as part of the data collection 80

97 process. Using the coordinates of each of the surveyed households, the researcher was able to identify its location, either in the evacuation or shadow evacuation zone. The measurement tool in Arcmap software was used to measure the distance from each surveyed household -- using the location of the household defined in the previous step -- to the coast or nearest body of water connected to the coast to get the distance. The elevation of each household also derived from the USGS (United States Geological Survey) DEM. Most of the surveyed households within the evacuation zone were with a distance of less than 1,929 feet to the coast. However, the majority of the surveyed households in the shadow evacuation zone were with a distance ranges between 1,930 3,859 feet to the coast. The majority of the surveyed households were located in elevations of less than 59 meters in both of the evacuation and shadow evacuation zone (See Table 5.43 for more details about the physical characteristics of the surveyed households). Consistency between the distance and the elevation of surveyed households is expected because as location of the household moves further inland, the elevation of the household also increases. The presented demographic and physical characteristics of the surveyed households in this study will be used in the following chapter (Chapter 6) to identify the significant variables that contribute most to the evacuation decision for each (10-, 20-, and 30-foot) tsunami scenario. Table 5.43 Physical characteristics of the surveyed household in Orange County, CA for the (10, 20, 30 ft) tsunami scenarios Variable Distance (ft) Evacuation Zone (n=185) Shadow Evacuation Zone (n=50) Less than % 30% % 42% % 12% % 12% 7720 and more 9% 4% Elevation Less than % 78% % 6% % 14% % 0% 240 and more 0% 2% 81

98 5.6 Demographic Characteristics of Orange County, CA As stated before in Chapter 3, Orange County, CA represents the third largest populated county (12.4%) in California, according to the 2010 U.S. Census, with a total population of 3,010,232. The goal of including the demographic section in the survey questionnaire in this research is to provide an overview of the demographic characteristics of the sample population. Furthermore, these demographic variables will also be tested against evacuation decisions that were made by the respondents for each of the tsunami scenarios (10-, 20-, and 30-foot) to identify the variable most contributing to the evacuation decision, as described in Chapter 6. However, in order to allow for an assessment of the survey sample used in this dissertation, it was critical to provide a general overview of the demographic characteristics of the county to evaluate the demographic variations of the population within the entire county. The same demographic variables (age, race, gender, education, income, and presence of a disabled person) that were used in the survey questionnaire for this research were downloaded from the U.S. Census website for year 2010 at the block group level. These variables represent the general demographic characteristics for the population of Orange County. In terms of age, the majority of Orange County residents (38%) range between years of age, as stated in Table This age-range conflicts with the age of the sample population, in which the age of the majority of the respondents ranged between in the evacuation zone, and in the shadow evacuation zone (See Table 5.34). Gender is another demographic variable that was used in the survey where the majority of the sampled population is female as shown in Table The female population is also in the majority when Orange County is viewed as a whole (See Table 5.44). While designing the survey questionnaire, the race variable was divided into six categories: White, Black or African American, Asian, Native Hawaiian and other Pacific Islander, American Indian and Alaskan Native, and Others. This division was based on the categories that were defined for race in the Census data. According to this data, most of the population of Orange County assigned themselves within the category White in terms of their racial makeup, and the same trend was noticeable for the survey respondents who chose White as their race (See Tables 5.35 and 5.44). Regarding educational level, according to the U.S. Census data, most of the residents of Orange County have some 82

99 college. This can be compared with the information obtained from the survey, in which the research sample indicated that the majority of the respondents within the evacuation zone held a bachelor degree and that a majority of respondents within the designated shadow evacuation zone had some college (See Tables 5.38 and 5.45). Income is considered to be another demographic variable that might affect the evacuation decision, so it was included in the survey questionnaire. Table 5.46 indicates that most of the population of Orange County has an annual income that ranges between $100,000 and $199,999. However, the majority of the survey respondents have their incomes ranging between $50,000 and $99,999 for both evacuation and shadow evacuation zones (See Table 5.40). In response to the questionnaire item relating to the presence of a disabled person within the household, most of the respondents who answered in the positive, characterized the nature of the disability as one that posed physical disability. The U. S. Census data indicated that the majority of the disable population in Orange County belongs to the category of having difficulty in performing job duties (See Tables 5.42 and 5.47). Table 5.44 The characteristics of age, gender, and race for the population of Orange County, CA % of population in Orange County (n=3,010,232) Age Less than 18 years old 24% years old 38% years old 25% years old 8% 80 years old and over 3% Gender Male 49% Female 51% Race White 64% Black 2% Asian 1% Native Hawaiian and other pacific Islander 18% American Indian and Alaskan Native 0% Other 15% 83

100 Table 5.45 The percentage of the educated people in Orange County, CA Education level % of population in Orange County (n=1,813,456) Less than high school 21% High school graduate 17% Some college 23% Associate degree 8% professional degree 3% Bachelor degree 20% Master's degree 7% Doctorate degree 1% Table 5.46 The income level for the population of Orange County, CA Income % of population in Orange County (n=984,503) Less than $29,999 18% $30,000-$49,999 15% $50,000-$99,999 31% $100,000-$199,999 27% $200,000 or more 9% Table 5.47 Different disability types for the population of Orange County, CA Disability % of population in Orange County (n=755,910) Loss of sight, hearing, smell 9% Physically disabled 19% Mentally disabled 12% Difficulty in dressing, bathing, or getting around inside the home 7% Difficulty in going outside the home alone to shop or visit a doctor's office 24% Difficulty in performing job duties due to disability 29% Providing an overview of the demographic characteristics of the overall population of Orange County led to identifying the similarities and differences between demographic characteristics of the survey respondents and those of the population at-large of Orange County. 84

101 It is clear there are demographic consistencies as these variables are defined within the survey responses and then compared with those calculated for the overall Orange County population, e.g., gender and race. However, there are notable variations in others of the demographic variables, such as age, education, income and, presence of disability, indicated in the answers of the survey respondents and those of the overall population of Orange County. These variations might be related to the sample size that was used in this research, which was limited to 235 samples, due to the available financial resources used in data collection. Another reason that might have led to the inconsistency in the demographic variables between the survey results and overall-county demographic characteristics is related to the fact that the survey samples were collected within specific geographic areas (evacuation and shadow evacuation zones that were delineated in Chapter 4). Limiting the sample to a specific geographic location might not capture the variations of demographic characteristics in the same degree of detail that was reported for the overall population of Orange County. Providing the general demographic characteristics of the overall population of Orange County allows for an assessment of the survey sample. The results of the telephone survey conducted in Orange County, California offered significant insights into the resident population s knowledge of tsunamis. Most of the sampled population knew about the major causes of tsunamis in both of the evacuation and shadow evacuation zone. However, slightly more than half of the surveyed population did not know about the last tsunami to hit coastal California. In terms of tsunami perceptions and beliefs, slightly more than half of the population in the evacuation zone thought that they were living in what would become the inundation zone of a possible tsunami and only around five out of fifty respondents in the shadow evacuation zone thought they were living in a tsunami inundation hazard zone. Also, in addition to being able to identify the sources of local and distant tsunamis, the surveyed households of Orange County were able to recognize the signs of an impending tsunami an earthquake and a rapid rise and fall of coastal water. Only a small proportion of the surveyed respondents were able to predict the estimated arrival times of local and distant tsunami waves in a manner close to reality in both of the evacuation and shadow evacuation zone. Not all of the sampled population would evacuate if they felt an earthquake, but reported that if they were to evacuate, it would be in response to an official order. When asked about the information sources they turn to for warnings of tsunami events, 85

102 most of the Orange County sampled population indicated they rely on local television channels for such alerts, as opposed to other sources such as the Internet, social media accounts, e.g., Facebook and Twitter, and NOAA radio. Further responses indicate that for each of the three tsunami scenarios (10-, 20-, and 30- foot waves), more than half of the population would plan to evacuate, and, evacuation participation rates would increase as the heights of the tsunami waves increase. Most of the respondents would evacuate their homes immediately if an official announcement were made that suggested an immediate evacuation for a tsunami affecting the Orange County coast, regardless of the tsunami scenario, and the majority of those respondents selected the house of a friend or relative as a shelter. With regard to selecting the evacuation routes, most of the surveyed respondents would choose their route with the following criteria: the distance between the route and the surveyed household, the location of the road relative to the ocean, the respondent s knowledge of the route, traffic conditions, and whether the road was the only option to get out of the area at risk. Regarding the demographic characteristics of the survey respondents, the age of more than half of the sampled population in this study ranged from 45 to 79. Racially, most are classified as white, non-hispanic. The majority of the respondents are married and female. While the majority of the respondents held a bachelor degree, on the whole, educational levels ranged from less-than-high school to a doctorate degree. Annual incomes of residents who made up the population sample ranged from $50,000 to $99,999 for the year 2013, and a small portion of the sampled population had a disabled person in the home, with the majority of these classified as having a physical disability. In terms of the physical characteristics of the surveyed households, most of the surveyed households were located in the evacuation zone with a distance of less than 1,929 feet between their households and the coast. The majority of the surveyed households were located in elevations of less than 59 meters. 86

103 CAHPTER SIX EVACUATION PARTICIPATION RATE AND TRIP GENERATION Evacuating populations from areas at risk of a possible tsunami is a critical task since the time between the generation of a tsunami wave and its arrival to coastal areas is limited. There are several factors that affect the evacuation process. Among such factors are: the magnitude of the tsunami, its estimated time of arrival, the available road network for evacuation, and the given population s response to an evacuation order. In addition to having an accurate picture of road-network capability, an effective evacuation depends on having knowledge of how people at risk of a specific tsunami threat may react to an evacuation order. However, as earlier stated (Chapters 2 and 5) communications between social scientists and transportation modelers are somewhat problematic: social scientists are interested in studying how people may behave in the context of an evacuation; transportation modelers focus on the technical aspects of modeling the evacuation process. This chapter provides information on how these two discipline-based approaches might be integrated. Survey questionnaire data from this study provide insights relative to the respondents evacuation decisions for the three tsunami scenarios and to the demographic characteristics of each of the respondents. Also, there are physical variables such as distance, elevation, and geographic location associated with each of the respondents that would likely impact the tsunami evacuation decision. Identifying the significant demographic and/or physical factors that contribute to the evacuation decision for each of the tsunami scenarios is critical to create a prediction models to estimate the evacuation rate for the whole population in order to come up with a total number of evacuees for each of the tsunami scenarios. This can be done through utilizing logistic regression models in which demographic and physical independent variables can be tested against the evacuation decision, specified as evacuating or not evacuating. Predicting the possible number of evacuees will help in providing estimates about the evacuation demand, which is a representation of the extent to which people may actually use the roads for evacuation. This will assist the transportation modelers in modeling the evacuation scenario, using both the evacuation demand and the current road network. The presented method will ease 87

104 communication between those concerned with the behavioral output of a population regarding specific threat and those involved with the needs of transportation modeling in the field of evacuation research. Chapter 6 will be concerned with the following questions: 1) How can behavioral responses be used to predict evacuation participation rates? 2) What are the demographic variables that might contribute to estimating evacuation rates in a tsunami event? 3) What are the physical variables that might contribute in estimating the tsunami evacuation rate? 4) What is the participation rate for each of the three tsunami scenarios? 6.1 Evacuation Participation Rate Evacuation demand generation has received little attention among researchers compared to that received by the traffic assignment models. This has been due to the complex nature of evacuation behaviors and the difficulty of modeling the decision-making process of evacuees (Yazici and Ozbay, 2008). However, with the availability of the appropriate tool to assist in predicting the evacuation decision for an entire population based on individual evacuee decisions within a specific surveyed population, estimating evacuation participation rates has become possible. The logistic regression model represents one of the more effective prediction tools in estimating evacuation participation rates, especially when the dependent variable is discrete (evacuate or not evacuate). There have been several studies that have utilized logistic regression to capture the significance of the socioeconomic and demographic characteristics in trip generation models, (Trainor et al, 2013; Yazici and Ozbay, 2008; Wilmot and Mei, 2004; Mei, 2002). Highlighting the potentials of the logistic regression model in predicting the evacuation participation rate is critical to evacuation research Logistic Regression Models Logistic regression models have a concept similar to that of the regular linear regression models. However, in the former there are: a) a dependent variable that has a categorical dichotomy, and, b) independent variables that are either continuous or are categorical variables 88

105 (Field, 2005). This means that the regular linear regression model has a continuous dependent variable that is measured on an interval or ratio scale (Azen and Walker, 2011). The regular linear regression model can be applied directly to a situation in which the outcome variable is dichotomous; however, the linear model should not be used for such situation due to the fact that the linear regression model assumes the relationship between variables to be linear (Field, 2005). There are several reasons for the selection of the logistic regression method in identifying the variables that contribute to the evacuation decision. First, logistic regression has the capability of estimating the probability of the occurrence of an event. Logistic regression offers the ability to analyze non-linear relationships through testing the relationship between a set of conditions and the event-occurrence probability (Sweet and Grace-Martin, 1998). Using the logistic regression model will assist in classifying an individual decision into one of two categories: dependent variables given specific information, and independent variables or predictors related to this individual respondent. Furthermore, Logistic regression can be used to model more than two response categories (Afifi and Clark, 1984; Afifi, May, and Clark, 2012). For this study, the researcher aims to create three models that predict the tsunami evacuation participation rate using the respondents evacuation decisions of the three tsunami scenarios (10-, 20-, 30-foot) and the demographic and physical characteristics of the respondents derived from the survey. Since the focus of this research is to determine if the targeted population will belong to the evacuate category or to the not evacuate category, the logistic regression model was selected to achieve this goal. Binary discrete phenomena, such as tsunami evacuation decision take the form of a dichotomous indicator or dummy variable. The values of 1 and 0 are usually used to represent the values of the dependent variable in the logistic regression. This is because the mean of the dummy variable equals the proportion of cases with value of 1 and can be interpreted as a probability (Pampel, 2000). The logistic regression model computes the log odds that a particular outcome will occur. The basic idea behind the logistic regression is that it begins by calculating the odds of the event using equation [6.1], where (A) represents the constant of the model, (B1, B2) represent the coefficient included in the model, and (X1, X2) represent the explanatory variables that is used for the prediction. In order to calculate the occurrence probability of an event, the odds should be converted to log-odds using equation [6.2]. In other words, the odds of an event occurring are given by the ratio of the 89

106 probability of it occurring to the probability of it not occurring. The log-odds of an event will range between infinity and +infinity with a high value representing an increased probability of occurrence. A positive value indicates that the event is more likely to occur than not (odds are in favor) while a negative value indicates that the event is more likely not to occur (odds are against) (Barce, Kemp and Snelgar, 2009). Predicting the log-odds of an event using the logistic regression model will produce a value range of between 0 and 1, which represents the log-odd of an event. Odds = Exp (A+B1(X1)+B2(X2) ) [6.1] Probability = Exp (A+B1(X1)+B2(X2) ) /1+ Exp (A+B1(X1)+ B2(X2) ) [6.2] The logistic regression model provides an important tool in identifying the significant independent variables that contribute to the dependent dichotomous variable. Using the SPSS (Statistical Package for the Social Sciences) software to create a logistic regression might produce a model that includes both the significant and insignificant variables even if the whole model is statistically significant. In order to create a logistic model that only includes the statistically significant independent variables and mainly contributes to the dependent variables, the stepwise procedure is suggested, especially in predictive research that focuses on defining a model not concerning causality and theoretical development (Menard, 2002). There are two types of stepwise procedures: forward and backward. The forward method starts with no independent variables in the model and then enters variables one at a time, at each time adding a variable with a statistical score p value that is less than 0.05, until the model comes to a point where no significant independent variables can be added. The backward method starts with the full model including all the independent variables in the model and then starts to remove the independent variables one by one until ending up with the most significant variables that contribute to the significance of the whole model. Field (2005) stated that the backward stepwise method is preferable to the forward method due to the support effect that occurs when a predictor has a significant effect, but only when another variable is held constant. Applying the backward stepwise method to the logistic regression model will assist in identifying the significant independent variables that contribute to the significance of the prediction model. 90

107 6.2 Data As described above, the two main sections of the survey (tsunami scenarios, and demographic information) are the major data sources that will be used to predict the evacuation participation rate of Orange County s population using the logistic regression model. This is because these two sections include information that relates to the respondents evacuation decisions, in addition to the demographic characteristics of the respondents, in the various tsunami evacuation scenarios. Research has shown that in reality, people tend to behave in ways similar to those they describe when responding to survey questions (Whitehead et al, 2000). This indicates that using the survey data to predict how people might behave in specific evacuation circumstances is a reliable measure. The answers to the questions 23, 38, and 52, relating to the respondent s evacuation decisions for the three tsunami scenarios (10-, 20-, and 30-foot waves, in the tsunami scenarios section of the survey) were used as dependent variables in the logistic regression model. In the survey, the respondents were asked to choose from four options regarding their intention to evacuate: Yes, No, Depends, and Do not know. All options as selected by each respondent were combined into two groups YES and NO. (The option NO made up of the remaining three options in order to have the dependent variable as a dichotomous variable. The demographic information (age, gender, income, education, race, disabled persons in the household, children, and persons over 65 in the house hold) in addition to the physical characteristics of each household such as location in the evacuation or shadow evacuation zone, distance of the household to the coast, and the elevation of the household are used as independent variables in the logistic regression models for the three tsunami scenarios. Data preparation is a necessary step prior to developing a logistic regression model using the SPSS since this tool requires the dependent and the independent variables to be coded in a way that can be read and analyzed by the software. Since the evacuation decision for each of the three tsunami scenarios (the dependent variable) was either evacuate or not evacuate, they were coded as (1,0) where 1= evacuate and 0 = not evacuate. The (1,0) codes were also used for the demographic variables (the independent variables) that are dichotomous such as (gender, race, household occupants over 65, and number of children). The race variable initially was a category; however, since the majority of the respondents recorded white as their racial group, 91

108 Table 6.1 Demographic variables codes in SPSS for logistic regression Variables SPSS Code Variables SPSS Code Evacuation Decision Age Evacuate Not evacuate Education Less than high school and high school graduate 1 80 or more 4 Some college and Associate degree and Professional degree 2 Didn't answer 5 Bachelor's degree 3 Gender Master's and Doctorate degree 4 Male 1 Income Female 0 Less than $29,999 1 Over 65 in Household $29,999 to $49,999 2 Over 65 1 $50,000 to $99,999 3 Not over 65 0 $100,000 to $199,999 4 Kids at Household $200,000 or more 5 Kids 1 Race No kids 0 White 1 Disable at the Household Other 0 Disable Person 1 No disable Persons 0 Table 6.2 Physical variables codes in SPSS for logistic regression Variable SPSS Code Variable SPSS Code Zone Distance Evacuation zone 1 Less than Shadow Evacuation Zone Elevation Less than and more and more 5 92

109 all other racial groups were merged into one group called others, making it a dichotomous variable. For the independent variables that are categorical such as (age, education and income), there are specific codes assigned to these categories, depending on the number of categories for each variable (see Table 6.1 for details about the codes of the variables). The age category consisted originally of four groups as shown in the survey (see Appendix). However, to avoid any missing values in the age variable, the decision was made to add another category, Did not answer, since some respondents chose not to provide information about their age. The physical variables were treated the same way where the location of the household (the evacuation or shadow evacuation zone) was coded as (1= evacuation zone and 0 = shadow evacuation zone) (see Table 6.2 for more details). The data concerning the location of each respondent (the evacuation or shadow evacuation zone) were provided by the Kerr and Down Company in coordinate points (x,y) format as part of the data collection process. These points of each of the respondents were also used to get the distance of each respondent s household to the coast using the measure tool in ArcMap software. The elevation of the respondent s household as another physical variable was extracted from USGS DEM using the extract value to point tool under spatial analyst tools in ArcMap software. The distance and elevation of the household were treated as continuous independent variables. Since there are some demographic and physical variables that are categorical, these variables were specified in the SPSS as categorical variables where the first group of each category was specified as the reference group. 6.3 Logistic Regression Models for Each of the Three Tsunami Scenarios (10-, 20-, 30-foot) Enter Method The intention for evacuation was modeled against the demographic and physical variables that were detailed in the previous section for each of the tsunami scenarios separately. The enter method was selected to insure that all variables were entered into the model at the same time. The output of the model using the enter method will result in a model that includes the entire number of variables even if they are not statistically significant. In the 10-foot tsunami scenario, the evacuation intentions of all the respondents (156 decided to evacuate and 79 decided not to evacuate) were modeled against the demographic and physical variables of each 93

110 respondent. Table 6.3 represents the output of the logistic regression model using the enter method for a 10-foot tsunami scenario. It is necessary to define the meaning of each of the statistical measures that are reported in Tables in order to interpret the outputs of the models. (B) stands for the coefficient of the independent variable, (S.E) represents the stander errors associated with the coefficients, (Wald) provides the Wild chi-square value, which tests the null hypothesis that the estimate equal 0, (df) lists the degrees of freedom for each of the variables that are included in the model, (Sig) represents the p-values of the coefficients and Exp(B) represents the odd ratios for the predictors. The overall significance of the 10-ft model using the enter method is Note that not all of the independent variables are statistically significant in Table 6.3 since the significance (Sig) value in some of the independent variables is more than The gender and zone are the significant independent variables that contribute to the significance of the whole model for the 10-foot tsunami scenario, using the enter method. This indicates that there are some independent variables not significantly contributing to the model even though these variables are included in the model, and the general model is statistically significant. The intention for evacuation in the 20-foot tsunami scenario for the entire sampled population (185 would evacuate, 50 would not) was modeled against the demographic and physical variables. The overall significance of the model, using the enter method, is indicating that the model is statistically significant in predicting the intention for evacuation in a 20-foot wave scenario. Table 6.4 summarizes the output of the logistic regression model for this 20-foot scenario using the enter method. The use of the this method, which enters all the independent variables into the model, resulted in only the gender and zone being statistically significant in the model even though the overall model is statistically significant. For the 30- foot tsunami scenario, the intention of evacuation (195 would evacuate and 40 determined they would not) was modeled against the demographic and physical variables of each of the respondents. The overall significance of the 30-foot tsunami evacuation logistic regression model, with the use of the enter method, is The significance of each of the dependent variables of the logistic regression model for the 30-foot tsunami scenario is summarized in table 6.5. Note that using the enter method with logistic regression for the 30-foot tsunami scenario indicated that the race, gender and zone the most statistically significant variables in the model. 94

111 Table 6.3 The output of the logistic regression model using the enter method for 10-ft tsunami Age (18-44) [reference group] B S.E. Wald df Sig. Exp(B) Age( 45-64) Age(65-79) Age(80 or more) Age(Didn t answer) Race Education (less than high school and high school graduate) [reference group] Education( some college, associate degree and professional degree) Education ( Bachelor s degree) Education ( Master s and Doctorate degree) Income (Less than $29,999) [reference group] Income( $29,999 - $49,999) Income( $50,000 - $99,999) Income( $ 100,000 - $199,999) Income( $200,000 or more) Disable Gender Zone Elevation (Less than 59 m) [reference group] Elevation( m) Elevation( m) Elevation(240 m and more )

112 Table 6.3 continued B S.E. Wald df Sig. Exp(B) Distance (Less than 1929) [reference group] Distance ( ft) Distance ( ft) Distance ( ) Distance ( 7720 and more) Over Kids Constant Table 6.4 The output of the logistic regression model using the enter method for 20-ft tsunami B S.E. Wald df Sig. Exp(B) Age (18-44) [reference group] Age( 45-64) Age(65-79) Age(80 or more) Age(Didn t answer) Race Education (less than high school and high school graduate) [reference group] Education( some college, associate degree and professional degree) Education ( Bachelor s degree) Education ( Master s and Doctorate degree) Income (Less than $29,999) [reference group] Income( $29,999 - $49,999) Income( $50,000 - $99,999) Income( $ 100,000 - $199,999) Income( $200,000 or more)

113 Table 6.4 continued B S.E. Wald df Sig. Exp(B) Disable Gender Zone Elevation (Less than 59 m) [reference group] Elevation( m) Elevation( m) Elevation(240 m and more ) Distance (Less than 1929) [reference group] Distance ( ft) Distance ( ft) Distance ( ) Distance ( 7720 and more) Over Kids Constant Table 6.5 The output of the logistic regression model using the enter method for 30-ft tsunami B S.E. Wald df Sig. Exp(B) Age (18-44) [reference group] Age( 45-64) Age(65-79) Age(80 or more) Age(Didn t answer) Race Education (less than high school and high school graduate) [reference group] Education( some college, associate degree and professional degree) Education ( Bachelor s degree) Education ( Master s and Doctorate degree)

114 Table 6.5 continued B S.E. Wald df Sig. Exp(B) Income (Less than $29,999) [ reference group] Income ( $29,999 - $49,999 ) Income ( $50,000 - $99,999) Income ( $100,000 - $199,999) Income( $ 200,000 or more) Disable Gender Zone Elevation ( Less than 59 m) [reference group] Elevation ( m) Elevation ( m) Elevation ( 240 m and more ) Distance (less than 1929 ft) [reference group] Distance ( ft) Distance ( ft) Distance ( ft) Distance (7720 and more) Over Kids Constant It is clear that using the enter method for the three tsunami scenarios produced statistically significant models; however, the significance of the independent variables is not consistent from one scenario to another even though all of the independent variables were included in the three models. Including only the significant independent variables in the logistic regression model might improve the significance of the whole model, which in turn would increase the reliability of the predictive of the whole model. Thus, it was necessary to select a method that only includes the significant variables in the logistic regression model, to improve 98

115 the overall significance of the model and identifies the independent variables that contribute the most to the whole model. The backward stepwise method is considered one of the best methods for identifying the most statistically significant variables in the model. It works by including all of the independent variables in the logistic regression model and then removes one variable at a time, based on its significance, until ending up with the most significant variables in the model. Including the significant independent variables in the model might improve the significance of the whole model, which in turn will be reflected in the precision of its predictions Backward Stepwise Method The same dependent and independent variables for each of the tsunami scenarios that were used in the enter method were applied also in the backward stepwise method. The general data coding for the logistic regression using SPSS software was the same in each case. The overall significance of the 10-foot tsunami evacuation model using the backward stepwise method is It is clear that the significance of the whole model improved using the backward stepwise method when it was compared with the significance level of the enter method. Generally, the significance of the whole model changes due to the inclusion or exclusion of the independent variables in the model. Including specific independent variables in the model may change the significance of these independent variables, where different independent variable combinations may produce different significance levels depending on the variables included in the model, which in turn will be reflected in the significance of the whole model. Table 6.6 summarizes the main outputs of the logistic regression model for the 10-foot tsunami scenario using the backward stepwise method. The location of each of the respondents (whether in the evacuation or shadow evacuation zone) is the only significant variable contributing to the evacuation decision for a 10-foot tsunami scenarios, based on the result of the backward stepwise method. The inclusion of this variable indicates its contribution to the significance of the whole model. Thus, the (B) value of the independent variable reported in Table 6.6 was used in the final prediction model for a 10-foot wave scenario, using equation 6.3. The (B) value measures the relationship between the independent variables and the dependent variable where the dependent variable is measured in a logit scale. The gender and zone are the 99

116 main contributing independent variables in the evacuation decision for the 20-foot tsunami scenario (See Table 6.7 for greater detail). Using these specific independent variables that resulted from the backward stepwise method for the 20-foot tsunami scenario will improve the significance of the whole model to ( ) when it is compared to the significance of the 20- foot tsunami model using the enter model ( ). The (B) value of these independent variables as noted in Table 6.7 was used for the 20-foot tsunami final prediction model using equation 6.4. Table 6.8 summarizes the main outputs of the logistic regression model for the 30-foot tsunami scenario using the backward stepwise method. Gender, zone are the main contributors to the significance of the 30-foot tsunami evacuation scenario logistic regression mode using the backward stepwise method. When compared to the overall significance of the same scenario, using the enter model ( ), the significance of the 30-foot scenario improved by (8.1205E-8) using only the independent variables reported in Table 6.8. This indicates that the backward stepwise method identifies the variables that make a statistically significant contribution to the model. Table 6.6 The output of the logistic regression model using the Backward Stepwise LR method for a 10-foot tsunami scenario Variables (B) Sig Exp (B) Zone Constant Table 6.7 The output of the logistic regression model using the Backward Stepwise LR method for a 20-foot tsunami scenario Variables (B) Sig Exp (B) Gender Zone Constant

117 Table 6.8 The output of the logistic regression model using the Backward Stepwise LR method for a 30-foot tsunami scenario Variables (B) Sig Exp (B) Gender Zone Constant Equation 6.3: Prediction equation for 10-foot evacuation = Exp ( (Zone)) /1+ Exp ( (Zone)). Equation 6.4: Prediction equation for 20-foot evacuation = Exp ( (Zone) (Gender)) /1+ Exp ( (Zone) (Gender)). Equation 6.5: Prediction equation for 30-foot evacuation = Exp ( (Zone) (Gender)) /1+ Exp ( (Zone) (Gender)). The above noted models are the most statistically significant for each tsunami scenario where various independent variables may contribute to the overall significance of the model. It is necessary to report the overall correct percentage of the predictions for each model -- in addition to Cox & Snell R Square and Nagelkerke R Square, which are statistical measures equivalent to R Square in multiple regressions for each model -- in order to evaluate their predictive strength (Brace et al, 2009). Both the Cox & Snell R Square and Nagelkerke R Square show the estimates of the amount of variance in the dependent variable that can be explained by the independent variables. The 10-foot tsunami model accounted for between 4.7% and 6.5 of the variability (See Tables 6.9), with 34.2% of the non-evacuating population and 85.3% for the evacuating population successfully predicted. The overall predictions of the 10-foot model improved to 68.1% when the variable zone was included the in the model comparing with the overall 101

118 prediction of having only the constant in the model with overall predictions of 66.4% (See Tables 6.10 and 6.11 for more details). Table 6.9 Equivalent R Square statistical measures for 10-ft tsunami scenario using Backward Stepwise method Cox & Snell R Nagelkerke R Step -2 Log likelihood Square Square Table 6.10 Percentage of correct predictions using only the constant in the 10-ft tsunami model Observed Predicted 10-ft Percentage Not Evacuate Evacuate Correct 10-ft Not Evacuate ft Evacuate Overall Percentage 66.4 Table 6.11 Percentage of correct predictions after including the variable zone in the 10-ft tsunami model Observed Predicted 10-ft Percentage Not Evacuate Evacuate Correct 10-ft Not Evacuate ft Evacuate Overall Percentage 68.1 The 20-foot tsunami model accounted for between 10.8% and 16.7% of the variability (See Table 6.12), with 28% of the non-evacuating population and 94.6% of the evacuating population successfully predicted. The overall predictions of the 20-foot model improved to 80.4% when the variables zone and gender were included the in the model comparing with the 102

119 overall prediction of having only the constant in the model with overall predictions of 78.7% (See Tables 6.13 and 6.14). For the 30-foot tsunami scenario, the model accounted for between 13% and 21.7% of the variability (See Table 6.15), with 35% of the non-evacuating population and 94.9% of the evacuating population successfully predicted. The overall predictions of the 30- foot model improved to 84.7% when the variables zone and gender were included the in the model comparing with the overall prediction of having only the constant in the model with overall predictions of 83% (See Tables 6.16 and 6.17). Reporting these statistics about the models would provide a general overview of the model predictive power, which might be reflected in the predicted results. Table 6.12 Equivalent R Square statistical measures for 20-ft tsunami scenario using Backward Stepwise method -2 Log Cox & Snell R Nagelkerke R Step likelihood Square Square Table 6.13 Percentage of correct predictions using only the constant in the 20-ft tsunami model Observed Predicted 20-ft Percentage Not Evacuate Evacuate Correct 20-ft Not Evacuate ft Evacuate Overall Percentage 78.7 Table 6.14 Percentage of correct predictions after including the variables zone and gender in the 20-ft tsunami model Observed Predicted 20-ft Percentage Not Evacuate Evacuate Correct 20-ft Not Evacuate ft Evacuate Overall Percentage

120 Table 6.15 Equivalent R Square statistical measures for 30-ft tsunami scenario using Backward Stepwise method -2 Log Cox & Snell R Nagelkerke R Step likelihood Square Square Table 6.16 Percentage of correct predictions using only the constant in the 30-ft tsunami model Observed Predicted 30-ft Percentage Not Evacuate Evacuate Correct 30-ft Not Evacuate ft Evacuate Overall Percentage 83.0 Table 6.17 Percentage of correct predictions after including the variables zone and gender in the 30-ft tsunami model Observed Predicted 30-ft Percentage Not Evacuate Evacuate Correct 30-ft Not Evacuate ft Evacuate Overall Percentage 84.7 Most of the demographic variables that were analyzed using logistic regression in this study were not statistically significant when compared with the zone variable (evacuation or shadow evacuation zone) which has a significant predictive power in calculating the evacuation rate. This might be related to the impact of societal factors such as social control, social cohesion, and social capital) in the evacuation decision. Ricchetti-Masterson and Horney (2013) conducted a study to test the impact of social factors on the relationship between demographic variables and the evacuation decision after hurricane Irene. The results indicated that there was no direct association between the demographic variables and the evacuation decision; however, some degree of association was found to exist between social capital or social cohesion and demographic variables that could impact the evacuation decision. Baker (1991) stated that 104

121 demographic variables are rarely, if ever, associated with the evacuation decision. Although there have been several studies that have reported an association between certain demographic variables and the evacuation decision (Riad and Norris, 1999; Bateman and Edwards, 2002), the findings of these studies have been inconsistent, indicating that a demographic variable might be significant for evacuation in one study, but not in another. For this dissertation, only gender was found to have a slight association with the evacuation decision for the 20- and 30-foot tsunami scenarios where females are somewhat more likely to evacuate than males. On the other hand, zone (evacuation or shadow evacuation) was shown to have a strong association with the evacuation decision due to the location of the respondents, i.e., most of the survey sample was collected within the evacuation zone. It is critical to highlight that the logistic regression model tries to make predictions based on the survey responses; however, in reality, people might not make the same decision they hypothesized making in the survey. Thus, varying evacuation trends may result. The evacuation prediction rates reported in this dissertation are based on the evacuation intentions stated by the survey respondents. The methodology of using the models that were built, using the survey data to estimate the evacuation participation rate, based on census block group data, will be discussed in the following section. 6.4 Predicting Number of Evacuees Predicting the evacuation participation rate from the evacuation and shadow evacuation zones for the three hypothetical tsunami scenarios is a critical step in estimating the number of evacuees for the purpose of trip generation and evacuation modeling. The prediction equations 6.3, 6.4, 6.5, produced using the backward stepwise method of logistic regression model, were used to predict the number of evacuees for each of the 10-, 20-, and 30-foot tsunami scenarios. These equations were used to predict the participation rate for each of the census block groups in both the evacuation and shadow evacuation zones that were specified in Chapter 4. The location (within the evacuation or shadow evacuation zone) was derived using the centroid of each census block group polygon in ArcMap software. This is done through overlaying the centroid of the polygons on the evacuation and shadow-evacuation-zone layers that were addressed in Chapter 4. Additionally, gender data for Orange County were downloaded from the US Census website for year 2010, at the census block group level. The gender data were spatially joined with the 105

122 census block group shapefile using geoid to represent these data spatially on the map. The evacuation and shadow evacuation zones that were delineated in Chapter 4 were used to identify which census block group belonged to which zone using the centroid of each polygon. Thus, if the centroid of the polygon were located in the evacuation zone, this polygon would be classified as within the evacuation zone, and if the centroid of the polygon were located in the shadow evacuation zone, this polygon would be classified as within the shadow evacuation zone. Since the prediction models for 10-, 20-, and 30-foot tsunami scenarios were built based on using individual level data (survey respondents), and the data that would be used to make predictions of tsunami evacuation for each scenario is on an aggregated level (census block groups), it was necessary to find a way to overcome different data level issues. This was done through using the average probability that is based on expectation or the expected value as it is known in statistics (Cohen, 1988). The expected value basically takes the summation of possible values from a random variable, representing the average probability of an event occurring (Clarke and Cook, 1978). In the current study, gender takes the form of (1) for male or (0) for female. Let the proportion of male be p1, and the proportion of female be p0. For male, the predicted evacuation rate is p_evaculate1, for female, it is p_evaculate0. Based on the expected value, the expected evacuation rate is p1*p_evaculate1 + p0*p_evaculate0. The example below explains the method in more detail. For instance, there is a block group with male and female populations of 4,303 and 4,388 respectively (a total of both equaling 8,691). This block group is located in the shadow evacuation zone, and there is a need to estimate the total number of evacuees in case of a 20-foot tsunami scenario. The total number of evacuees for this scenario can be calculated first through getting the predicted evacuation rate for both male and female populations, using the 20-foot tsunami scenario prediction equation that was referenced above as the following: Predicted evacuation rate for Male = EXP ( (1)+1.758(0)/(1+EXP( (1)+1.758(0)) =

123 Predicted evacuation rate for Female = EXP ( (0)+1.758(0)/(1+EXP( (0)+1.758(0))= The predicted evacuation rate for both male and female populations then should be multiplied by the proportion of males and females within the same block group to get the observed probability, where in our example the male and female proportions are equal, i.e., (50%). Observed probability for Male= *0.50= Observed probability for Female= *0.50= The output of adding the observed probability for male and the observed probability for female will result in the expected evacuation rate for both male and female populations in the block group as follows: Expected evacuation rate for male and female= = This expected evacuation rate of male and female represents the overall evacuation probability of the population in the block group. In order to get the total number of evacuees from the same block group, the calculated expected evacuation rate should be multiplied by the total number of residents of the same block group as follows: Total number of evacuees= *8,691= 4,493 evacuees The same approach was applied for each block group located within the evacuation and shadow evacuation zones for all three of the tsunami scenarios (10-, 20-, and 30-foot). The overall evacuation participation rate for each of these scenarios, using the prediction equations produced from the logistic regression, is reported below in Table

124 Table 6.18 Evacuation participation rate using the prediction models Tsunami Scenario Evacuation Rate (n=1,119,940) 10 ft 48% 20 ft 54% 30 ft 58% Table 6.19 Tsunami evacuation participation rate based on the survey results Tsunami Scenario Evacuation Rate (n=235) 10 ft 66% 20 ft 79% 30 ft 83% It is clear that as the size of each tsunami scenario increases, the evacuation participation rate increases as well. However, when comparing the evacuation rate produced from the prediction models to the evacuation rate reported from the survey as stated in Tables 6.18 and 6.19, a sizable gap exists between the survey evacuation rate and that of the prediction models. The reason for the big gap between the survey evacuation rates and the prediction models evacuation rate was due to deliberately oversampling survey samples in the evacuation zone, which prevented the comparison of the simple average of the combined sample (evacuation and shadow evacuation zone) from survey to a simple average of the predicted value for the combined (evacuation and shadow zones) based on census block groups. Collecting more samples in the evacuation zone while conducting the survey was based on the assumption that more evacuees would belong to this zone as compared to those belonging to the shadow evacuation zone. The conflict involved in selecting the appropriate sample size from each of these zones (evacuation and shadow evacuation), in a way that would match the real situation, resulted in having the predictions of the model lower than those indicated by the survey results. Had the majority of survey samples been collected from the shadow evacuation zone, this would have better insured that the predicted evacuation participation rate from the model using census block group would have more closely matched the evacuation participation rate reported from the survey. 108

125 Despite the inconsistency between the survey evacuation rate and the predicted evacuation rate, the evacuation prediction produced from the model indicates the same trend in the evacuation participation rate reported from the survey: i.e., as the magnitude of the tsunami increases in a scenario, the evacuation participation rate increases as well in both the survey and model results. The output of the prediction models is used in the following section for the purpose of trip generation. 6.5 Estimating Number of Evacuating Vehicles (Trip Generation) The total number of evacuees as well as the total number of vehicles varies from one census block group to another due to variations in the residents demographic and socioeconomic characteristics. These variations will be reflected in the total number of vehicles expected to be used in any evacuation event. The predicted number of evacuees for each of the tsunami scenarios, calculated using the logistic regression model, represents the total number of evacuees, including children, adults, and the elderly, for each block group. In order to calculate the total number of evacuating vehicles from each block group, it was necessary to identify the total number of adult evacuees within the same block group since it was supposed that only adults would drive a vehicle. This was done through dividing the total number of adult residents (age range between years old) by the total number of residents of the same block group. The output of this division represents the percentage of adult residents out of the total residents in the same block group. This percentage was then multiplied by the total number of evacuees of the same block group; this step led to the identification of the total number of adult evacuees for that block group. In order to calculate the total number of evacuating vehicles, vehicle occupancy had to be taken into consideration. Since there was no existing data for vehicle occupancy that could be used for such calculation, four different scenarios were proposed to calculate the number of evacuating vehicles, based on vehicle occupancy for each of the tsunami scenarios. In the first vehicle occupancy scenario (the worst-case scenario) the researcher assumes that every adult evacuee will evacuate using a vehicle. Therefore, the total number of adult evacuees was divided by one, resulting in the total number of evacuating vehicles. For the second scenario, two adults were expected to use one vehicle for evacuating; therefore, the total 109

126 number of evacuating adults was divided by two to calculate the total evacuating vehicles while accounting for vehicle occupancy. For the third scenario, three adults were expected to ride in one vehicle; in this case the total evacuating vehicles was calculated by dividing the total number of adult evacuees by three. In the fourth scenario, four adults were expected to ride in one vehicle; the total number of adult evacuees was then divided by four to calculate the number of evacuating vehicles. The total number of vehicles that were calculated, based on vehicle occupancy and the proportion of the adult population who planned to evacuate, would be used as an input for evacuation modeling, in order to measure the clearance time required for each of the tsunami scenarios (Chapter7). Identifying the total number of evacuees and vehicles expected to be used in the evacuation process based on the logistic regression prediction models that were built according to the significant physical and demographic variables derived from the survey results of the sampled population -- reflects the possibility of communicating the behavioral output into data that can be used by transportation modelers to model the evacuation process. 110

127 CHAPTER SEVEN EVACUATION MODELING AND CLEARANCE TIME A key factor in creating models or simulations of an evacuation process is the variable of human behavior. As mentioned earlier in Chapter 5, a survey is considered the main data source of behavioral analysis. Behavioral assumptions that are derived from individual surveys describe the way people respond to an evacuation order. Specifically, the results of the behavioral analysis in this research will indicate the population s decision for evacuation and how the evacuees will choose their evacuation destinations, routes, and departure times. Integrating the spatial distribution of the vulnerable population with their behavioral responses in the case of an evacuation order, i.e., the population s probable actions and decisions in such an event, will help in identifying the possible number of evacuees, evacuation routes and destinations; this information then can be used to model the evacuation process (Cova and Church, 1997; Lambert, 2013). Routes for evacuation purposes vary in their capacities, to accommodate a specific number of vehicles (Hobeika and Kim, 1998). Such a variable indicates the importance of the role human factors play in the evacuation process, along with the characteristics of the designated road network. There have been a number of evacuation behavioral studies developed side-by-side with evacuation transportation modeling in efforts to understand the evacuation process and plan for it (Lewis, 1985; ORNL, 1995; PBS&J, 2000; Franzese and Han, 2001; Urbina, 2002; Lindell and Prater, 2007; Yazici and Ozbay, 2008; Trainor et al, 2013). However, a gap exists in evacuation research between behavioral and transportation research due to the different focuses of these two lines of research. This chapter will utilize the geospatial tool CASPER (Capacity-Aware Shortest Path Evacuation Routing) to model the output of the behavioral evacuation decisions, using an existing road network for Orange County, California. This tool will assist in modeling the findings from behavioral analysis, which in turn will improve the estimation of the evacuation clearance time. This method may enhance evacuation planning and help in developing new policies to protect coastal communities. 111

128 This chapter will answer the following questions:1) How might behavioral data be used to simulate the evacuation process? 2) What are the issues that should be taken into consideration while conducting the transportation analysis? 3) What are the advantages and the limitations of using the ArcCASPER tool to model evacuation? 4) What is the clearance time for each tsunami evacuation scenario? 7.1 Overview of CASPER Tool CASPER (Capacity-Aware Shortest Path Evacuation Routing) is a network analyst tool which, through the use of an algorithm, develops routes for evacuating populations to the closest safe locations, during disaster events (Shahabi, 2012). The CASPER tool is classified as a prescriptive evacuation routing method. The main goal of the prescriptive method is to reduce the evacuation time, minimize traffic congestion, and minimize exposure to danger. There are several reasons behind selecting the CASPER tool to model evacuation for this research. First, CASPER is not combined with zonal scheduling, as in the case of the shortest-path routing to solve routing problems. Instead, it uses an algorithm to connect each source node to its nearest destination. This is done through intelligently integrating transportation network capacity (number of lanes for each road) and traffic flow (number of evacuating vehicles) to minimize congestion and reduce the evacuation time, in turn eliminating the need for scheduling evacuation time. Also, the CASPER method scales to larger areas with limited memory use, which allow the user to perform the routing on a larger network. It determines the optimal evacuation routing strategies to achieve evacuation goals without performing fine scale modeling. Furthermore, CASPER works with any traffic model as long as it meets certain conditions such as flow and capacity (Shahabi and Wilson, 2014). The CASPER method produces realistic routes that decrease the global transportation time while minimizing the traffic congestion, without any loss of performance. The tool combines the length of the road with the road capacity in order to predict speed under different traffic conditions. The tool gives the user the opportunity to select from five different traffic models: Flat, Step, Linear, Power, and Exponential; CASPER then utilizes road segment capacity and traffic flow as inputs into the selected traffic model. A new speed estimate will be 112

129 produced for each road segment based on the road saturation density per unit capacity; this estimate then affects the path finding process. The tool keeps updating the speed estimates and repetitively generates routes specific to each evacuee in order to minimize the global evacuation time (Shahabi and Wilson, 2014). 7.2 Assumptions The CASPER tool has been selected to be used in this dissertation research to simulate the different tsunami evacuation scenarios; therefore, it is critical to state the basic assumptions that were made in order to evaluate the performance of this tool. The first assumptions is related to the evacuation clearance time of evacuating vehicles based on vehicle occupancy of a given tsunami scenario. It is assumed that as the number of vehicles increases for a certain evacuation scenario (i.e., 10-foot), the total evacuation time should increase as well. Another assumption is related to the evacuation clearance time comparison between one tsunami evacuation scenario and another, where is it assumed that congested evacuation (numbers of evacuating vehicles increase) time produced by the CASPER tool should increase as the severity of the tsunami scenario will increase. Also, it is assumed that CASPER tool doesn t take into consideration the dynamics of the road network during an evacuation. Using the CASPER tool, it is assumed that every road in the network will be treated the same way in terms of calculating congested and uncongested evacuation time. Furthermore, it is expected that CASPER will produce similar evacuation routes for 10-, 20-, and 30-foot tsunami scenarios since the same road network is used for all scenarios using seven seconds instead of one second initial delay cost per evacuees. 7.3 The CASPER Construct The CASPER tool models the evacuation routing problem as a graph, path-finding problem with four inputs: the graph, the traffic model, the source flow, and the destination. Represented in the graph is the road network, with edges and vertices where each vertex has nonnegative impedance and capacity. Source points and destination points are necessary inputs in CASPER to model the evacuation: source points contain all the vertices of the evacuating population and the destination points (safe locations or shelters) contain the vertices for all of the 113

130 population in the source points. When source points generate evacuation routes to a destination, each source point will generate only one path to one of the destination points. This evacuation path, produced by the CASPER tool, can be described as an ordered set of edges that will direct evacuees to safety. Thus, the total flow on an edge represents the sum of all flows from all paths that pass through the same edge. The power traffic model is implemented in the CASPER tool to predict the edge congestion based on edge capacity (number of lanes) and total flow (number of evacuees) (see equation [7.1] for more details). The capacity definition in CASPER tool is totally different from the traditional capacity definition for roadway that depends on geometric characteristics of the roadway as well as the traffic demand. The cost of traversing an edge is calculated based on the traffic model specified and the saturation density per unit capacity, which represents the number of vehicles within the lane of the road segment. Assigning specific value to the saturation density of the traffic model will affect the road traversal speed. This is due to the fact that if the road is routing its specified quantity of evacuees, the traversal speed of the road will reduce to half of its original traversal speed, which will affect the traversing cost of the road. Calculating the cost of traversing an edge is followed by calculating the cost of traversing a path. The final output will include the delay time restricted by the traffic model (congested evacuation time) and the time of traversing the whole evacuation path (evacuation time without congestion). As already noted, the user of the CASPER tool has the option to choose from five different traffic models while simulating the evacuation process to select the best routes and minimize the global evacuation time. One of these traffic models is the Flat model, which is a constant function. This model ignores the congestion ratio and generates the shortest path for each evacuation source point to a destination, if a path exists. The second traffic model is the Step model, created based on the Capacity Constrained Route Planner (CCRP) algorithm, which stops using an edge as soon as it runs out of capacity. Each evacuee is assigned an evacuation route by the CCRP algorithm; in doing so, this algorithm is also able to insure that both intersections and roadway segments are not strained beyond their capacities (Shahabi and Wilson, 2014). The Linear model is another traffic model that allows evacuation routing problems to be reduced to a max-flow, min-cost problem which, in turn, is solved as polynomial time while assuming that evacuee movements can be modeled as flows. The Power 114

131 traffic model is the result of the empirical curve fitting and is considered to be fixable. The predicting traffic congestions and estimating the evacuation time can be enhanced using the power model. The Linear and Exponential models are classified as analytical traffic models of CASPER tool (Shahabi and Wilson, 2014). Among these traffic models, the Power model is considered to be the best since it never underestimates the evacuation time, and predictions of the total evacuation model are close to the simulation time. The Exponential traffic model predicts some of the congestion, but such predictions are lower than the simulation. This is due to the Exponential model not calculating full congestion on roads that have large width (number of lanes) when compared to the Power model. The Flat model is unconcerned with congestion; thus, it produces the same evacuation time for every scenario. The Step model behaves in a way similar to the Flat model, especially when the graph is not saturated (Shahabi and Wilson, 2014). For further details, with regard to these traffic models, the reader can refer to Shahabi and Wilson (2014). 7.4 Data in the CASPER Tool The CASPER tool, as an evacuation routing extension to ArcGIS Network Analyst, can serve to improve the research communication level between the behavioral scientists and evacuation modelers: this tool represents a promising solution for modeling an evacuation process that is based on the evacuation demand determined by a behavioral analysis. The data the CASPER tool uses to model the evacuation routing problem may come from different sources with various levels of detail. These data include: the geographic location of safe points, the geographic location of evacuee points, and the roadway network. The geographic location of safe points can be located anywhere outside the risk zone (evacuation and shadow evacuation zones in this research). For the purpose of this dissertation, the locations of safe points were selected outside the risk zone that was specified in Chapter 4. The evacuee points are basically the centroids of each census block group polygon that were used in Chapter 6 to calculate the number of evacuees. Each evacuee point has four different specifications of number of evacuees representing the number of evacuating vehicles, based on vehicle occupancy. Therefore, the locations of the centroid of each polygon within the risk zone represent the locations of the evacuees. 115

132 The roadway network that was used in this research was purchased from Korem Company, a company specialized in geospatial technologies. This roadway dataset was selected for use in this research since it has most of the attributes that are needed to create the road network and perform the evacuation modeling. These attributes include: type of roadway, speed limits, number of lanes, bridges, etc. All of these attributes and others are necessary to be included in the road network in order to allow the CASPER tool to simulate the evacuation process with real data that represent the real world. (For more detail about the roadway data see Chapter 3.) 7.5 CASPER Tool Processing Phases Phase 1 Phase one focuses on building up the road network from the roadway data. It is necessary to insure that all of the necessary attributes, such as number of lanes, speed limits, travel time, directions, etc., are included in the roadway dataset. While creating the roadway network in ArcCatalog, a new attribute named Capacity was created. In the original road dataset, there is an attribute named Lanes, which represents the number of lanes for each roadway segment. The values of the Lane attribute will be used to fill in the Capacity attribute. The Power traffic model within the CASPER tool depends on flow (number of evacuating vehicles) capacity (number of lanes for each road) to calculate the evacuation time with congestion. Therefore, it is necessary to create the capacity attribute while building the roadway network Phase 2 The main purpose of Phase Two of CASPER tool is to identify and load the locations of evacuees and safe points. The safe zone in this research would be any area that is outside the shadow evacuation zone, since the shadow evacuation zone is an extension of the evacuation zone as described in Chapter 4. A twenty equally spaced points are created as safe points outside the shadow evacuation zone. A new attribute Capacity is created in the safe points layer. In order to insure that all safe points can hold an unlimited number of evacuees, the new capacity 116

133 attribute in the safe point layer should have a value of -1 (Shahabi and Wilson, 2014). The evacuee points are located on the centroid of the census block groups that were used in Chapter 6 to calculate the number of evacuating vehicles. As detailed in Chapter 6, the number of adults from each block group within the evacuation and shadow evacuation zones was used as input into the logistic regression model to predict the number of evacuating vehicles based on vehicle occupancy. These numbers represent the evacuating vehicles with differing vehicle occupancies for each block group. Thus, for the purpose of the CASPER tool modeling, the centroids of these block group polygons are used as evacuee points Phase 3 The final phase of the CASPER tool is Phase Three where evacuation settings are managed. In the evacuation options, the user should specify the capacity attribute that was created while preparing the road network in Phase One. In the routing options, the user should select the minutes attribute as the cost of the network routing. As noted above, the CASPER tool provides the user with the option of using various traffic models. However, for the purpose of this dissertation, the researcher selected the Power model since it is fixable, and it improves the predictions of traffic congestion and the estimates of evacuation time (Shahabi and Wilson, 2014). The Power traffic model enables the user to control the traversal speed of the road through the Saturation Density per Unit Capacity, one of the traffic options of the evacuation setting in CASPER; thus, once the user has selected the Power model as the traffic model under the traffic options of the evacuation settings, the Saturation Density per Unit Capacity should be specified. There are different link performance functions that are used in the literature in addition to the Power model used for this research; however, since the CASPER tool specifically uses the Power model only, it will be used for the analysis in this dissertation If the Saturation Density per Unit Capacity indicates there is an increase in the demand on the road, through a routing of a greater-than-anticipated number of evacuees, the traversal speed of the same road will decrease to half of its original traversal speed. That is, if the number of evacuating vehicles was to exceed the cut-off indicated by the user through saturation density per unit capacity, and the traffic slows to this extent, there is the danger of congestion. The 117

134 CASPER tool keeps updating the traversal speed of the roads that make up the evacuation route for each evacuee, which in turn will affect the global evacuation time. Since the final evacuation time with and without congestion depends on the saturation density per unit capacity, the researcher decided to use a set of saturation densities 500, 250, 225, 200 and 100 to indicate evacuation time differences based on these densities. The results will help emergency managers and policy makers to develop evacuation plans based on the available resources. The final output of the CASPER tool is evacuation routes, polylines from each evacuee point to the selected safe point. These evacuation routes report the evacuation time with and without congestion. The evacuation time with congestion represents the evacuation time based on the reduced speed that is controlled by the saturation density per unit capacity, whereas the evacuation time without congestion sums the cost of the road segments that make up the evacuation route itself. When interpreting the final results, the user should report the maximum evacuation time with congestion and the maximum evacuation time without congestion to indicate the total evacuation time with and without congestion. That is, the tool assumes that all evacuees will start the evacuation at the same time, and the last evacuee with the longest evacuation time represents the total evacuation time. 7.6 Comparing the Power Model to the Bureau of Public Roads Function Power E. C = T ( f, c) = 1 ρ f e [7.1] The Power model is a traffic model that is used to calculate the congestion ratio based on the traffic flow and road capacity. (T) stands for the congestion ratio; (f) represents the traffic flow, which is basically the number of evacuating vehicles. (C) stands for the road capacity, which is the number of lanes for the road. The (f) and the (C) values are the main two inputs that the user should enter into the model to calculate the congestion ratios. In this research, the (f) value represents the number of evacuating vehicles and is derived from the behavioral survey and the (C) value represents the number of lanes for each road segment and is derived from the road network. (P) stands for the model ratio that is calculated based on the saturation density per unit capacity. (P) = 0.5 / (sqrt(saturation density Per unit capacity) * exp( )).the value of 118

135 (P) will change depending on the value of the saturation density per unit capacity that the user uses while setting up the evacuation options. As the number of the saturation density per unit capacity decreases, the congested evacuation time will increase. (E) stands for Epsilon with a value of ( ), which is a constant in the Power model. The value of (E) represents the optimum value of epsilon, the rate at which extra lanes can improve traffic according to the Power model (Shahabi and Wilson, 2014). The Power model represents a promising method for calculating congestion while simulating the evacuation. However, it is necessary to compare it to the Bureau of Public Roads (BPR) function, which is one of the most widely used methods to calculate traffic congestion. The BPR function is demonstrated in equation [7.2]: Time β V Congested = Time Freeflow α [7.2] C The (BPR) function depends on the volume and the capacity to calculate the evacuation congestion time. The Time Congested represents the final, congested travel time calculated using the (BPR) function. The Time Free flow stands for the free flow travel time on a link. (V) stands for the assigned volume on a link and (C ) stands for the capacity of the link. The (α) and (β) are constants and their values are based on the designed speed limit and road type. It is critical to compare the Power model to the (BPR) function to highlight the similarities and difference between these two methods of calculating the congestion time. The main objective of each of these two methods is to calculate the congestion travel time. Although they share the same objective, the mathematical settings (variables in the formula) of these two methods vary, which in turn affect the final result. While the Power model uses (P), model ratio and (E), Epsilon as constants in the model, the (BPR) function uses (α) and (β) as constants in calculating the congestion time. As mentioned above, the (P) value of the Power model is calculated based on the saturation density per unit capacity entered by the user, and the (E) value of the Power model represents the best optimum value of epsilon as a result of the simulation. 119

136 There are some limitations associated with the CASPER tool as compared to other traffic evacuation methods. First, CASPER uses capacity as number of lanes for each road segment; however, the concept of capacity in the transportation discipline is totally different, which is the number of vehicles per hour for each road segment. This is due to the fact that CASPER was not designed to solve traffic problems but rather to simulate evacuation and calculate congestion based on various variables. In reality, roadways with the same number of lanes may have different capacities (number of vehicles per hour) depending on the speed limit and the time needed to traverse each road. Second, calculating the travel time for each road depends on the speed limits of each road segment where the CASPER tool uses the speed as part of the route forecasting process. This is done through finding the best evacuation routes, if there are any, based on the number of lanes for each roadway segment and number of evacuees from each evacuee point. Then the tool utilizes vehicle density values (critical density per unit capacity and saturation density per unit capacity) to affect the speed, where the speed limit of each roadway may start to decrease as the number of evacuees meets the assigned critical density value. The speed of the road segment will become half of its original speed when the number of evacuating vehicles reaches the second density threshold, which is the saturation density per unit capacity. At this point, the speed will start to decrease exponentially and reflect on evacuation-congested times. In addition, the variables that are used in the CASPER tool to model the evacuation i.e., critical density per unit capacity, saturation density per unit capacity, and initial delay cost per evacuee are considered global parameters where all roadways of the network are treated in the same way using these parameters. Due to these limitations, the CASPER tool can be described as an algorithm that assists in finding the best evacuation routes based on the given variables; this is a qualification better than one which would require running an "all or nothing" assignment that basically sent people along the shortest path to safety. 7.7 Tsunami Evacuation Modeling Approaches The predicted number of evacuating vehicles based on the number of adult evacuees for each of the tsunami scenarios (10-, 20-, and 30-ft) is used as input into the CASPER tool. However, in order to simulate the evacuation process using the various variables of CASPER tool, two modeling approaches are adopted based on initial delay cost per evacuee: one second 120

137 and seven seconds. In describing these two approaches, it is critical to define the tool's parameters and how these parameters may affect the results when using each of the two approaches. Tool parameters include: 1) critical density per unit capacity, 2) saturation density per unit capacity, and, 3) initial delay cost per evacuee. The critical density per unit capacity means that each road can route up to that calculated equivalent of evacuating vehicles without affecting its traversal speed. Thus, the critical density per unit capacity is considered to be the minimum threshold of a road segment, and, that should the number of evacuating vehicles go beyond this value, the speed limit will drop, affecting the travel time for crossing that particular segment. The saturation density per unit capacity means that if a road segment routes this many evacuees, its traversal speed will decrease to half of its original traversal speed. The saturation density per unit capacity represents the maximum threshold of a road segment; if the number of evacuating vehicles goes beyond this value, the travel time of traversing this road segment will increase exponentially. The initial delay cost per evacuee represents the initial space between evacuees that are sharing their start locations. Changing the value of the initial delay cost per evacuee will result in different congested evacuation times depending on the space between evacuees who share the same location. As the value of the saturation density per unit capacity decreases, the congested travel time increases depending on the values of the initial delay cost per evacuee, the number of evacuees, and the capacity of the road network (number of lanes for each road segment).since there are no specific values that are considered to be the best and can work for all evacuation scenarios, the researcher decided to model the three tsunami evacuation scenarios using two approaches. The first approach is based on modeling the evacuation process using a one-second initial delay cost per evacuee with varying saturation densities per unit capacity (500, 250, 225, 200, and 100) while holding the critical density constant at 20 as a default setting. The second approach is based on modeling the evacuation process using a seven-second initial delay cost per evacuee with varying saturation densities per unit capacity (500, 250, 225, 200, and 100) while holding the critical density constant at 20 as a default setting. The value of the critical density per unit capacity was held constant at the default setting since it does not have a significant impact on the road's traversal speed compared with the saturation density per unit capacity. Modeling 121

138 the evacuation scenarios using these two approaches allowed the researcher to get different congested and un-congested evacuation times for each of the tsunami evacuation scenarios (10-, 20-, and 30-ft). The following section reports the results of both approaches and describes the reasons behind the variations within these results. 7.8 Modeling Tsunami Evacuation Scenarios Results One-Second, Initial-Delay Cost per Evacuee Approach As noted earlier in the previous section, the tsunami evacuations for the 10-, 20-, and 30- foot scenario events were modeled using two different approaches (one- and seven-second, initial-delay cost per unit evacuee). This section deals with the results of using a one- second, initial delay cost per evacuee, in the context of varying saturation density per unit capacity values (500, 250, 225, 200, and 100) for each of the three, tsunami evacuation scenarios. Tables summarize the evacuation time with and without congestion for each of the tsunami scenarios using one-second initial delay cost per evacuee. It is clear that the congested evacuation time, shown in Table 7.1 (and in Tables7.2 and 7.3) for each of the evacuating vehicle groups increases as the value of the saturation density per unit capacity decreases. For instance, when the saturation density per unit capacity for the first group of evacuating vehicles was 500 in Table 7.1, the congested time was 61 minutes; however, when the saturation density per unit capacity changed to 250, the congested time changed to 65 minutes. This is due to the fact that the roadway segments able to handle 500 evacuating vehicles per lane have a specific traversal speed that drops to half at the point when 501 vehicles are on the roadway. The traversal speed of the road continues to decrease exponentially as the number of vehicles on the road increases. However, when the saturation density changed to 250, the tool tried to find other roadway segments that could best handle the evacuation with these settings, which resulted in adding more roads to the existing ones that were used in case of 500 saturation density. Adding more road segments that had various traversal speeds led to an increase in the travel time along the route, which in turn increased the congestion resulting from the one-second 122

139 initial-delay cost per evacuee. The same explanation can be applied for the congested time of different evacuating vehicles using various saturation densities per unit capacity values (500, 250, 225, 200, and 100) in the Tables 7.2 and 7.3. The uncongested evacuation time reported in Tables represents the total cost of the roadway segments that made up the evacuation route without taking congestion into consideration. As also stated for the congested evacuation time, shown in Table 7.1, the value of the uncongested evacuation time increases due to adding more road segments to the evacuation routes and results in adding more cost to the existing cost, without considerations of congestion. While such increases may appear confounding, the uncongested evacuation time in the event of a 10-foot tsunami using 100 saturation density per unit capacity for the third and the fourth groups of the evacuating vehicles does, in fact, increase (Table 7.1) while it was supposed to decrease as a result of decreasing the number of evacuating vehicles. The best explanation for such a situation is that every time the tool models the evacuation, it may use routes different to the ones that were used for the previous number of evacuating vehicles, with different road segment costs. Thus, selecting the evacuation route that can handle the number of evacuating vehicles based on the specified saturation density per unit capacity affects the congested and uncongested evacuation time. The same explanation of the variation of the congested and uncongested evacuation time discussed for a10-foot tsunami scenario and reported in Table 7.1, can be applied in the results for tsunami scenarios of 20- and 30-feet reported in Tables 7.2 and 7.3. Table 7.1 Tsunami evacuation time with and without congestion for 10-ft tsunami scenario using 1 second Initial Delay Cost approach Adult per vehicle Total Evacuating Vehicles Time SDPUC(a) (500) SDPUC (250) SDPUC (225) SDPUC (200) SDPUC (100) C(b) U(c) C U C U C U C U 1 344,952 Minutes , ,476 Minutes , ,984 Minutes , ,238 Minutes , (a): Saturation Density per Unit Capacity, (b): Congested time, (C): Uncongested time 123

140 Table 7.2 Tsunami evacuation time with and without congestion for 20-ft tsunami scenario using 1 second Initial Delay Cost approach Adult Total Time SDPUC(a) SDPUC SDPUC SDPUC SDPUC per Evacuating (500) (250) (225) (200) (100) vehicle Vehicles C(b) U(c) C U C U C U C U 1 390,184 Minutes , ,092 Minutes , ,061 Minutes , ,546 Minutes , (a): Saturation Density per Unit Capacity, (b): Congested time, (C): Uncongested time Table 7.3 Tsunami evacuation time with and without congestion for 30-ft tsunami scenario using 1 second Initial Delay Cost approach Adult per vehicle Total Evacuating Vehicles Time SDPUC(a) (500) SDPUC (250) SDPUC (225) SDPUC (200) SDPUC (100) C(b) U(c) C U C U C U C U 1 418,883 Minutes , ,442 Minutes , ,628 Minutes , ,721 Minutes , a): Saturation Density per Unit Capacity, (b): Congested time, (C): Uncongested time Modeling the evacuation using 100 saturation density per unit capacity on one second initial delay cost per evacuee indicate that the CASPER tool does not provide a means for rerouting the evacuation traffic, the roadways on the evacuation traffic keep getting congested more and more which is clear from the unrealistic evacuation travel times Seven-Seconds, Initial-Delay Cost per Evacuee Approach Tables summarize the results of tsunami evacuation time, with and without congestion, for 10-, 20-, and 30-foot tsunami scenarios using a seven-second variable for an initial delay cost per evacuee with varying saturation densities per unit capacity values (500, 250, 225, 200, and 100). Tables indicate no significant change in the reported congested evacuation times for the same evacuation group (number of evacuating vehicles) while using various values for saturation density per unit capacity (500, 250, 225, 200, and 100). The major 124

141 differences in the congested evacuation times in Tables are due to using different numbers of evacuating vehicles. For instance, in Table 7.4, the congested evacuation time for evacuating 344,952 vehicles, using 500 as the saturation density and a seven-second initial delay, is 360 minutes. Whereas, congested evacuation time for evacuating 172,476 vehicles, using 500 as the saturation density and a seven-second initial delay, is 182 minutes. Since the saturation density per unit capacity and the initial delay per evacuee stays the same, the results of the congested evacuation time may vary depending on the number of evacuating vehicles. This is due to the fact that having seven seconds as the initial delay per evacuee gave more time for the rest of the evacuees to load into the road network. Using the same saturation density value, the same value for the initial delay and different numbers of evacuating vehicles will produce the same evacuation route with different congestion evacuation times. The same explanation will apply for all congestion times reported in Tables Using different saturation density per unit capacity might not produce noticeable change in the congested evacuation time in case of using a 7 seconds for initial delay cost per evacuee. For example, in Table 7.5, changing the saturation density per unit capacity to evacuate 130,061vehicles using the seven-second, initialdelay per evacuee option resulted in increasing the congested evacuation time by one minute for some values of the saturation density per unit capacity. Sometimes, a small change can be reported as in the above example, and, at other times, the change is too small to be reported since the difference might be in decimals. Table 7.4 Tsunami evacuation time with and without congestion for 10-ft tsunami scenario using 7 second Initial Delay Cost approach Adult per vehicle Total Evacuating Vehicles Time SDPUC(a) (500) SDPUC (250) SDPUC (225) SDPUC (200) SDPUC (100) C(b) U(c) C U C U C U C U 1 344,952 Minutes ,476 Minutes ,984 Minutes ,238 Minutes (a): Saturation Density per Unit Capacity, (b): Congested time, (C): Uncongested time 125

142 Table 7.5 Tsunami evacuation time with and without congestion for 20-ft tsunami scenario using 7 second Initial Delay Cost approach Adult per vehicle Total Evacuating Vehicles Time SDPUC(a) (500) SDPUC (250) SDPUC (225) SDPUC (200) SDPUC (100) C(b) U(c) C U C U C U C U 1 390,184 Minutes ,092 Minutes ,061 Minutes ,546 Minutes (a): Saturation Density per Unit Capacity, (b): Congested time, (C): Uncongested time Table 7.6 Tsunami evacuation time with and without congestion for 30-ft tsunami scenario using 7 second Initial Delay Cost approach Adult per vehicle Total Evacuating Vehicles Time SDPUC(a) (500) SDPUC (250) SDPUC (225) SDPUC (200) SDPUC (100) C(b) U(c) C U C U C U C U 1 418,883 Minutes ,442 Minutes ,628 Minutes ,721 Minutes (a): Saturation Density per Unit Capacity, (b): Congested time, (C): Uncongested time The uncongested time reported in Tables is consistent when using various saturation density per unit capacities for different numbers of evacuating vehicles for the 10-, 20-, and 30-foot tsunami scenarios. This is because all of the road segments that made up the evacuation route can handle any number of evacuees with differing congestion times depending on the number of evacuees since the initial delay cost per evacuee is uniform at seven seconds. The same explanation applies for all of the congested and uncongested evacuation times reported in Tables Comparing the results of the one-second, initial-delay approach with the results of the seven-second, initial-delay approach indicated that the results of the latter, the seven-second approach are closer to reality. This is because having evacuating vehicles loading into the road network within one-second intervals from the same location is not realistic. Using one second for 126

143 the initial delay in the three tsunami scenarios may produce different evacuation routes for every saturation density per unit capacity value, which in turn results in different congested and uncongested evacuation times (See Figures 7.1 and 7.2 for illustration). However, using the seven-second, initial delay resulted in selecting the same evacuation routes for the various saturation densities per unit capacity, but with different congested evacuation times depending on the number of evacuating vehicles (See Figures 7.3 and 7.4 for details). In the absence of the exact capacity (number of vehicles per hour) of the roads of the road network, one can use the results of the seven-second initial delay per unit capacity results with different saturation densities, per unit capacity values, to report the congested and uncongested evacuation times, as done for several tsunami evacuation scenarios for Orange County, California in this research CASPER Limitations Although CASPER tool has the ability to produce the shortest evacuation routes and estimate the evacuation time, with and without congestion, using different road capacities (number of lanes), it has several limitations. One of these limitations is that CASPER uses number of lanes to represent the road capacity, whereas in the transportation engineering discipline, roadway capacity is represented by number of vehicles per hour (Transportation Research Board, 2000). CASPER calculates road congestion based on critical density, saturation density per unit capacity, and the initial delay cost per evacuee all of which make up the global parameters where all of the road networks are treated entirely the same way, which is not realistic. In reality, the congestion rate may vary from one roadway to another even if these two roads have the same number of lanes and speed limits. Using one-second initial delay cost per evacuee in CASPER tool resulted in large amounts of travel time(s) due to the impact of the global parameters which treat all the roadways the same way; conversely, traffic assignment models, which do not treat all roadways the same, are able to re-route the traffic to an alternate route to reduce the overall travel time. 127

144 The CASPER tool also does not account for background traffic, which may affect the congestion rate and the evacuation clearance time. Furthermore, CASPER tool does not implement a traffic assignment methodology, but rather it finds the shortest path between evacuation origins and destinations. In addition, the CASPER tool does not take into consideration the complex turn restrictions that a network dataset may have. Another limitation of CASPER is that it does not have a dynamic module that specifically focuses on time-varying demand in order to identify the queues. CASPER tool does not also load the demand to the network based on an S-curve, which limits the demand loading methodology. However, CASPER sorts the evacuees at the beginning and starts the evacuation from the farthest evacuee to the safe point though, in reality, evacuees may overpopulate network bottlenecks and increase evacuation time. Finally, CASPER does not have the ability to identify the locations of critical bottlenecks during evacuation on the network. All of these limitations should be stated in order to evaluate the reported clearance time with and without congestion. The goals of this chapter were to identify evacuation routes in addition to estimate the evacuation clearance time with and without congestion using CASPER tool for different tsunami scenarios. Two different evacuation approaches were utilized, one with a one-second initialdelay cost per evacuee and the other with a seven-second initial-delay cost per evacuee, to simulate the evacuation process for each of the tsunami scenarios, i.e., 10-, 20-, and 30-foot. The results of using the one-second initial-delay cost per evacuee approach indicated that the congested and uncongested evacuation times increase as the value of the saturation density per unit capacity decreases. This is due to the fact that the time between loading each evacuee into the road network is one second, and with the limited number of evacuees per lane (saturation density per unit capacity), the congested evacuation time increased. This was clear specifically when using 100 saturation density per unit capacity on one second initial delay cost per evacuee, which indicated that the CASPER tool does not provide a means for rerouting the evacuation traffic, the roadways on the evacuation traffic keep getting congested more and more which is clear from the unrealistic evacuation travel times. However, when using the seven-second initialdelay cost approach, the congested evacuation time changes only due to the number of evacuating vehicles, where the same routes are used in every evacuation scenario. A small change of one to two minutes is detected in the congested evacuation time from changing one 128

145 saturation density per unit capacity value for another. The uncongested evacuation time, however, stays the same over all of the evacuation routes in various saturation densities per unit capacity values. This indicates that having the time space of seven seconds between each evacuee, while loading into the road network, will not change the selection of the evacuation route; however, it will affect the evacuation congested time based on the number of evacuating vehicles. The results from the use of the CASPER tool, employing the seven-second initial-delay per cost for evacuee approach, indicate this approach as the better alternative for assigning evacuation routes in the event of a tsunami threat. 129

146 Figure ft tsunami evacuation routes using one second for the initial delay cost per evacuee and 500 saturation density per unit capacity 130

147 Figure ft tsunami evacuation routes using one second for the initial delay cost per evacuee and 200 saturation density per unit capacity 131

148 Figure ft tsunami evacuation routes using seven seconds for the initial delay cost per evacuee and 500 saturation density per unit capacity 132

149 Figure ft tsunami evacuation routes using seven seconds for the initial delay cost per evacuee and 200 saturation density per unit capacity 133