Florida's Tornado Climatology: Occurrence Rates, Casualties, and Property Losses Emily Ryan

Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2018 Florida's Tornado Climatology: Occurrence Rates, Casualties, and Property Losses Emily Ryan

Follow this and additional works at the DigiNole: FSU's Digital Repository. For more information, please contact [email protected] FLORIDA STATE UNIVERSITY

COLLEGE OF SOCIAL SCIENCES & PUBLIC POLICY

FLORIDA’S TORNADO CLIMATOLOGY:

OCCURRENCE RATES, CASUALTIES, AND PROPERTY LOSSES

EMILY RYAN

A Thesis submitted to the Department of Geography in partial fulﬁllment of the requirements for the degree of Master of Science

2018

James B. Elsner Professor Directing Thesis

David C. Folch Committee Member

Mark W. Horner Committee Member

The Graduate School has veriﬁed and approved the above-named committee members, and certiﬁes that the thesis has been approved in accordance with university requirements.

ii TABLE OF CONTENTS

List of Tables ...... v List of Figures ...... vi Abstract ...... viii

1 Introduction 1 1.1 Deﬁnitions ...... 1 1.2 Where Tornadoes Occur ...... 3 1.3 Tornadoes in Florida ...... 4 1.4 Goals and Objectives ...... 6 1.5 Tornado Climatology as Geography ...... 6 1.6 Outline of the Thesis ...... 7

2 Data and Methods 9 2.1 Data ...... 9 2.1.1 Tornado Data ...... 9 2.1.2 Tropical Cyclone Tornado Data ...... 11 2.1.3 Property Value Data ...... 13 2.2 Statistical Methods ...... 15 2.3 Analysis Variables ...... 16 2.3.1 Occurrence Rates ...... 16 2.3.2 Casualties ...... 16 2.3.3 Property Exposures ...... 17

3 Time Variations in Occurrence 19 3.1 Yearly ...... 19 3.2 Monthly ...... 21 3.3 Daily ...... 27 3.4 Hourly ...... 29

4 Characteristics and Regional Variations 32 4.1 Tornado Characteristics ...... 32 4.1.1 Magnitude ...... 33 4.1.2 Path Length and Width ...... 37 4.2 Regional Variation ...... 39 4.2.1 Panhandle ...... 42 4.2.2 Peninsula ...... 44 4.2.3 Distance to Coast ...... 46

iii 5 Structural Property Losses 49 5.1 Property Value Analysis ...... 49 5.1.1 Structural Values ...... 49 5.1.2 Tornado Overlay and Exposure ...... 51 5.1.3 Case Study ...... 52 5.2 Modeling Property Loss ...... 56 5.2.1 Monte Carlo Simulation ...... 56 5.2.2 Actual Loss Estimates ...... 58 5.2.3 Probable Maximum Loss Estimates ...... 60

6 Major Findings, Conclusions, Limitations, and Future Work 62 6.1 List of Major Findings ...... 62 6.2 Conclusions and Limitations ...... 62 6.3 Future Work ...... 64

Bibliography ...... 66 Biographical Sketch ...... 71

iv LIST OF TABLES

1.1 Enhanced-Fujita Damage Scale. Source: Storm Prediction Center ...... 2

4.1 Tornado characteristics by F/EF rating ...... 37

4.2 Tornado characteristics as a function of distance to coast for the whole state of Florida 47

4.3 Tornado characteristics as a function of distance to coast for the Panhandle of Florida 47

v LIST OF FIGURES

2.1 Genesis location of Florida tornadoes from 1950-2016 by EF rating ...... 10

2.2 Genesis location of tornadoes in Florida associated with tropical cyclones from 1995- 2015 by EF rating ...... 12

2.3 One kilometer grid cells of property values in Florida (2014 USD) ...... 14

3.1 Yearly distribution of Florida tornadoes from 1987 to 2016 ...... 20

3.2 Yearly distribution of tornado casualties from 1987 to 2016 ...... 21

3.3 Monthly distribution of Florida tornadoes from 1987 to 2016 ...... 22

3.4 Monthly distribution of tropical cyclone tornadoes in Florida from 1995 to 2013 . . . 23

3.5 Monthly distribution of Florida tornado casualties from 1987 to 2016 ...... 25

3.6 Average monthly property exposure in Florida in 2014 dollars ...... 26

3.7 Daily tornado count distribution based on yearly and monthly trends in tornado occurrence from 1950 through 2016 for Florida ...... 27

3.8 Number of days with at least one tornado on a logarithmic scale...... 28

3.9 Hourly distribution of Florida tornadoes from 1987 to 2016 ...... 30

4.1 Monthly distribution of EF Rating to all tornadoes by percentage on a variable y-scale 34

4.2 Number of tornadoes that produced at least one casualty in Florida from 1987 through 2016...... 35

4.3 Genesis location of casualty producing tornadoes by EF Rating in Florida ...... 36

4.4 Annual average path length of Florida tornadoes in kilometers ...... 38

4.5 Annual average path width of Florida tornadoes in meters ...... 39

4.6 Annual average path area of Florida tornadoes in kilometers ...... 40

4.7 Monthly distribution of Florida tornadoes by county from 1987 to 2016 ...... 41

4.8 Monthly Tornado Distribution between the Panhandle and Peninsula ...... 43

5.1 Residential and non-residential structural value distribution based on one kilometer grid cells in Florida ...... 50

vi 5.2 One-kilometer cell with the largest residential structural value ...... 51

5.3 One-kilometer cell with the largest non-residential structural value ...... 52

5.4 Tornado paths in Florida from 1950 through 2016 ...... 53

5.5 1956 F3 Pembroke Pines to Fort Lauderdale tornado ...... 54

5.6 Structural value cell with the greatest tornado occurrence ...... 54

5.7 April 4, 1966 Tampa tornado path ...... 55

5.8 Monte Carlo simulated tornado paths ...... 57

5.9 Loss versus exposure from 2007 through 2015 ...... 58

5.10 Average annual property loss from Florida tornadoes ...... 59

5.11 Annual exceedance probability of property loss from Florida tornadoes as modeled in the Monte Carlo simulation ...... 61

vii ABSTRACT

Florida has a high frequency of tornadoes that occur throughout the United States. Together, Florida’s large population and expensive property, provides a great risk for injuries, fatalities, and damage to structures for when a tornado occurs. This risk of death or damage continues to increase as the population expands. The goal of this research is to better understand the tornado hazard in Florida by creating a climatology of Florida tornadoes through examining occurrence rates, casualties, and property loss. The tornado reports are obtained from the Storm Prediction Center’s Severe Weather database. Descriptive statistics are used to analyze temporal distributions, characteristics, and geographical distributions of tornadoes. Tropical cyclone tornado data from 1995 through 2013 is used for examining temporal distributions throughout the state. In addition, a new property value dataset put together by Georgianna Strode at the Florida Resources and Environmental Analysis Center is used to evaluate property loss from tornadoes throughout the state. Inferential statistics are used for testing hypotheses and modeling future tornado paths using a Monte Carlo simulation. Over the period from 1987 though 2016, there were 1,765 tornado reports in the state. The peak frequency occurs during the month of June with the overall tornado distribution mimicking the tropical cyclone distribution of the North Atlantic hurricane season. Majority of tornadoes occur in the peninsular region of the state, with tornadoes in the panhandle likely being stronger. There is a strong positive correlation between the amount of property exposed and the number of casualties produced by tornadoes. Although the majority of tornadoes that occur throughout Florida are very weak, the path length and width are shown to be increasing in recent years. Additionally, the annual average property loss estimate from tornadoes in Florida is $53 Million. Results of the Monte Carlo simulation indicate a 5% chance that the annual loss will exceed $203 million, a 1% chance that it will exceed $430 million, and a 0.1% chance that it will exceed $1 billion.

viii CHAPTER 1

INTRODUCTION

In this chapter, I deﬁne and describe the salient features of tornadoes and tornado activity and their various classiﬁcations. In addition, I provide background details on tornadoes and where they occur throughout the world, the United States, and in Florida. I address the main goals and objectives behind this research. Lastly, I provide a brief outlined description of each chapter.

1.1 Deﬁnitions

A tornado is a high speed, violently rotating vortex of wind, that can cause catastrophic damage to property and loss of life [57]. Most tornadoes develop as a result from thunderstorms, which develop when the atmosphere is unstable. Lower level warm air breaks through a stable layer and rises above a cooler and drier air mass. The rising air is warmer than its environment producing an upward acceleration of the air (updraft) due to buoyancy. The updraft results creates deep upward motion (convection) that feeds the thunderstorm. As the warmer air rises and these winds begin to twist the updraft, a rotating column of air is formed, known as the mesocyclone. As the rotating column of air aligns vertically, a tornado is formed [44]. In the United States, tornadoes are classiﬁed according to an estimated wind speed and rated based on the amount of damage they cause. The scale for rating tornadoes was originally a wind speed scale created by tornado researcher T. Theodore Fujita in 1971, called the Fujita (F) scale. There were six diﬀerent rankings on which tornadoes were rated, ranging from F0 (weak) to F5 (violent) [31, 3]. The Fujita scale was later revised, and in 2007, the Enhanced Fujita scale (EF) was implemented. This new rating scale rates tornadoes based on the damage they cause, in contrast to their wind speed [53, 3]. It ranges from EF0 (weak) to EF5 (violent), similar to that of the previous scale. The EF scale is more accurate because wind measurements from tornadoes have been relatively rare, therefore looking at the amount of damage caused tends to be a more useful indicator [18]. Only about one-third of tornadoes are rated higher than an F/EF2 and just a small percent of them are considered violent with a rating of F/EF4 or greater (Table 1.1)

1 The wind speeds in tornadoes are estimated based on the observed damage [29]. Table 1.1 shows the corresponding wind speeds for each EF rating, and a description of the damage that is observed at each rating. [17]. Although, direct wind measurements of tornadoes are relatively rare, with the use of WSR-88D Doppler radars, wind measurements are more accurately being obtained [18]. That being said, a majority of tornadoes tend to have wind speeds less than 70 meters per second [26]. However, built structures in the path of a tornado are vulnerable to destruction from winds that can exceed 80 meters per second with the greatest wind speed recorded being approximately 135–142 meters per second from the May 3, 1999 Bridgecreek, Oklahoma tornado [23, 18, 65, 11].

Table 1.1: Enhanced-Fujita Damage Scale. Source: Storm Prediction Center

Rating Wind Speed (m/s) Description EF0 29–37 Minor or no damage: Peels surface off some roofs; some damage to gutters or siding; branches broken off trees; shallow-rooted trees pushed over. EF1 38–49 Moderate damage: Roofs severely stripped; mobile homes overturned or badly damaged; loss of exterior doors; windows and other glass broken. EF2 50–60 Considerable damage: Roofs torn off well-constructed houses; foundations of frame homes shifted; mobile homes completely destroyed; large trees snapped or uprooted; light-object missiles generated; cars lifted off ground. EF3 61–73 Severe damage: Entire stories of well-constructed houses destroyed; severe damage to large buildings such as shopping malls; trains overturned; trees debarked; heavy cars lifted off the ground and thrown; structures with weak foundations are badly damaged. EF4 74–90 Extreme damage: Well-constructed and whole frame houses completely leveled; cars and other large objects thrown and small missiles generated. EF5 90+ Total destruction of buildings: Strong-framed, well-built houses leveled off foundations are swept away; steel-reinforced concrete structures are critically damaged; tall buildings collapse or have severe structural deformations; some cars, trucks and train cars can be thrown approximately 1.5 km.

Tornado path length and path width have been shown to positively correlate to one another. As the average path length increases, so does the path width. These two characteristics have also been proven to increase with EF rating [26, 7]. Based on damage path statistics from 2007 to 2013, the average path length of tornadoes in the United States ranges from approximately 2.27

2 kilometers to 71.95 kilometers and average path width ranges from approximately 54.9 meters to 1635.8 meters depending on the magnitude rating of the tornado [25]. It has also been observed that the mean path length and width of tornadoes has slightly increased over time [25].

1.2 Where Tornadoes Occur

Tornadoes are known to occur in all regions of the world where thunderstorms develop [29]. They have been observed on every continent, expect Antarctica [37, 8] and are most likely to occur at latitudes between 20 and 60 degrees latitude both north and south of the equator [32, 37, 4]. Tornadogenesis, or the process by which tornadoes form, is most favorable in these geographic areas because it is often where polar air meets subtropical air, which creates a boundary between the air masses and leads to convective precipitation. Storms that develop in this region often rotate due to the diﬀerent speeds and directions the air travels in the troposphere, further facilitating tornadogenesis [51]. Some tornadoes, however have occurred outside of this region and are not limited to these given latitudes. If conditions are favorable, tornadoes may occur anywhere on Earth. Some countries, such as the Netherlands, Spain, Italy, and Greece, [19, 20, 36, 34, 56] are better known for waterspouts occurring from non supercell storms, whereas countries such as United States, Germany, Argentina, Canada, and Mexico [37, 8, 20, 23, 63] are more likely to have tornadoes caused by supercell storms. The United States leads the world with the highest frequency of tornadoes occurring, followed by Canada. Other places, such as Europe, western Asia, Japan, Bangledesh, Australia, New Zealand, South Africa, China, and Argentina also have a relatively fair amount of tornadoes that occur each year [32, 51, 37]. Although tornadoes occur in other countries throughout the world, tornadoes in the United States are by far the most frequent and most intense. They are considered one of the most catastrophic natural disasters. Roughly 1,000 to 1,200 tornadoes occur each year in the United States [17, 16], which is more than any other country in the world [25]. Tornadoes in the United States are most likely to occur east of the Rocky Mountain Range in what is referred to as Tornado Alley, as well as in the southeastern region known as Dixie Alley. Tornado Alley or the Central Plains region, extends from central Texas northward into Nebraska, and Dixie Alley includes tornadoes that occur over Louisiana, Arkansas, Mississippi, Alabama, Georgia, and the western part of Tennessee [33]. These areas of the United States are more prone to tornadoes compared to other regions

3 because the topography and atmospheric conditions are particularly favorable for the development of storms that produce tornadoes [3]. The inherent mechanism that drives tornado development in the Central Plains and Eastern regions of the United States is the formation of supercell storms [8].

1.3 Tornadoes in Florida

Tornadoes in Florida, on the other hand, occur within lines of thunderstorms ahead of an advancing cold front from the northwest, within single isolated thunderstorms, or within the rotating thunderstorms associated with a tropical cyclone [40, 41]. Due to its proximity to the Gulf of Mexico and the Atlantic Ocean, Florida typically has enough water vapor in the air to support thunderstorms. Under these conditions, the main ingredient for tornadoes are winds above the ground moving with increasing speeds and veering directions, known as wind sheer. These tornadoes can still be intense, causing a lot of damage to property, in turn impacting many people. Many people do not see Florida as a tornado prone state, however the frequency of tornadoes that occur in Florida rivals that of states that are better known for tornadoes such as Kansas, Illinois, Iowa, and Oklahoma. The annual statewide occurrence rate of tornadoes in Florida over the past thirty years is four per 10,000 square kilometers. This happens to be the same rate of tornado occurrence in Kansas and actually exceeds the rate in Oklahoma by 22% [27]. According to the Southeast Regional Climate Center, Florida has the highest frequency of tornadoes in the southeastern United States on an annual basis [54]. On the other hand, Florida has the lowest frequency of strong tornadoes (rated EF2 or greater) in the top ten tornado prone states [64]. Strong tornadoes, however, do occur in the state and can have extreme effects. An example of a strong tornado in the state would be the 1925 Miami Tornado which caused five deaths and produced damage totals that were estimated between $200,000 and $300,000 1925 USD [39]. The state of Florida has the third largest population in the United States as of 2017 [10]. This should not be shocking considering the many large cities such as Miami, Orlando, Tampa, Jack- sonville, Tallahassee, and more that have very dense populations. This creates a high population exposure as well as property exposure for when a tornado occurs in the state. According to a national study conducted in 2014, Florida tornadoes were found to be the deadliest in the nation because “the state’s vulnerable population and flimsy housing make the tornadoes that hit here

4 deadlier” [1]. As the population continues to rise, it is important to understand the temporal and spatial variability of Florida tornadoes. Based on population statistics, Kansas has 2.9 million people as of 2016, and Florida has just under 21 million, therefore it is about seven times more likely that a tornado will affect a person in Florida because the distribution of tornadoes favors coastal regions that tend to have a higher population [10]. Although, tornadoes are still considered to be rare events, they are known to cause extreme damage to structure and property, resulting in high financial losses, in addition to injuries and deaths. On average, there are 1,500 injuries and 80 deaths that are reported due to tornadoes each year [16]. Furthermore, average annual losses as a result of tornadoes has been estimated to be $982 million dollars, which is more than double that of previously reported annual tornado loss estimates from the National Weather Service (NWS) [13]. There has also been an increase in occurrence in weather-related disasters that has been shown to be a function of climate and society. This has led to a greater exposure rate because of economic growth and society, which is an important influence in the rising impacts of disasters [59]. Economic losses due to tornadoes have been increasing over time. This is due to inflation, the concentration of population, and urban sprawl which is the tendency for cities to expand [17, 8]. With increasing prices, rising population, and growing cities, tornadoes are sure to be an important hazard to both human life and property. For comparison, the median price of a home in Kansas is $128,400 and in Florida it is $217,300 [66]. The price of a home in Florida is 1.7 times higher than in Kansas, so if a tornado touches down, the damage that could occur is likely to be more costly in Florida due to the increased value of homes. By examining the climatology of Florida tornadoes through researching occurrence rates, casualties, and property losses, it can help answer questions about the tornado climatology in the future. As the average global temperature continues to warm and the climate continues to change, characteristics of tornadoes may be altered. Some studies have predicted that tornadoes may be larger, stronger, and more intense, while others show changes in the frequency distributions and clustering effects of tornadoes [47, 25]. More people, as well as property will be exposed to tornadoes, which puts a greater risk for damage to homes, injuries, and/or death as Florida's population continues to grow. The most recent study describing a climatology of Florida tornadoes

5 focused on days where four or more tornadoes occurred [40]. No study has analyzed the full scope of the Storm Prediction Center’s tornado data set for the state of Florida as a whole.

1.4 Goals and Objectives

The overall goal of this research project is to better understand the tornado hazard throughout the state of Florida. The main objectives of my research are to create a climatology of tornadoes in Florida through analyzing occurrence rates, casualties, and property losses. Here, I show the characteristics of Florida tornadoes and their distributions spanning from 1987 through 2016 for a true (30-year) climatological period. I ﬁnd that some aspects of my tornado climatology contradicts previous studies of tornadoes in Florida. Toward this end, I will ﬁrst analyze the temporal variability of tornadoes in Florida by focusing on the yearly, monthly, daily, and hourly distributions of tornadoes over time. I will also compare distributions of tornado casualties and property exposure from tornadoes. Next, the characteristics of tornadoes in Florida will be analyzed and compared to tornadoes throughout the United States. In addition, I will examine the geographical distributions of tornadoes in Florida by separating the panhandle region from the peninsula to get a more in depth understanding of the climatology. Furthermore, the tornadoes will be overlaid onto Florida property value grid cells to evaluate the average annual property loss from tornadoes in Florida. Lastly, a Monte Carlo simulation method will be used to estimate the annual average property loss from tornadoes. Throughout this research, I show a variety of new graphs portraying spatial and temporal distribution of tornadoes in a new manner, which have never been used before in a tornado climatology.

1.5 Tornado Climatology as Geography

Tornado climatology is part of the broad discipline of geography. Geography is a diverse, interdisciplinary science that can be split into two diﬀerent areas, human geography and physical geography. Human geography is the study of people and their interaction with the environment in relation to place and space, while physical geography is the study of processes in the natural environment relating to the atmosphere, biosphere, hydrosphere, and geosphere. Over the years, geography has been known to bridge the gap between social sciences and natural sciences. In a way, it is the science of all sciences, which is where my research ﬁts in. Within my research, my

6 analysis ties into both the human and physical areas of geography. I am interested in the spatial relationship between tornadoes, humans, and property. Altogether, this research will combine the components of both human and physical geography into one to create a climatology of tornadoes in Florida.

1.6 Outline of the Thesis

In chapter two, I describe the data used, the statistical methods employed, and the variables that will be analyzed. The primary data are the records of tornadoes from the Storm Prediction Center’s severe weather database, tropical cyclone tornado data, and property values for the state of Florida. The statistical methods used can be divided into descriptive statistics, hypothesis testing, and predictive inference. All three methods will be used in relation to one another. The diﬀerent variables used for analysis are the occurrence rates of tornadoes, casualties caused by tornadoes, and the amount of property exposed to tornadoes. Chapter three looks at the climatology of Florida tornadoes by analyzing the temporal distributions. By using descriptive statistics, I provide information on yearly, monthly, daily, and hourly frequency distributions of tornadoes. I also take into account casualty and property exposure distributions to have a better understanding of the tornado hazard with respect to changes over time. In chapter four, I study tornado characteristics in Florida by looking at the magnitude, path length, and path width of tornadoes. I use a combination of descriptive statistics and hypothesis testing to examine regional variations in tornado frequencies between the panhandle and peninsular regions in the state. Furthermore, I hypothesize that tornadoes become stronger as they travel away from the coast and inland. Chapter ﬁve examines the structural property exposure to tornadoes in Florida. In this chapter, I analyze residential and non-residential structural values across the state. Then, I overlay the past tornado events onto the property value grid cells and calculate exposure rates for each tornado event. In addition, I evaluate the amount of structural exposure for the worst case scenario that occurred in Florida. Lastly, I use predictive inference to create a Monte Carlo simulation of possible future tornado paths. I use these characteristics to evaluate actual and maximum loss estimates of property exposed to tornadoes.

7 Chapter six summarizes my research project by highlighting the major ﬁnding in my thesis. I give a brief summary of each chapter, while also addressing some limitations observed from the data sets and methods used. Lastly, I provide some insight on future initiatives for dealing with tornadoes in the state of Florida.

8 CHAPTER 2

DATA AND METHODS

In this chapter, I provide a brief description of the datasets used, statistical methods employed, and variables analyzed for conducting my research. I state the underlying issues with each dataset and the reasons for using only certain years. I provide background information on each diﬀerent statistical method that is applied to my research. Additionally, each analysis variable is brieﬂy described according to how it corresponds with the analysis of tornadoes in Florida. Together, the datasets, statistical methods, and analysis variables that are described in this chapter illustrate how I obtain the latest climatology for tornadoes in Florida.

2.1 Data 2.1.1 Tornado Data

The tornado data was obtained from the Storm Prediction Center’s (SPC) severe weather database, which is derived from all recorded tornado reports from the National Weather Service’s (NWS) Storm Data. The data provides information on nearly every tornado occurring in the United States from 1950 to 2016. Tornado records contain information on initiation point and endpoint (longitude and latitude), date and time of occurrence, length and width of the damage path, and maximum damage rating on a scale from zero to five based on the Fujita (F) or Enhanced-Fujita (EF) scale. There were 3,278 tornadoes that are associated with the state of Florida. Figure 2.1 shows the genesis location of these tornadoes from 1950 through 2016 with their respective F/EF ratings. The tornado dataset also contains an estimate of property loss in millions of dollars since 1996 and whole dollar amounts since 2007. Loss columns that contain a zero value does not necessarily mean there were no losses. The tornado reports in the database are compiled by the NWS offices and then reviewed by the National Centers for Environmental Information (formally known as the National Climate Data Center) [62]. The database is available in a shapefile format with each tornado being represented by a straight-line track. The tornado track is the line between the initiation point and the end location, with no width. The recorded locations in the attribute

9 Tornado Genesis (1950−2016) F/EF0 F/EF1 F/EF2 F/EF3 F/EF4

Figure 2.1: Genesis location of Florida tornadoes from 1950-2016 by EF rating

10 table have a two-digit precision prior to 2009 and four digits afterwards. Later records have greater precision because the estimates were made using a Global Positioning System (GPS). Limitations of the SPC tornado dataset have been very well documented in past research. One limitation is the lack of accuracy and temporal consistency of reports [9, 30]. There is also bias in the magnitude rating of the tornado’s intensity because the original F scale was used to estimate the maximum wind speed associated with the tornado. After January 2007, the rating of tornadoes was determined by the EF scale, providing a more accurate estimation of damage to structures that were surveyed by the NWS [18, 24]. However, the surveys may be biased due to the opinion of the surveyor. Tornadoes that occur in lower populated areas may often go unwitnessed and unreported [26]. These tornadoes are also less likely to cause structural damage or casualties, however the intensity of these tornadoes may also be underestimated [5] leading to another bias in the dataset. As the population of the United States grew, the number of reported tornadoes increased. Fuhrmann et al. (2014) observed a doubling of tornado reports from the 1950’s to 2000’s which are likely attributed to better technology, population growth, and more tornadoes being reported [30]. Since the EF scale is a damaged based scale, there may be bias in the rating of the tornadoes if they occur in flat areas with little or no structures. An accurate estimation of the wind speed within the vortex is unlikely to be obtained due to the lack of structures damaged, resulting in the rating being underestimated. An example of this is the El Reno, Oklahoma tornado that touched down on May 31, 2013. Two different Mobile Doppler Radars sampled the tornado’s wind speeds, with one radar’s data measuring wind speeds greater than 295 miles per hour (approx. 132 meters per second). This would have put the tornado’s overall rating at an EF5, however, the NWS survey teams indicated no structural damage beyond an EF3 according to the visual eye, resulting in an official rating of an EF3 [52]. Thus, the overall EF ratings of tornadoes in the data set may be slightly biased by NWS surveyors, leading to a limitation in the EF scale.

2.1.2 Tropical Cyclone Tornado Data

The tropical cyclone tornado data used is provided by Todd Moore, an Assistant Professor for the Department of Geography and Environmental Planning at Towson University. He used Roger Edwards’ TCTOR database which is a combination of the SPC’s ONETOR tornado database and the National Hurricane Center’s (NHC) HURDAT database. The TCTOR database includes

11 Tropical Cyclone Tornado Genesis (1950−2015) F/EF0 F/EF1 F/EF2

Figure 2.2: Genesis location of tornadoes in Florida associated with tropical cyclones from 1995-2015 by EF rating

12 information on all tropical cyclone tornadoes from 1995 to 2015 and each tornado’s association with its parent cyclone. All tornadoes in the database have been veriﬁed with the use of surface and upper air maps, archived satellite photos and/or imagery derived from archived NEXRAD Level II data [21]. Figure 2.2 shows the genesis location of tornadoes in Florida by EF rating that are associated with tropical cyclones from 1995-2015. There are a total of 313 tornadoes reported in this dataset. The major limitation of this dataset is that it only dates back to 1995, providing only 20 years for analysis. The dataset starts in 1995 to correspond to the National deployment of the WSR- 88D radar [21]. Although it is still helpful, thirty years of data is considered optimal for a full climatology. Another limitation is due to the infrequent nature of tropical cyclones occurring and spawning tornadoes, which creates small sample sizes.

2.1.3 Property Value Data

Risk assessment of property structural values can be diﬃcult to obtain due to the large number of individual properties and inconsistent tools for measurement. The dataset used in this research was developed to provide a consistent method for measuring property values across the state of Florida. In this study, the source information is from the 2014 Florida Department of Revenue’s cadastral database. This data was created by Georgianna Strode at the Florida Resources and Environmental Analysis Center. The high-resolution property values were determined by each county’s local property appraiser and aggregated to the U.S National Grid, which are one kilometer by one kilometer cells. The data contains assessed value, land value, and structural (material) value for all land use types and for residential parcels. I am particularly interested in the residential and non-residential structural values in each grid cell. Georgianna Strode calculated these by subtracting the land value from the assessed value and then aggregating them to the one kilometer cells. Figure 2.3 shows a map of the one kilometer grid cells of property values in Florida. All dollar amounts in the dataset are represented as 2014 USD. Lighter shades correspond to smaller property values, while darker shades correspond to larger property values, with areas of white having no value. There are 91,180 one-km cells with at least some structural property [27]. Areas with high structural values are located along coastal areas such as Miami, Fort Meyers, and Tampa as well as in cities such as Orlando, Jacksonville, and Tallahassee.

13 Structural Property Values (2014) $1−10M $10−25M $25−50M $50−75M $75−100M $100−200M $200−250M > $250M

Figure 2.3: One kilometer grid cells of property values in Florida (2014 USD)

14 2.2 Statistical Methods

The different statistical methodologies that I use in my research can be split into two different categories: descriptive and inferential. Descriptive statistics used in my research include tables and graphs of frequency distributions to see how the variables in the data are distributed. The frequency refers to the number of observations that are in a group [38]. Descriptive statistics provides a way to simply and quantitatively describe what is happening in the data. Some tables are used for describing variables, however the use of graphs provides a better visual impression of the data. Therefore, I tend to focus on using different types of graphs, such as histograms, line graphs, and scatter plots to depict relationships between variables. Histograms are bar charts that show the distribution of a variable and are measured at the interval or ratio level. Line graphs show the frequency value for a class and are connected by a line. Scatter plots show interval or ratio variables which contain information about the extent of differences between pairs of observations [38]. These types of graphs create a more useful visual representation and understanding of the data, which is an important factor in the analysis process. Inferential statistics are used to measure the relations and effects between variables by coming to conclusions that extend beyond the data itself [61]. In my research, I use two different types of inferential statistics. The first technique is used to make inferences about what I think and to test explanatory theories. The scientific method provides the foundation for hypothesis testing. It is the “process by which we acquire new knowledge and refine our existing understanding about a phenomena we observe” [38]. It involves making observations, developing hypotheses, experi- menting, and analyzing the information gathered. In my research, I use the scientific method to draw inferences about my data and test hypotheses to refine my knowledge and understanding of tornadoes throughout Florida. The second technique I use is predictive inference. I use this to test predictive theories on future observations that are based on past events. An example of this in my research is using a Monte Carlo simulation to create future tornado events that are based on the historical past events. I can then have a better understanding of when, where, and how tornadoes will effect Florida in the future.

15 2.3 Analysis Variables 2.3.1 Occurrence Rates

I look at the long-term trends and climatology of tornadoes throughout the state of Florida. The analysis starts in 1987 and continues through 2016 to get a full thirty-year climate period. It is important to note that Doppler (WSR-88D) Next-Generation Weather Radar was done being commissioned by the NWS in 1988 [45]. In 1994, two Doppler radars were installed in Florida and another four were installed in 1995 [2], therefore reports for the latter half of the years are considered to be more accurate and provide more precision. I use descriptive statistics to plot tornado frequency distributions as histograms and line graphs to analyze the temporal distribution of tornadoes throughout Florida. This research includes yearly, monthly, daily, and hourly patterns of tornadoes. I am interested to know when the highest occurrence rate of a tornado is likely to take place, based on the temporal aspects. In addition, I also analyze the characteristics of tornadoes such as magnitude, path length, path width, and path area. By looking at the characteristics of Florida tornadoes, I am able to validate past research and better understand the relationship of the strength and intensity with respect to temporal and geographical distributions. By comparing the similarities and differences between the panhandle and peninsular portions of Florida, I also analyze the regional distributions of tornadoes. I then compare my results with the findings from previous studies. In addition, I hypothesize about the relationship of storm strength and tornadoes and their distance from the coast. As tornadoes in Florida move further away from the coast onto land, specifically in the panhandle region of the state, they strengthen. I calculate the distance each tornado genesis location was from the Florida coastline and analyze scatter plots and spatial statistics to prove or disprove my hypothesis.

2.3.2 Casualties

I analyze information relating to the number of injuries and fatalities caused by tornadoes in Florida. A tornado casualty is deﬁned as the number of injuries plus the number of fatalities caused from any single tornado that touched down in the state. A tornado that causes a casualty is considered a casualty producing tornado. To better understand where the increased potential of

16 tornado hazards are to human life in Florida, I analyze the temporal and geographical distributions of casualty producing tornadoes. A tornado casualty must be directly attributed to a tornado, or to impact by airborne, falling, or moving debris. An example of a direct fatality would be a driver killed when a motor vehicle is tossed over. To be considered a direct injury, the injury must require treatment by a ﬁrst- responder or at a medical facility. An example would be a person treated for a laceration caused by ﬂying debris. Indirect casualties are not considered. An example of an indirect casualty would be electrocution during debris removal. I take into account the number of casualties per tornado and plot distributions relating to the temporal aspects observed with the occurrence rates of tornadoes in the state. I analyze the yearly and monthly distributions of casualty producing tornadoes, while comparing and contrasting them with the tornado’s respective frequency distributions. I also determine what day was associated with the highest casualty rate from tornadoes, as well as when the highest number of casualties is likely to occur throughout the day. I then analyze the geographical distributions of casualty producing tornadoes by F/EF rating. Using the tornado characteristics, I compare and contrast their distributions to casualties. This helps me better understand where people are more at risk to be injured or killed by tornadoes in Florida. I hypothesize that when controlling for storm intensity, casualties are higher in the panhandle portion of the state than in the peninsula because structures in the peninsular portion of the state are likely to withstand higher wind speeds.

2.3.3 Property Exposures

To look at property exposures and losses from tornadoes, I overlay all tornadoes from the full data set onto the property grid cells and calculate the amount of structure under each tornado path. For this analysis, all tornadoes that occur in Florida from 1950 through 2016 were used because the property value analysis is not inﬂuenced by a possible bias in the tornado records. With this information, I calculate the average annual total, residential, and non-residential property exposures. I compare and contrast property exposure by year and month to the frequency distributions of tornadoes to understand when the highest risk to structures is likely to occur in Florida. In addition, I also analyze the F/EF rating of tornadoes to property exposure and look at geographical aspects to better understand where the tornado hazard is most likely to cause a substantial amount of structural damage throughout the state.

17 With the use of inferential statistics, I also use a Monte Carlo simulation algorithm to permute potential future tornado paths. I then graph and analyze the distribution of the annual average loss of tornadoes and the annual exceedance loss. I use these distributions to estimate the annual probability maximum loss (PML) as the value of the largest loss that could occur from tornadoes. Insurance companies can then use this information to allocate money where it is best suited before and after a tornado disaster occurs.

18 CHAPTER 3

TIME VARIATIONS IN OCCURRENCE

This chapter examines the climatology of Florida tornadoes by analyzing the yearly, monthly, daily, and hourly distributions of tornadoes in Florida from 1987 through 2016. Tornadoes throughout the world, the United States, and Florida vary both spatially and seasonally in their distribution and frequency [6]. According to the Southeast Regional Climate Center, Florida has the highest frequency of tornadoes in the southeastern United States on an annual basis [54]. This is due to there being a high number of very weak tornadoes produced from hurricanes, sea breeze interaction, and water spouts moving over land [16]. By looking at the diﬀerent temporal distributions, forecasters are able to use risk assessment analyses to determine the probability of a tornado striking based on the time of year or time of day with diﬀerent weather events. This leads emergency managers and other community members to better prepare and/or mitigate an oncoming potential tornado disaster. It may also help to reduce the number of fatalities and/or injuries when a tornado strikes.

3.1 Yearly

The yearly distribution of tornadoes in Florida for the thirty-year period from 1987 to 2016 is shown in Figure 3.1. There were 1,765 tornadoes recorded in Florida, with a median of 56 and mean of 59 occurring each year. The years with the highest frequency of tornadoes occurred in 1997, 1998, and 2004, with the number of tornadoes being 115, 109, and 104, respectively. The lowest frequency of tornadoes occurred in 2010, with only 19 reports. There is a multimodal distribution observed, with an average yearly tornado occurrence that appears to be decreasing in the 2000’s. Between 1987 and 2001, there was a total of 1,088 tornadoes with an average of 71 tornadoes occurring each year, where as between 2002 and 2016, 677 tornadoes occurred with an average of 50 each year. The Welch two-sample t-test is used to determine whether or not there is a shift toward fewer tornadoes since 2002. The reason for the t-test being used as opposed to a diﬀerent statistical test is due to the smaller sample size. After conducting the t-test between the average number of tornadoes per year in both periods, it appears that the latest 15 years (2002-2016) does

19 120 115 109 104

90 85 77 77 71 70 66 60 60 60 56 56 56 56 49 48 45 46 45 43 44 42 37 39 37 30 Number of Tornadoes 30 28 19

1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 Year

Figure 3.1: Yearly distribution of Florida tornadoes from 1987 to 2016 have a decrease in the frequency of tornadoes per year compared to the earlier 15 years (1987-2001) (p-value < 0.001). In addition, the Mann-Whitney U test was used as a nonparametric test to further test the shift in tornado occurrences because it does not require the distribution to be normal. The p-value obtained from this test was < 0.001, further validating the hypothesis of a decrease in activity. While we do not know the cause of this statistical shift, there is clear evidence that tornado frequency in Florida has declined and should be further investigated. Figure 3.2 shows the yearly distribution of tornado casualties in Florida from 1987 through 2016. There is an average of 38.3 casualties with a median of 16.5. The peak number of casualties occurred in the year 1998 with 109 tornadoes causing 351 casualties. As observed in the yearly distribution, the frequency of tornadoes throughout the state is declining. This observation, seems to be mirrored in the annual casualty rate, however it is masked by outliers in the data. Between 1987 and 2000, there was an average of 59 and a median of 31 casualties per year. From 2001 to 2016, there was an average of 17 casualties per year with a median of 4. Because the mean is much greater than the median when comparing the years, it is safe to say that the average yearly

20 300

200

Number of Casualties 100

1990 2000 2010 Year

Figure 3.2: Yearly distribution of tornado casualties from 1987 to 2016 casualty rate is being slightly skewed by the outliers. The obvious trend toward few casualties is likely due to increased awareness from the advancement of technology, more accurate watches and warnings being issued by the National Weather Service, but also due to fewer tornadoes occurring as well.

3.2 Monthly

Tornadoes that are spawned from cold fronts and squall lines are most commonly produced in the winter and spring months. Fall tornadoes are likely generated by tropical cyclones or hurricanes, and tornadoes during summer months likely result from air-mass and sea-breeze inﬂuenced thunderstorms [43]. By analyzing the seasonal and monthly distributions, we can obtain a more accurate understanding of the tornado occurrence rate variability. The monthly frequency of tornadoes in Florida for the thirty-year period is shown in Figure 3.3. The month of June has the highest frequency of tornadoes with a total of 270, followed by September, August, and July, having 233, 194, and 190 tornadoes, respectively. With less than 100 tornadoes occurring, the months of

21 300 270

223

200 190 194

154 156 144

101 104 100 95 82 Number of Tornadoes 52

April May June July March August January February October September November December Month

Figure 3.3: Monthly distribution of Florida tornadoes from 1987 to 2016

December (52), November (82), and January (95) have the lowest frequency of all months. There is a bimodal distribution with peak frequencies in the spring and summer. Hagemeyer (1997) observed a minimum in November for the peninsular portion of Florida, however the minimum for this study shows a minimum in December [40]. This same study showed a peak in June, which is consistent with my ﬁndings for the state as a whole. Climatologically speaking, the peak in the summer months and into fall can be attributed to tornadoes that are spawned from tropical cyclones, whereas the peak in the spring is likely due to severe weather outbreaks. Brooks et al. (2003) suggest that the summer maximum peak is attributed to weak tornadoes (F/EF0 and F/EF1) and speculate that the causes of these tornadoes are due to non-supercell events such as waterspouts moving onto land or from sea-breeze interactions [9]. Waterspouts in Florida are most likely to occur along the coast of Florida during the summer months. There are two types of waterspouts which are either associated with fair weather or accompanying storms [64]. The fair weather type of waterspouts are weaker, more common, and tend to develop over shallow water where solar heating is greatest. Places where this type is

22 127

100

61 51 50 45

12 7

Number of Tropical Cyclone Tornadoes Number of Tropical 0 0 0 0 2 2 0

April May June July March August January February October September November December Month

Figure 3.4: Monthly distribution of tropical cyclone tornadoes in Florida from 1995 to 2013 likely to be seen include the Tampa Bay region, Pensacola Bay, Biscayne Bay, and the Keys. The second type of waterspouts are stronger, less common, and can originate from squall lines, cold fronts, thunderstorms, hurricanes, and land tornadoes that move over water [64, 43]. Waterspouts over bay regions are common because the sea breeze enhances cumulonimbus cloud development, allowing for stronger, more destructive waterspouts to develop [43]. To further observe the speculation of Brooks et al. (2003), I analyzed the monthly tropical cyclone distribution in Florida. The tropical cyclone data provided me with all tornadoes that were spawned as a result of tropical cyclones from 1995 through 2015 in Florida. Altogether, there were 313 tornadoes associated with tropical cyclones. Each tornado in the SPC tornado data set that occurred from a tropical cyclone was labeled. The tropical cyclone tornado trends and distributions were then analyzed separately from the rest. The monthly tropical cyclone tornado distributions for Florida are shown in ﬁgure 3.4. The Atlantic hurricane season starts June 1st each year and goes through November 30th [12]. Agee & Hendricks (2011) suggest that Florida mirrors the seasonal patterns of tropical cyclones in the

23 North Atlantic [2]. Tropical cyclone tornadoes peak in September with a high of 127 tornadoes for the period from 1995 to 2015, which does correspond to the peak of the Atlantic hurricane season. In addition, tropical cyclones have spawned two tornadoes outside of this season in both May and December. It is likely that the speculation from Brooks et al. (2003) is correct from comparing the monthly distributions of tornadoes to tropical cyclone tornadoes in Florida. As stated earlier, Florida has the highest frequency of tornadoes in the southeastern United States [54]. This monthly analysis is particularly interesting because the frequency of tornadoes in Florida has been assumed to be similar to that of the southeastern states as a whole. The Southeast Regional Climate Center (SERCC) has found that most tornadoes, as well as tornado casualties, occur during the months of March to May, with a peak in April. In the SERCC analysis there was also a secondary peak in November [54]. Compared to the southeastern states as a whole, Florida is unique in that its peak frequency is in June, with a minimum in December. According to my monthly analysis of Florida tornado casualties, the highest number of casualties has occurred in February with a total of 427. This peak is influenced by a high casualty rate resulting from a tornado outbreak that occurred on February 22-23, 1998. Figure 3.5 shows the distribution of tornado casualties by month from 1987 through 2016 in Florida. There is bimodal distribution with a small secondary peak in casualties that occurs during October and November. The peak in casualties for southeastern states occurs in April, which corresponds to the frequency of tornadoes for the region as well [54]. The peak shown in Florida, which occurs in February, corresponds to the secondary peak in casualties for southeastern states. The difference between the April peak in the southeastern states and the February peak in Florida, may have to do with the difference in tornado climatology for the regions. Tornadoes in February for Florida tend to be stronger and rated higher, which may be linked to the high casualty rate for the month. Chapter 4 provides a more in-depth analysis on the magnitude of tornadoes in Florida. Now, many tornadoes occur during the summer and early fall as see in figure 3.3, however there are very few casualties during those months (figure 3.5). I computed the correlation between the frequency of tornadoes by month and the number of casualties by month using a Pearson-product moment correlation test. The value is -0.31 [(-0.75, 0.33) 95% uncertainty interval). The result indicates a negative correlation that is not statistically significant. This is because many of the

24 427

400

300

200 187

118 108 100 82 Total Number of Casualties Total

46 38 46 38 25 19 15 0

April May June July March August January February October September NovemberDecember Month

Figure 3.5: Monthly distribution of Florida tornado casualties from 1987 to 2016 tornadoes during these months are very weak tornadoes spawned from tropical cyclones and do not result in a signiﬁcant amount of casualties. A study conducted by Belanger et al. (2009) on Hurricane-induced tornadoes (HITs) explored the frequency of tornadoes that were associated with land falling tropical cyclones and their variability of occurrence [42]. In conjunction to this study, there was a similar study published on this same concept by Agee & Hendricks (2011), but focused on the role of technology used in the state of Florida [2]. The authors counted all potential HIT days, in addition to all land falling TCs and HITs. It has been shown that Doppler radar is more accurate in predicting the estimate of HIT occurrences than previous methods, which can be very useful for predicting HITs in the future with respect to a warmer climate [2]. These previous studies are helpful to understand the frequency of HITs in Florida and are of interest when analyzing casualties and property exposure. Most tropical cyclones that make landfall in the United States spawn tornadoes, which is often times overlooked [35]. Although these tend to be weak tornadoes, cases of signiﬁcant damage and deaths have occurred [42].

25 $15,000,000

$10,000,000

$5,000,000 Average Exposure Average

April May June July March August January February October September NovemberDecember Month

Figure 3.6: Average monthly property exposure in Florida in 2014 dollars

In addition to monthly distributions of casualties and casualty producing tornadoes, I also analyzed the average monthly property exposure in Florida. All dollar amounts are in 2014 dollars no matter which year the tornado occurred. More details about property values, exposure, and loss are explained later in Chapter 5. Figure 3.6 shows the average monthly distribution of property exposure in Florida. The important thing to note is the peak exposure occurring in February, which also corresponds to the peak in casualties. Again, this may correspond to the climatology of tornadoes during the transition of weather patterns from winter into spring. The peak average exposure in February is $15,200,816 with a minimum average exposure in July of $723,898. Because the distributions in the number of tornadoes, number of casualties, and average monthly exposure diﬀer, it may be more beneﬁcial to distinguish Florida as its own entity when developing a climatology of tornadoes for southeastern states. The casualty and property exposure distributions by month are fairly similar, with both having a maximum occurring in February and a minimum occurring in July. I computed the correlation between the total number of casualties and average property exposure by month using the Pearson-

26 2010 2000 1990 1980 1970 1960

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

1 2 3 5 9 16 28 Daily Tornado Count

Figure 3.7: Daily tornado count distribution based on yearly and monthly trends in tornado occurrence from 1950 through 2016 for Florida product moment statistic. The value is .93 [(.77, .98) 95% uncertainty interval). This results in a strong positive correlation between values, which makes sense for the fact that more densely populated areas correspond to higher property values. If a tornado hits one of these areas, the resulting number of casualties and amount of property damaged will likely be higher.

3.3 Daily

Florida has been known to have the highest rate of thunderstorm activity compared to anywhere else in the United States [64]. There are anywhere between 65 and 80 days reported to have thunder and lightning each year across the state [64]. This high rate of thunderstorm activity corresponds to the high frequency rate of tornadoes. By understanding daily occurrence rates, we can obtain a more accurate representation of tornadoes for the climatology. In this section, I look at daily distributions of tornadoes in Florida. Figure 3.7 shows the daily tornado count distributions with respect to both yearly and monthly trends in tornado occurrence from 1950 through 2016. This ﬁgure portrays a new way of examining the daily tornado count throughout the state. The lighter shades of gray indicate a lower number of tornadoes per day,

27 657

158

100 61

16 17

10 8 Number of Days 7 5

2 2

1 1 1 1 1 1 1

0 5 10 15 20 Number of Tornadoes

Figure 3.8: Number of days with at least one tornado on a logarithmic scale. while the darker shades indicate a higher number of tornadoes per day. This figure reveals that throughout history, Florida tends to have very few days with many tornadoes occurring. While analyzing the daily tornado count for the most recent thirty years of data, figure 3.8 shows an exponential decrease in the number of days with at least one tornado occurring. In the period from 1987 to 2016, there were 657 days with an occurrence of only one tornado and 158 days with two tornadoes. There were only two days with a maximum of 22 tornadoes, which occurred on April 23, 1997, causing one casualty and June 23-24, 2012, causing two casualties. The series of tornadoes that occurred in June of 2012, were spawned in the outer rain bands from tropical cyclone Debbie. Most of these tornadoes were rated an EF0 on June 23 and affected the southwestern portion of Florida. However, on June 24, a series of tornadoes broke out in the southern and central Florida peninsula causing widespread damage with many of them being rated EF1 and EF2 [46]. In contrast, the day with the highest number of casualties attributed to tornadoes occurred on February 22-23, 1998. There was a series of eight tornadoes that were reported, which were linked to 301 casualties. Of these, 259 were injuries with the remaining 42 being fatalities. These

28 tornadoes were associated with a tornado outbreak that occurred in east-central Florida, when a super cell thunderstorm came ashore from the Gulf of Mexico [64]. The outbreak took place between 11 pm and 2:30 am EST, and three of the eight tornadoes were rated an F3 [15]. In addition, more than 3,000 structures were damaged, 800 homes destroyed, and more than 700 left uninhabitable [64]. The estimated amount of damage produced by the tornadoes is over 100 million dollars (1998 USD). This outbreak is considered one of the most deadly outbreaks in Florida’s history [15].

3.4 Hourly

Hourly distributions for Florida tornadoes are also plotted from 1987 to 2016. Figure 3.9 shows the frequency of tornadoes in local time. This type of ﬁgure has not been used to analyze hourly distributions of tornadoes but provides a more natural way of observing trends. The Apalachicola River separates the Central Standard time (CST) zone from the Eastern Standard time (EST) zone. It ﬂows from north to south through the panhandle of Florida, between Panama City (CST) and Tallahassee (EST). The local time of each tornado is calculated based on the start longitude and latitude location of the tornado touchdown, in its corresponding time zone. Since the time column in the original data set is in CST, which accounts for daylight savings, one hour was added to the time of the locations of tornadoes that initiated in the eastern time zone. There is a slightly left skewed distribution of the hourly frequency of tornadoes, which peaks in the afternoon at 4 pm local time with an occurrence rate of 183 tornadoes. Most tornadoes in Florida tend to occur during daylight hours from around 7 am to 7 pm, with majority of them occurring in the afternoon between 1 pm and 5 pm. During these times, the temperature is likely to be near its high for the day and sea breezes are most active, creating an environment that is conducive to thunderstorm formation [43]. Tornadoes in the southeastern part of the United States occur most frequently in the late afternoon and early evening, which is consistent with the timing of tornadoes in Florida. The peak in tornado frequency likely corresponds to the peak in convective processes, where the diurnal heating is maximized and thunderstorms develop. The tornadoes that occur during the day, account for the majority of casualties from tornadoes with a peak of 14 casualties occurring between the hours of 2 pm to 3 pm and 5 pm to 6 pm. In contrast, very few tornadoes occur during the night time, however it is not unlikely for them to take place. Tropical cyclone tornadoes often times occur at night, leaving research to suggest

29 Midnight 10 pm 2 am 150

100 8 pm 4 am

0 6 pm 6 am Number of Tornadoes 4 pm 8 am

2 pm 10 am Noon

Figure 3.9: Hourly distribution of Florida tornadoes from 1987 to 2016 that night time occurrence of tornadoes may be in association with tropical cyclones making landfall [22]. In the southeast, night time tornadoes are often deadly. They are hard to see and most of the time people are asleep [54]. In Florida, the most casualties caused from night time tornadoes occurred between 3 am to 4 am with nine casualties, followed by midnight to 1 am with eight casualties, and 11 pm to midnight with seven casualties. Hagemeyer (1997) isolated outbreak tornado events in his study of peninsular Florida tornadoes. While doing this, he found that most tornadoes occurred from 9 am EST through 2 pm EST for extratropical cases. However, when I factored in all tornado cases throughout the state, my peak hourly distribution varied. This indicates that the tornado distributions were skewed when factoring in non-tornado outbreak days. After analyzing the yearly, monthly, daily, and hourly trends of tornadoes in Florida, I have provided some insight on why an updated climatology of tornadoes is important. There are some limitations to my methods, which relate to the the lack of a reliable long term tornado record. The occurrence of tornadoes in the earlier years of the SPC tornado data set are uncertain because of

30 the inherent nature of the data itself. An example where this limitation may be apparent is while testing whether or not the frequency of tornadoes per year has been declining in Florida. In my methods, I split the data in to two diﬀerent ﬁfteen year periods (1987-2001 and 2002-2016) to test whether or not the number of tornadoes occurring each year is declining. This results in a small sample size for both periods, which limits the strength of inference when drawing conclusions. Comparing the temporal aspects of tornado frequency distributions, tropical cyclone tornadoes, casualties, and property exposure can help one better understand the tornado hazard throughout the state. Tornadoes in Florida can and do cause damage to property as well as many injuries and fatalities. It is important to fully understand the potential for loss of life and property from tornadoes that touchdown in the state. In the next chapter, I examine the geographic distribution of Florida tornadoes. I look the characteristics of tornado path length, width, and area. I also observe regional variations between the panhandle and peninsula regions of the state. Lastly, I hypothesize about tornadoes and their distance from the coast.

31 CHAPTER 4

CHARACTERISTICS AND REGIONAL VARIATIONS

In this chapter, I discuss the tornado characteristics in Florida by examining magnitude, path length, path width, and path area. I also look at geographical locations of where casualties from tornadoes have occurred in Florida. Because of diﬀerences in the climate between the northern and southern portions of the state, I analyze regional variations of Florida tornadoes. I split the state between the panhandle and peninsular region to observe similarities and diﬀerences between tornado frequency distributions. Lastly, I explain the science of tornadoes and hypothesize about the relationship between the strength of a tornado and its distance from the coast.

4.1 Tornado Characteristics

The variability in both distribution and frequency of tornadoes diﬀers across seasons and spatial patterns throughout the United States [6]. Tornado frequency is often based on the area of damage in the tornado path during the time of the event. The damage area in the path of a tornado tends to better account for the measure of incidence because of the high variability in the length, width, and area of the tornado path [55]. Tornado characteristics in Florida are diﬀerent compared to tornadoes throughout the United States. This section focuses on the magnitude of tornadoes in Florida with respect to both casualties and property values. I analyze the number of tornadoes versus the number of casualties throughout the state and provide maps of the locations of casualties based on the magnitude of the tornadoes that produced these casualties. I observe the structural property exposure of tornadoes by F/EF rating throughout the state. In addition, I analyze the path length, path width, and path area characteristics of tornadoes and compare them to the characteristics of tornadoes in the United States. I also test my hypothesis to determine whether or not casualties are higher in the panhandle than in the peninsula.

32 4.1.1 Magnitude

Most tornadoes that occur in Florida are very weak tornadoes that do not cause a lot of damage. Of all tornadoes that have occurred in Florida from 1987 through 2016, 78 percent of them are rated an F/EF0, followed by 18 percent being an F/EF1, and only about ﬁve percent being rated greater than or equal to an F/EF2. There were a total of ten reports of tornadoes rated an F/EF3, with no reports of F/EF4 or F/EF5 in the thirty years of data used (Table 4.1). Figure 4.1 shows the monthly distribution of tornadoes in Florida by EF rating as a percentage compared to all tornadoes. Notice, the scale of the y-axis is not ﬁxed across plots. The month of February has the highest chance of a tornado rated an F/EF3 or higher occurring, most likely due to the shift of temperatures from the winter to spring season in Florida. There have also been ratings of F/EF3 occurring in April, October, and November. Strong tornado occurrences in April are likely similar to those occurring in February. This is likely due to large temperature gradients across frontal systems the tornadoes spawn from, as well as fast high-altitude winds which provide the energy needed allowing the tornado to further intensify [43]. However, it is quite obvious that weak tornadoes (

33 Percentage of EF Rating Tornadoes to All Tornadoes EF0 EF1 2.5

10 2.0

1.5 Ratio 5 1.0

April May June July April May June July March August March August JanuaryFebruary October JanuaryFebruary October September NovemberDecember September NovemberDecember

EF2 EF3 0.8 0.4

0.6 0.3

0.4 0.2 Ratio 0.1 0.2 0.0

April May June July April May June July March August March August JanuaryFebruary October JanuaryFebruary October September NovemberDecember September NovemberDecember Month Month

Figure 4.1: Monthly distribution of EF Rating to all tornadoes by percentage on a variable y-scale

34 57

10 8

5 5

3 Number of Tornadoes

2 2 2 2 2

1 1 1 11 1 11 1 1 11 1 1 1 1

1 10 100 Number of Casualties

Figure 4.2: Number of tornadoes that produced at least one casualty in Florida from 1987 through 2016 mainly occur in the peninsular portion of the state, with a few occurring in the panhandle away from the coast. In addition, the highest casualty count occurs in the peninsular portion from EF3 tornadoes (> 100). These casualties are mainly centered around the Tampa and Orlando areas. The area between Tampa and Daytona Beach tends to be a hot spot for tornadoes with a high incident rate of injuries and deaths [64]. I hypothesized that when controlling for storm intensity, casualty rates are likely to be higher in the panhandle portion of the state than in the peninsula. The reason for this is because structures built in the peninsular portion of the state have more strict building codes and will withstand higher wind speeds due to a higher frequency of hurricanes or tropical storms making landfall. I calculated the number of casualties from tornadoes for each region and found that the peninsular portion of the state has a higher average number of casualties per tornado (9.1) than the panhandle (4.1). I then used a Pearson’s chi squared test of independence to determine whether or not there is a relationship between the number of casualties by F/EF category in the panhandle versus the peninsular region of the state. The results indicate that I can reject the null hypothesis (X2 (3,

35 EF0 EF1

EF2 EF3

Number of 50 100 150 Casualties

Figure 4.3: Genesis location of casualty producing tornadoes by EF Rating in Florida

36 n = 4) = 63.154, p-value < 1.245 x 10−13) and conclude that there is a statistically significant relationship found between the number of casualties and F/EF category in both the panhandle and peninsular portions of the state. Because the two variables are dependent on each other, the number of casualties is likely to be higher as the F/EF rating of a tornado increases. In addition to tornadoes causing casualties throughout the Florida, they can also cause large amounts of structural damage. According to Boruff et al. (2003), of all natural disasters in the United States, tornadoes ranked third in total dollar losses based on damage totals, right behind floods and hurricanes [5]. Like casualty rates, weaker tornadoes tend to cause less damage than more intense tornadoes. While EF0 tornadoes account for 59% of all Florida tornadoes they represent only 9% of the property exposure. In contrast, tornadoes rated EF2 or higher account for less than 12% of all tornadoes but 64% of residential and 69% of non-residential property exposure. Although weak tornadoes dominate the distribution, strong tornadoes cause a significantly larger amounts of damage when they touchdown. Tornadoes associated with tropical cyclones account for 21% of all residential and 17% of all non-residential property exposure.

4.1.2 Path Length and Width

Table 4.1: Tornado characteristics by F/EF rating

F/EF Mean Mean Mean Rating Counts Length (km) Width (m) Path Area (km2) Percent 0 1370 2 26 54 78 1 319 4 80 506 18 2 66 10 175 1962 4 3 10 23 306 6581 1

Table 4.1 shows the characteristics of tornadoes that occurred in Florida from 1987 through 2016. Past research has proven that as the magnitude (F/EF rating) of a tornado increases, so does the average path length and width [26, 7]. This is also true for tornadoes in Florida. The mean path length of tornadoes in Florida ranges from two kilometers to 23 kilometers, with the mean width ranging from 26 meters to 306 meters depending on its magnitude rating. Of all tornadoes that have touched down in the state, the average path length is 2.4 kilometers, and the average path width is 43.4 meters. Tornadoes in Florida as compared to the United States have smaller

37 6

2 Average Length (km) Average

1990 2000 2010 Year

Figure 4.4: Annual average path length of Florida tornadoes in kilometers average path lengths and path widths. Based on these statistics, it is safe to say that the path length and width are strongly driven by smaller, weaker tornadoes that touch down. Figure 4.4 shows the annual average path length of tornadoes in Florida for each year and figure 4.5 shows the annual average path width. The average path length and width of tornadoes, in recent years, has increased. When using the Pearson product-moment correlation and Spearman’s rho tests, both showed p-values that were statistically significant (p-value < 0.02 for path length and p-value < 0.001 for path width). This increase may be attributed to more accurate surveys, the use of technology such as Doppler Radar, and/or climate change. These findings are consistent with findings from Elsner et al. (2014), which observed slight increases in path length and width over time for tornadoes in the United States. However, future studies should be conducted to determine the direct cause of the increase. In comparison, the mean path area of tornadoes in Florida ranges from 54 km2 to 6,581 km2 with an average of 2276 km2, depending on its magnitude rating. The tornado path area also increases with EF rating, which is to be expected. Recent studies have observed an upward trend

38 90

60 Average Width (m) Average 30

1990 2000 2010 Year

Figure 4.5: Annual average path width of Florida tornadoes in meters in the path area of tornadoes in Florida since 1988 [27]. Figure 4.6 shows the approximate annual path area of tornadoes since 1987. As seen by the regression line in ﬁgure 4.6, there is a slight increasing trend in the path area of a tornado in Florida. This increase in path area corresponds to the increases in path length and path width as well.

4.2 Regional Variation

Tornado distributions in Florida vary geographically throughout the state. During the winter season, Florida is said to have two different climates with one being considered continental and the other being peninsular [64]. These different climates during the winter months affect the distribution and strength of tornadoes throughout the state. Tornadoes have found to be most common along the coast, from just north of Tampa to Fort Myers, from Tampa to Daytona Beach, in the western panhandle, and the southeastern portion of the state around the Miami and Fort Lauderdale region [43, 64]. For the first time ever, a county analysis of tornadoes is portrayed in a geographical map of the state of Florida. Figure 4.7 shows the monthly distribution of tornadoes

39 1000 ] 2

100

0.1

Approximate Total Tornado Path Area [km Path Tornado Total Approximate 0.01

1990 1995 2000 2005 2010 2015 Year

Figure 4.6: Annual average path area of Florida tornadoes in kilometers in Florida with respect to each county. It is important to point out that tornado distributions in most counties located in the peninsular portion of the state show fairly normal distributions with peak frequencies ranging from late spring into early fall. However, most counties in the panhandle portion of the state show peak frequencies during the winter months ranging from November to March. This signifies that there are clear regional differences in the tornado climatology for the state. These differences in climate between the panhandle and peninsular portions of Florida can lead to shifts in weather systems, alterations in thunderstorm patterns, and contrasting tornado frequency distributions for the two regions. While it is important to think about the state of Florida as a whole, there are some variations in the climatology of tornadoes between the panhandle and peninsula. This variation may be small, however it affects how meteorologists, climatologists, emergency managers, insurers, and other personnel forecast or respond to tornadoes in the state. Hagemeyer (1997) focused on looking only at tornadoes that occurred in the peninsular portion of Florida. He defined this as being the area at or south of 30 degrees latitude and east of 84 degrees west. For comparison in this study, the same

40 Nassau

0 1 2 3 4 Escambia Okaloosa Holmes Jackson Gadsden Leon Jefferson Madison Hamilton Columbia Baker Duval

0 1 2 3 4 5 0 2 4 6 0 2 4 6 0 1 2 3 0 1 2 3 4 0 1 2 3 0 3 6 9 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.00 0.25 0.50 0.75 1.00 Santa Rosa Walton Washington Calhoun Liberty Wakulla Taylor Lafayette Suwannee Union Bradford Clay

0 2 4 6 0 2 4 6 8 0 1 2 3 4 0 1 2 3 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 Bay Gulf Franklin Dixie Gilchrist Alachua St. Johns

0 2 4 6 0 2 4 6 0 1 2 3 0.0 2.5 5.0 7.5 0.0 2.5 5.0 7.5 0.00 0.25 0.50 0.75 1.000.00 0.25 0.50 0.75 1.00 10.0 12.5 Levy Marion Putnam Flagler

0 2 4 6 8 0 2 4 6 0 1 2 3 4 5 0 2 4 6 8 Citrus Sumter Lake Volusia

0 2 4 6 0 1 2 3 4 5 0 2 4 6 8 0.0 0.5 1.0 1.5 2.0 Pinellas Hernando Pasco Seminole Brevard Month

0 5 0 1 2 3 4 5 0 2 4 6 0 3 6 9 Month 10 15 0.0 0.5 1.0 1.5 2.0 January Hillsborough Polk Osceola Orange Indian River February March

April 0 3 6 9 0 5 0 2 4 6 0 1 2 3 10 15 20 0.0 0.5 1.0 1.5 2.0 May Manatee Hardee Okeechobee St. Lucie June July August 0 5 0 1 2 3 0 2 4 6 10 0.0 0.5 1.0 1.5 2.0 September Sarasota DeSoto Highlands Martin October November December 0 2 4 6 0 1 2 3 4 0 1 2 3 4 5 0 1 2 3 Charlotte Glades Hendry Palm Beach

0 2 4 6 0 1 2 3 4 0 5 0.0 2.5 5.0 7.5 10 15 10.0 Lee Collier Broward

0 5 0 3 6 9 0 5 10 15 10 15 Monroe Miami−Dade

0 2 4 6 0 5 10 Number of Tornadoes

Figure 4.7: Monthly distribution of Florida tornadoes by county from 1987 to 2016

41 latitude is used for the peninsular portion. Anything north of 30 degrees latitude is considered the panhandle. The 30 degree north latitude line also separates the peninsular and continental climate zones in the state. It runs almost parallel from St. Augustine to the mouth of the Suwannee River. This line is well-known to both climatologists and soil scientists. The soil temperature zones are divided into the thermic zone in the north and the hyperthermic zone in the south [64]. Because of the obvious temperature gradients in the atmosphere and geosphere, separating the panhandle from the peninsular region becomes increasingly more important for analyzing tornado occurrence rates.

4.2.1 Panhandle

High pressure systems that occur in northern Florida during the winter are often associated with cold fronts that cause heavy winds, thunderstorms, and a rapid decrease in temperatures [64]. Storms in the panhandle portion of the state are associated with fronts and squall lines, therefore these storms are likely to be the ones causing many of the tornadoes in the region. The panhandle receives the larger effect of the cold, stormy, winter weather compared to the peninsula. These tornadoes are often stronger than the tornadoes that touch down in the peninsula as well [43]. Brooks et al. (2003) analyzes the climatology of tornado days at any given location in the United States from 1980 through 1999 [9]. The primary focus of this research was to develop a climatology of the daily probability of a tornado occurring within the United States and to look at the potential threat at any point on the map. While looking at the date of maximum threat, they found that it peaks in April for the southeastern states. The Tallahassee region, on the other hand, has a slightly higher peak in November, than in April. This further indicates differences in the tornado climatology for Florida as compared to other southeastern states. Based off my regional and county distribution analysis in the panhandle, my frequency distributions do not align with the statistical characteristics that Brooks et al. (2003) postulated based off their probabilistic model. Since the statistical tornado distributions differ between the Florida panhandle and peninsula, I isolate each region to get a better understanding of their climatology and inherent differences. Tornadoes in the panhandle portion of Florida are hypothesized to be similar to tornadoes that occur along the Gulf Coast region [6]. Figure 4.8 shows the monthly distributions of tornadoes between the panhandle (top) and peninsula (bottom). Tornadoes in the panhandle portion of Florida account for about 23 percent of the overall number of tornadoes in the state. There is a

42 Panhandle Tornadoes 80 71

60 53

39 40 40 36 31 29 25 23 24 19 20 13

Number of Tornadoes 0

Jul Jan Feb Mar Apr May Jun Aug Sep Oct Nov Dec

Peninsula Tornadoes

250 230

200 167 165 152 150 115 113 103 100 82 91 70 46 50 28

Jul Jan Feb Mar Apr May Jun Aug Sep Oct Nov Dec

Figure 4.8: Monthly Tornado Distribution between the Panhandle and Peninsula trimodal distribution with a peak frequency in September, June, and March with 71, 40, and 39 tornadoes occurring, respectively. October, however has the second highest number of tornadoes with 53 occurrences. The frequency shown from June to November slightly mimics the monthly tropical cyclone distribution indicating that the North Atlantic tropical cyclone season drives the overall distribution of tornadoes in the panhandle. Strong tornadoes (> F/EF2) in the panhandle region have occurred in March, April, and December. The month of March leads the distribution with six occurrences, followed by April and December, each with four. I divided the winter and spring season from the summer and fall seasons to see if there was a large discrepancy between the strength of tornadoes between the seasons as observed in past research. The winter and spring months account for 71 percent of strong tornadoes in the panhandle. Seven percent of tornadoes in the panhandle from 1987 through 2016

43 are considered strong tornadoes, compared to only 3.5 percent in the peninsula being strong. To determine the signiﬁcance of these percentages, the Welch two-sample t-test was used. A p-value of < 0.02 was obtained, signifying a statistical signiﬁcance between strong tornadoes in the panhandle versus the peninsula. To conclude, past research is correct in stating that a tornado touching down in the panhandle is likely to be stronger than one occurring in the peninsula.

4.2.2 Peninsula

Thunderstorms that develop over Florida as a result of the convergence of air masses from the Gulf of Mexico and Atlantic ocean are fairly common during the summertime. These thunderstorms form squall lines, which produce many of the tornadoes that commonly spawn over the peninsular region of the state [64]. The monthly distribution of tornadoes in the peninsular portion of Florida, are shown on the bottom of figure 4.8. This figure shows a bell-shaped distribution with a peak number of occurrences taking place during the month of June. The peninsula of Florida accounts for just over 77 percent of tornadoes in the state, which is about three times higher than in the panhandle. The distribution of tornadoes in the peninsula is also quite similar to the distribution for the state as a whole. Therefore, it is safe to say that the overall frequency of tornadoes throughout the state of Florida is largely driven by occurrences in the peninsula. In contrast to the Florida panhandle, there have been a few studies that have analyzed the trends of tornadoes in the peninsula of Florida. Hagemeyer (1997) presented the first climatology of tornadoes for the peninsula of Florida by only looking at tornado outbreak events [40]. In his monthly analysis, he found that tornadoes spawned from extratropical systems peaked in March and April, while tornadoes from tropical systems caused outbreaks in June, as well as September through November. These results are to be expected and seem to be quite consistent with my analysis for the peninsular portion of the state. Hagemeyer (1997) also found a maximum frequency in June, with a minimum in November, where as the minimum in my study was in December. He concluded that most peninsular tornadoes occur from May to August during the peak thunderstorm season, with a high population of the tornadoes being weak. Compared to my analysis, most tornadoes occur between June and September, both of which are a month later than Hagemeyer (1997) concluded. Although Hagemeyer (1997) only took into account outbreak scenarios, his results are fairly consistent with my findings [40]. However, the maximum and minimum tornado distributions have shifted to be a month later than previously observed. This shift may be attributed to more

44 accurate tornado reports from recent technological advances in radar and recording mechanisms, or they may be attributed to climate change. Either way, these shifts should be made known for future research. In addition to Hagemeyer (1997), Brooks et al. (2003) also found some notable results about tornadoes in the Florida peninsula. While focusing on the climatology of tornado days from 1980 through 1999 in the United States, the authors indicated that Florida has the second highest frequency maximum threat of tornado days. The frequency maximum threat of tornado days is described in the literature by the mean number of days per year with at least one tornado touchdown. The peak value for the Florida maximum is around 1.5 tornado days per year [9]. The probability of a tornado occurring starts to increase in the middle of February and is greatest in Florida, as well as from Louisiana into southern Alabama. This is consistent with the fact that a higher rated tornado (≥ EF2) is more likely to occur during the month of February in Florida, than compared to any other month. While looking at the annual cycle, the authors also found that the southern United States has the highest tornado threat. This threat is also isolated to the Texas panhandle and the Florida peninsula, however they did conclude that the summer maximum of tornadoes in peninsular Florida is attributed to weak F0 and F1 tornadoes [9]. The authors also note that the date of maximum threat for the southeastern United States occurs in April, however the peninsular portion of Florida has a maximum threat occurring during the summer months. These findings are not surprising and remain consistent with my overall analysis for Florida. One last study to note, was recently conducted by Farney & Dixon (2015), where they compared spatial variables of tornadoes for the contingent United States between 1960 through 1989 and 1990 through 2011 [28]. They found that the Tampa region of Florida had an average of four tornado days annually, which was the greatest in the different regions in the United States. Their results also showed that during the last 22 years of data, the average annual number of tornado days increased by 1.86 tornado days in southwestern Florida. Central Florida was found to have the greatest risk of an annual tornado during the 1960 to 1989 period in conjunction with characteristic ’C-shaped’ high risk area which extends from the Midwest into the Plains, and through the southeast. Perhaps the most significant finding relating to Florida tornadoes from Farney & Dixon (2015) is that Central Florida, notably the Tampa region, reveals that the highest percentage of years with at least two tornado days occurring in both periods. Just over 83 percent of the years for the first period and

45 almost 91 percent of the years during the most recent period saw increases of at least two tornado days [28]. Although my analysis does not take into account tornado days, it is apparent that the peninsular portion of Florida is more susceptible to tornadoes as compared to the panhandle. These studies on the climatology of tornadoes, whether it be for Florida or the United States as a whole, are very helpful in determining areas of maximum tornado potential. All three studies analyze tornado climatologies using diﬀerent variables and methods. However, they all come to similar conclusions when looking at the statistics over Florida. The studies show a signiﬁcant tornado potential occurring over much of Florida. This further validates the need for a full tornado climatology.

4.2.3 Distance to Coast

To further understand the tornado characteristics across different regions in Florida, I look at the relationship of tornadoes as a function of distance to coast. I hypothesize that tornadoes are likely to be more damaging as they travel further away from the coastline. This idea comes from the science and physics of tornadoes. Strong tornadoes are more likely to occur further away from the coast because of the buoyancy of air. Thunderstorms thrive based on three ingredients: moisture, instability, and a lifting mechanism. Warm, moist air at or near the surface and cold, dry air aloft creates an unstable air mass conducive for thunderstorm formation. This produces an updraft, which feeds the into the convection of a thunderstorm. As warm air rises in a supercell thunderstorm, the shearing winds twist the updraft to create a rotating column of air, which is known as the mesosyclone. When the mesocyclone shifts vertically, a tornado is formed [44]. The strength of the mesocyclone is strongly dependent on the updraft speed, which is controlled by the buoyancy of the air. Coastal regions tend to have cooler air temperatures due to the difference in the rates of heating between the ocean and land. Cooler air is less buoyant, or more stable, which weakens the updraft in a thunderstorm. As thunderstorms move inland, the air feeding into the updraft becomes warmer. This allows the thunderstorm to intensify, further increasing the chance of generating a strong tornado. To determine each tornado’s distance from the coast, I consider only tornadoes whose track is completely over land. I use the coastline of Florida which is an ArcGIS shapefile that is ex- tracted from CloudMade data and derived from OpenStreetMap. It was made accessible through MapCruzin and is available for download online [48]. The distance to coast in units of meters is

46 computed as the cartesian length between the tornado track centroid and the coastline. To test my hypothesis, I ﬁrst look at the statistics of distance-to-coast tabled by EF rating. Table 4.2 shows the number of tornadoes, mean distance (meters), and median distance (meters) of tornadoes from the coast for each magnitude that occurred in Florida as a result of the distance to coast function. Tornadoes rated as an F/EF2 and F/EF3 were combined into one category because of the low sample size. Although there is a small change in distance from the coast compared to F/EF rating, my hypothesis is proven to be correct. Both the mean and median distance of tornadoes from the coast increases as the magnitude increases.

Table 4.2: Tornado characteristics as a function of distance to coast for the whole state of Florida F/EF Number of Mean Median Rating Tornadoes Distance (m) Distance (m) 0 1287 13908 4694 1 300 15205 5803 2+ 73 18454 7682

Because of the weaker relationship for the state of Florida as a whole, I decided to examine the distance to coast relationship for the panhandle of the state as a separate entity. The results of this relationship are shown in table 4.3. The relationship between mean and median distance to coast and the magnitude rating of a tornado is much more strongly correlated in the panhandle region of the state. There are likely two factors driving this relationship in the distance to coast analysis for the panhandle of Florida. The first one is due to the inherent differences of buoyancy between the cooler coastal region and the warmer inland area. The second factor is the latitudinal location of Florida’s panhandle, which places itself closer to the polar jet streak. Since the polar jet streak promotes a favorable shear profile for tornado genesis, the Florida panhandle would have a higher chance of stronger tornadoes as well. Thus, the coastal buoyancy gradient and latitudinal location factors likely explain this distance to coast relationship for the Florida panhandle.

Table 4.3: Tornado characteristics as a function of distance to coast for the Panhandle of Florida F/EF Number of Mean Median Rating Tornadoes Distance (m) Distance (m) 0 252 20758 8433 1 96 23863 19617 2+ 26 31337 29590

47 In conclusion, the two different climates between the panhandle and peninsular regions in Florida contribute to the variation in tornado characteristics and distributions. There is no significant difference between casualty rates from tornadoes rated higher than F/EF2 in the panhandle versus the peninsula. However, not very many strong tornadoes (> F/EF2) have occurred in Florida, therefore, the sample size when conducting the casualty rate per F/EF rating test is extremely small for stronger tornadoes which results in a larger variance. The limitation of rating tornadoes is important to note in this section because magnitude ratings of tornadoes may not be consistent throughout the data set. This is due to the change between using the original F scale in the early years of the data and using the updated EF scale starting in 2007. In addition, there may be an inherent bias in the EF rating system when conducting surveys. The tornado occurrence rate in the panhandle region of the state is much lower than in the peninsula. Although the sample size is small when comparing strong tornadoes in the panhandle versus the peninsula, tornadoes that touch down in the panhandle have a higher chance of being stronger than those in the peninsula. In the next chapter, I discuss property exposure and losses from tornadoes in Florida. I examine structural values throughout the state by looking at residential and non-residential values. I overlay the tornadoes onto the property value grid cells to understand where the greatest property exposure values are in the state. I also create a Monte Carlo simulation of tornado events to further examine property loss estimates.

48 CHAPTER 5

STRUCTURAL PROPERTY LOSSES

This chapter is an extension of the paper [27] I helped research and write with my major professor. Here, I examine property values across the state of Florida. I evaluate the amount of property exposed to tornadoes by overlaying past events onto the property value grid cells. I conduct a case study to determine how much property is exposed for the worse case scenario tornado event. Additionally, I use a Monte Carlo simulation to model property loss from tornadoes in Florida, which is used to obtain actual and probable maximum loss estimates.

5.1 Property Value Analysis

The state of Florida has many expensive structures located along the coastal regions in the state. This is largely due to the areas having a high population density, many businesses, and providing easy access for the imports of goods. Over the past two decades, Florida’s population has been steadily increasing and the rate of growth nearly doubled the national average since 1990 [60]. Therefore, it is important to further understand the tornado exposure to structure values around the state of Florida. In this section, I analyze the property value dataset with respect to the overall property value distribution, as well as the residential and non-residential structural values. I plot histograms of the property values and show maps of the one-kilometer grid cells of importance. In addition, I brieﬂy describe the tornado overlay analysis as well as a case study for future reference.

5.1.1 Structural Values

The structural values of interest are located in the residential and non-residential structural values in each grid cell. These values are calculated by subtracting the land value from the assessed value and aggregating them to the one-kilometer cells. If the land value is considered to be unrea- sonable (e.g., it is zero or it is greater than the assessed value), the structural value is 90% of the assessed value and the total structural value is based on all land use categories. The residential

49 Florida Structural Values 25000

20000

15000

10000

5000 Number of One Kilometer Grid Cells

$1 $100 $10,000 $1,000,000 $100,000,000

Residential Non Residential

Figure 5.1: Residential and non-residential structural value distribution based on one kilometer grid cells in Florida structural value is based on the residential land use category only, while the non-residential value is the diﬀerence between the total and the residential value. The total statewide structural value of all property in Florida is $942 Billion. At the per cell level, the average of all structural property values is $10 Million with a median of $484,881. Since the mean is signiﬁcantly greater than the median, there is a strongly right skewed distribution of property values. Figure 5.1 shows the statewide distribution of residential and non-residential structural values in Florida. On average, residential property values exceed non-residential values. The most common residential exposures at the cell level are between $100,000 and $1,000,000 while the most common non-residential exposures are between $50,000 and $500,000. The total residential structural value is $619 Billion with a mean of $6.8 Million and a median of $184,876. The cell with the highest residential structural value ($410 M) is located in downtown

50 Figure 5.2: One-kilometer cell with the largest residential structural value

Orlando and includes the area around Lake Eola Park (ﬁgure 5.2). In comparison, the total non- residential structural value is $323 Billion with a mean of $3.5 Million and a median of $87,495. The cell with the highest non-residential structural value ($8.7 B) includes Allen D. Nease Senior High School in Ponte Vedra, in St. John’s County (ﬁgure 5.3). The percentage of cells in the state with values exceeding $50 M is 3.3% for residential values and 1.2% for non-residential values.

5.1.2 Tornado Overlay and Exposure

For the tornado overlay analysis, all tornadoes that occurred in Florida from the original data set are used. Since the path length of a tornado is related to its path width, tornado tracks are calculated using the length and width columns in the data set. There were 3,278 tornadoes that occurred in Florida from 1950 through 2016 and their paths are shown in ﬁgure 5.4. These tornado paths are then overlaid onto the structural value grid cells to calculate the amount of property that is exposed to each individual tornado, assuming that the tornado occurred in 2014. Figure 5.5 shows the 1956 Pembroke Pines F3 tornado overlaid on the property value grid cells, which caused 20 casualties. Statistics of the tornado paths are then calculated at the per cell level. It was found that four percent of tornado paths have an area that exceeds one square kilometer. The total area of all

51 Figure 5.3: One-kilometer cell with the largest non-residential structural value paths with genesis locations in Florida is 733 km2, which is .5% of the total area of the state. The total number of cells aﬀected by tornadoes is 11,181 with 8,275 of the cells having been hit only once and 809 cells having been hit twice. The average number of tornadoes per cell with property is .12 with a minimum of zero and a maximum of 20. The cell with the greatest tornado occurrence rate is located in the in the Palmetto Park area of St. Petersburg and is hit about once every 3 or 4 years on average (ﬁgure 5.6). The annual average total structural value exposure is $158.9 Million with $96.6 Million being associated with residential values and the remaining $62.3 being associated with non-residential values. As stated previously, the month of February has the highest average exposure, followed by March and April. This is most likely attributed to stronger weather systems that spawn longer, wider tornadoes during the shift in seasons.

5.1.3 Case Study

After analyzing the results of property exposure to tornadoes, I was particularly interested in the worst case scenario. The tornado with the highest exposure rate to touchdown in Florida was an F4 tornado which occurred on April 4, 1966. This tornado was the ﬁrst of two tornadoes to

52 Figure 5.4: Tornado paths in Florida from 1950 through 2016

53 Figure 5.5: 1956 F3 Pembroke Pines to Fort Lauderdale tornado

Figure 5.6: Structural value cell with the greatest tornado occurrence

54 Figure 5.7: April 4, 1966 Tampa tornado path touch down and that are associated with the Tampa Tornado Family [64, 43]. The other tornado was rated an F3. Figure 5.7 shows the start and end location of the F4 tornado that touched down. These tornadoes spawned from a squall line that moved over the state. Both tornadoes origi- nated as waterspouts over the Gulf of Mexico that later moved onto land [64]. The ﬁrst tornado touched down around 8 am EST in the Gulf of Mexico near Clearwater, Florida in Pinellas County. It traveled northeast across the peninsula, ending near Cape Canaveral by the Atlantic Ocean [43]. The tornadoes damaged hundreds of homes, a junior high school, and dormitories on the Univer- sity of South Florida’s campus. Both tornadoes are listed as being continuous events, however the damage paths did not cross the entire state, and some damage is shown to be caused by downbursts within the system. Nevertheless, the destruction between the tornadoes and the downbursts was continuous throughout central Florida. There were 541 casualties as a result of the tornadoes, which is the highest of any tornado occurring throughout the 67 year period. Losses from the tornadoes were estimated between $5 Million to $50 Million 1966 USD, which is between $38 Million and $363 Million 2014 USD [64]. In this study, the total average structural value exposure from the F4 tornado was found to be $875 Million dollars with $513 Million being residential exposure and the remaining $362 Million being non-residential exposure. By analyzing the potential property exposure from tornadoes in

55 the state, insurance companies are able to have a better perspective of when and where to allocate money. In addition, other residents are able to better understand the tornado hazard and risk of casualties or structural damages in the state.

5.2 Modeling Property Loss

Each year, tornadoes cause property loss in Florida, which is widely dependent on the location, strength, and size of the tornado. As Florida’s population continues to increase, more structures are being built, creating a larger property exposure to tornadoes. In this section, I discuss the methods and results of property losses from tornadoes in Florida. It is important to note that the amount of property exposed to a tornado is an estimate of the potential damage that could occur if the event were to take place in 2014. Actual losses from a tornado event may be less because not every structure in the tornado’s path is damaged (see table 1.1 on page 2). For this reason, the actual and future property losses from tornadoes were analyzed and modeled using a Monte Carlo simulation.

5.2.1 Monte Carlo Simulation

Monte Carlo simulation is a computational modeling technique that calculates results over and over, each time using a different set of random values specified. It then produces distributions of possible outcome values and probabilistic solutions. It is a unique technique because it includes iteration and randomness when computing. In addition, it provides more information about the best case and worse case scenarios, which is what insurance companies are particularly interested in [58]. In my research, the Monte Carlo simulation is used for modeling tornado occurrences in Florida. The Monte Carlo algorithm I use is similar to that used in Strader et al. (2016) and Meyer et al. (2002) [58, 49]. It uses the path characteristics, locations, and orientations from the tornadoes in the original data set to generate potential future paths. The simulation is run over a period of 10,000 years to obtain an accurate representation by testing a significant portion of outcomes per grid cell. If provided sufficient data for a given grid cell, it has the capability to take into account the long-term patterns and variability of tornadoes throughout the state. The sample is drawn from a random Poisson distribution which models the number of events occurring within a specific

56 Figure 5.8: Monte Carlo simulated tornado paths time interval and the results must be greater than zero. The rate used is equal to the long-term annual rate of tornado occurrence in Florida, which is 49 tornadoes per year for the period from 1950 through 2016. It also uses a non-parametric re-sampling method to sample the angles, genesis locations, and other path characteristics of tornadoes. This means it will never generate a tornado in an area that has not previously had a tornado. The resulting synthetic paths are shown in ﬁgure 5.8. These paths are then overlaid onto the property exposure cells, and exposure values are determined by multiplying the corresponding regression coeﬃcients to obtain the loss estimates.

57 $100,000,000

$1,000,000 Losses

$10,000

$100 $10,000 $1,000,000 $100,000,000 Total Structural Property Value

Figure 5.9: Loss versus exposure from 2007 through 2015

5.2.2 Actual Loss Estimates

To obtain actual loss estimates, the loss column in the SPC tornado data set, which are available since 2007, is used along with the calculated exposures. The logarithm of the per tornado loss estimates are regressed onto the logarithm of exposure, while controlling for year. This determines the percentage change in losses for a percentage change in exposure. The average relationship can be used to estimate the amount of losses from the exposure. I create a scatter plot to show the relationship between actual loss and exposure, which is shown in ﬁgure 5.9. It turns out that there is a relationship between losses and structural property value. A structure can have a large exposure with little loss, which would likely occur from a weaker tornado. The amount of losses may exceed total structural value, which is due to non-structural type losses such as a vehicle or personal belongings. While controlling for the changes over time, these results show that a doubling of the residential exposure increases actual recorded losses by 26% since 2007 and a doubling of non-residential exposure by 21%. The average annual loss of property from tornadoes in Florida was found to be valued at $53 million. To put this in perspective, the

58 0.75

0.50 density

0.25

0.00

$1,000,000 $100,000,000 Annual Florida Property Loss from Tornadoes [2014 Dollars]

Figure 5.10: Average annual property loss from Florida tornadoes amount of direct spending in Tallahassee, Florida by visitors during the six games in the 2014 home football season was about $48.8 million dollars [14]. With population and property values rising throughout the state, the annual expected loss becomes increasingly important for homeowners and insurance companies to evaluate and assess the potential risk. As a result of the Monte Carlo simulation, I create a probability density function of the average annual loss of property values from tornadoes in Florida, which is the expected amount of loss per year. The results of the probability density function are shown in ﬁgure 5.10. The red line indicates the $53 million average annual loss line that was previously calculated using the loss estimates from the actual tornadoes. By looking at the annual average losses compared to the density distribution of property loss from the Monte Carlo simulation, the state of Florida has been rather unlucky with the amount of property damaged from tornadoes. Insurance companies, on the other hand, are not particularly interested in knowing the actual loss. Instead, they want to know what happens under the right tail of the graph.

59 5.2.3 Probable Maximum Loss Estimates

Although the annual average loss is important to know and understand, insurance companies are more interested in the largest loss that could occur. In this section, I estimate the annual probable maximum loss (PML), as the largest loss that could occur from tornadoes. These estimates are for annual expected losses. This helps insurance underwriting decisions which include the amount to cede to a re-insurer. Figure 5.11 shows the annual exceedance probability of property loss from tornadoes in Florida. In other words, it shows what is under the right tail of figure 5.10, just with the axes flipped. These results conclude that there is a 5% chance that the annual loss will exceed $203 million, a 1% chance that the it will exceed $430 million, and a 0.1% chance that it will exceed $1 billion. For comparison, in 2017, there were 16 weather and climate disaster with losses at or exceeding $1 billion each across the United States [50]. Of these, three were attributed to tornado outbreak events occurring in the southeastern United States (January), central and southeast (March), and Midwest (March) resulting in losses of $1.1 billion, $1.8 billion and $2.2 billion, respectively. These findings should be of wide interest to the property insurance and reinsurance industries for helping the gauge the risk of losses and prioritize management actions. While the Monte Carlo simulation has been verified as a strong statistical technique in simulating tornado characteristics [49], its techniques rely on the specified distribution given in the model. The assumed Poisson distribution is a good approximation for tornado characteristics, however it does not take into account changes in the distribution that may occur as a result of a changing climate. Thus, the Monte Carlo simulation may slightly underestimate extreme tornado events given their low frequency within the tails of the Poisson distribution. Additionally, the long-term variability in the characteristics of tornadoes is also not accounted for because the model is designed to resemble the raw data in which it is drawing its samples from. Another limitation in the methodology lies in the dollar value inherent in the property value data itself. The dollar values used to estimate property exposures and losses are in 2014 USD. The overall values are likely to increase over time due to inflation and rises in population, therefore leading to possible changes in the amount of losses in the future.

60 $5,000,000,000

$2,000,000,000

$1,000,000,000

$500,000,000 Property Loss

$200,000,000

$100,000,000

2.0% 1.5% 1.0% 0.5% 0.0% Annual Exceedance Probability

Figure 5.11: Annual exceedance probability of property loss from Florida tornadoes as modeled in the Monte Carlo simulation

61 CHAPTER 6

MAJOR FINDINGS, CONCLUSIONS, LIMITATIONS, AND FUTURE WORK

This chapter lists the major ﬁndings of this study, provides some conclusions, and discusses possible future work.

6.1 List of Major Findings

The major ﬁndings of this thesis are:

• The overall temporal variation of tornadoes in Florida is driven by the tropical cyclone season, resulting in a large number of weak tornadoes occurring throughout the state during the months of June through November.

• The seasonal distribution of tornadoes occurring in the peninsula of Florida peaks in June, while the seasonal distribution for the panhandle peaks in September indicating there are clear regional diﬀerences in tornado frequencies.

• The number of casualties from tornadoes in Florida peaks in the month of February and has a strong positive correlation with the amount of property exposed.

• The frequency of tornadoes has been on the decline in recent years, however the average path length, path width, and path area of tornadoes has increased.

• The relationship between mean and median distance to coast and the magnitude rating of a tornado is positively correlated with one another.

• Results of the Monte Carlo simulation indicate a 5% chance that the annual loss will exceed $203 million, a 1% chance that the it will exceed $430 million, and a 0.1% chance that it will exceed $1 billion.

6.2 Conclusions and Limitations

There are a variety of ﬁgures used throughout this research that provide new ways of analyzing tornado distributions both temporally and spatially. Many of these ﬁgures provide a more natural

62 way for examining tornado trends. On average, there are about 59 tornadoes that occur in Florida each year, with a declining trend in the average frequency. The monthly frequency distribution of tornadoes mirrors the Atlantic hurricane season, with peak occurrences taking place in June and September. In contrast, peak casualty and property exposure rates occur in February and positively correlate with one another. This is likely due to the weather shifts from winter into spring months that cause larger temperature gradients, which provides a more conducive environment for supercell thunderstorms to spawn strong tornadoes. There are very few days in Florida that have a large number of tornadoes occurring. Tornadoes tend to spawn during daylight hours, with most of them occurring in the afternoon when heating by the sun is the greatest. Although many tornadoes that spawn in Florida are weak, strong tornadoes have occurred in the past. These tornadoes can cause a large number of casualties and high amounts of structural damage. Areas with the highest casualty rate are located in the peninsula near the Tampa Bay and Orlando regions. Regions with the highest structural values are located along the coast as well as large cities such as Tampa Bay, Orlando, Fort Lauderdale, and Miami. The average path length, path width, and path area of tornadoes each year in Florida has shown an increasing trend over the thirty years of data. These increases are consistent with trends studied in past research. Variations in tornado frequency between the panhandle and peninsular regions in Florida diﬀer. The location in variation corresponds to the diﬀerent climate regions and soil types attributed to the area around 30 degrees latitude. Tornadoes in the panhandle are often times spawned from cold fronts and squall lines, which typically result in stronger tornadoes as compared to the peninsula. The frequency of tornadoes in the panhandle peaks in September, however the panhandle accounts for less than one-fourth of tornadoes that occur throughout the state. Tornadoes in the peninsular portion of the state peak during the month of June and strongly drive the overall distribution in Florida. According to my distance to coast hypothesis, tornadoes strengthen as they move further inland. This hypothesis is more apparent in the panhandle region of the state as compared to the state as a whole. The total structural value for all property in Florida is $942 Billion with residential values exceeding non-residential values. The annual average total structural value exposure is $153.9 Million with the highest exposure rates occurring in February, March, and April. The tornado with the highest exposure rate touched down in the Tampa Bay region and had a total exposure valued

63 at $875 Million. The resulting actual recorded property loss estimates from tornadoes showed a doubling of residential exposure by 26% and a doubling of non-residential exposure by 21%. The average annual loss was valued at $53 Million. Using the Monte Carlo simulation, potential future tornado paths were obtained. The results indicate a 5% chance that the annual loss will exceed $203 million, a 1% chance that the it will exceed $430 million, and a 0.1% chance that it will exceed $1 billion. One major limitation of the results presented in this thesis is the length of the data record used. Although I used the longest record over which the data sets are considered reliable, the thirty-year period is short given the infrequent nature of tornadoes at any one location. Thirty-year periods are commonly used for temperature and precipitation climatologies, but they may not fully represent the range of variation in extreme weather events like tornadoes. This creates a small sample size when testing hypotheses, which limits the strength of inference. There are also multi-decadal oscillations in synoptic scale weather patterns, therefore using only a thirty year time period may not fully capture the natural variability in tornado frequency, leading to another limitation associated with drawing too ﬁrm of conclusions. There is a possible inherent bias in the reported magnitude rating of the tornadoes, which is caused from diﬀerences between how tornadoes were rated using the F scale and EF scale. Finally, tornadoes sometimes go unreported if they occur in sparsely populated regions, leaving the actual frequency of tornadoes in Florida somewhat unknown. Lastly, the Monte Carlo model uses a non-parametric approach, so the characteristics of the tornado paths are drawn from the historical record and re-sampled. A place that has never been hit by a tornado, will not be hit by one in the model, although the tornado could theoretically pass over a point that was not previously hit. This can be problematic because a tornado can statistically occur anywhere throughout the state. In addition, the sampling was done with out regard to the correlation between the path length and path width of tornadoes.

6.3 Future Work

What this research means for the future is that residents in Florida should be more aware and alert when under a tornado watch or warning. Local, county, and state emergency managers should be more conscious when a tornado occurs to help prevent injuries and deaths. The state of Florida

64 does not currently have a dedicated state siren or warning system to alert residents. Instead, the State of Florida uses NOAA weather radio and Wireless Emergency Alerts through the Federal Government to alert residents of tornado warnings issued by the NWS. Having a siren system is considered a local issue. This means that the cost, operation, and maintenance is completely handled by the county or city government. Some cities in Florida such as Oviedo, and Winter Park do have a tornado siren system. Universities, such as Florida State University, has an all-hazards siren system to alert faculty and students, and the Florida State Capitol complex has its own siren system as well. Therefore, more research on population exposure and risk may be beneficial to residents of the state. It may also provide emergency management officials with the necessary research needed to further implement a state tornado siren system that may help save lives, while also saving money in the future as well. The work presented here is a start for more in-depth research in the future. There are many ways of expanding on this research. One way is to study the climatology of tornadoes by looking at tornado days, outbreak events, or clustering effects. This may provide more information about how tornadoes have been changing with alterations in the climate. In addition, research on tropical cyclone tornadoes and the effects can further be expanded upon. By analyzing tornado characteristics and frequencies with respect to oscillations between El Niñoand La Niñaevents, one can further expand on the climatology of Florida tornadoes. Potential influences of climate factors can alter weather patterns around the globe, which can have an effect on the path length and width, strength, intensity, and rate of occurrence of tornadoes. Given the past temporal span of the dataset, it is currently not well understood about how the tornado climatology in the United States is affected by multi-decadal weather patterns, such as the Atlantic Multidecadal Oscillations (AMO) and the Pacific Decadal Oscillations (PDO). Now, future research can begin analyzing tornado frequency and intensity based on the positive and negative phases of the multi-decadal weather patterns. Additionally, this work can be adopted by other states around the nation that face similar problems with tornadoes, casualties, and property loss.

65 BIBLIOGRAPHY

[1] Study: Florida tornadoes the most deadly in the nation. http://www.bizjournals.com/tampabay/blog/morning-edition/2014/06/study-ﬂorida- tornadoes-the-most-deadly-in-the.html, June 2014.

[2] Ernest M. Agee and Alyssa Hendricks. An assessment of the climatology of Florida hurricane- induced tornadoes (HITs): Technology versus meteorology. Journal of Climate, 24(20):5218– 5222, 2011.

[3] Donald C. Ahrens. Meteorology Today. Cengage Learning, 2012.

[4] G. Berz, W. Kron, T. Loster, E. Rauch, J. Schimetschek, J. Schmieder, A. Siebert, A. Smolka, and A. Wirtz. World map of natural hazards a global view of the distribution and intensity of signiﬁcant exposures. Natural hazards, 23:443–465, 2001.

[5] Bryan J. Boruﬀ, Jaime A. Easoz, Steve D. Jones, Heather R. Landry, Jamie D. Mitchem, and Susan L. Cutter. Tornado hazards in the United States. Climate Research, 24:103–117, 2003.

[6] John B. Bounds and Peter T. Soul´e. Aspects of tornado climatology in Alabama and Floridas panhandle, 1962-1986. Southeastern Geographer, 33(2):172–193, 1993.

[7] Harold E. Brooks. On the relationship of tornado path length and width to intensity. Weather and Forecasting, 19:310–319, 2004.

[8] Harold E. Brooks and Charles A. Doswell III. Normalized damage from major tornadoes in the United States: 1890–1999. Weather and Forecasting, 16:168–176, 2001.

[9] Harold E. Brooks, Charles A. Doswell III, and Michael P. Kay. Climatological estimates of local daily tornado probability for the United States. WF, 18:626–640, 2003.

[10] United States Census Bureau. Quick facts. https://www.census.gov/quickfacts/fact/table /NY,FL/PST045217, 2017.

[11] Donald W. Burgess, Michael A. Magsig, Joshua Wurman, David C. Dowell, and Yvette Richardson. Radar observations of the 3 May 1999 Oklahoma City tornado. Weather and Forecasting, 17(3):456–471, 2002.

[12] National Hurricane Center. Tropical cyclone climatology. https://www.nhc.noaa.gov/climo/.

[13] Stanley A. Changnon. Tornado losses in the United States. Natural Hazards Review, 10(4):145– 150, 2009.

66 [14] University Communications. Latest facts and ﬁgures about Florida State Universitys economic impact on Tallahassee and Leon County. http://economic-impact.fsu.edu/, Jan 2018.

[15] William M. Daley, D. James Baker, and John J. Kelly Jr. Central Florida tornado outbreak. https://www.weather.gov/media/publications/assessments/cntrlﬂ.pdf, June 1998.

[16] John Van de Lindt, Mohammad Omar Amini, Christine Standohar-Alfano, and Thang Dao. Systematic study of the failure of a light-frame wood roof in a tornado. Buildings, 2(4):519–533, 2012.

[17] Charles A. Doswell. Societal impacts of severe thunderstorms and tornadoes: Lessons learned and implications for Europe. Atmospheric Research, 67-68:135–152, 2003.

[18] Charles A. Doswell, Harald E. Brooks, and Nikolai Dotzek. On the implementation of the enhanced Fujita scale in the USA. Atmospheric Research, 93:554–563, 2009.

[19] Nikolai Dotzek. Tornadoes in Germany. Atmospheric Research, 56(1-4):233–251, 2001.

[20] Nikolai Dotzek. An updated estimate of tornado occurrence in Europe. Atmospheric Research, 67-68:153–161, 2003.

[21] Roger Edwards. Tropical cyclone tornado records for the modernized NWS era. 25th Con- ference on Severe Local Storms, Denver, CO, American Meteorological Society, page P3.1, 2010.

[22] Roger Edwards. Tropical cyclone tornadoes: A review of knowledge in research and prediction. Electronic J Severe Storms Meteor, 7(6):1–61, 2012.

[23] Roger Edwards. The online tornado FAQ. http://www.spc.noaa.gov/faq/tornado/, May 2017.

[24] Roger Edwards, James G. LaDue, John T. Ferree, Kevin Scharfenberg, Chris Maier, and William L. Coulbourne. Tornado intensity estimation: Past, present, and future. BAMS, 94:641–653, 2013.

[25] James B. Elsner, Thomas H. Jagger, and Ian J. Elsner. Tornado intensity estimated from damage path dimensions. PLoS ONE, 9 (9):e107571, 2014.

[26] James B. Elsner, Richard J. Murnane, Thomas H. Jagger, and Holly M. Widen. A spatial point process model for violent tornado occurrence in the U.S. great plains. Mathematical Geosciences, 45:667–679, 2013.

[27] James B. Elsner, Emily Ryan, and Georgianna Strode. Structural property losses from tornadoes in ﬂorida. Weather, Climate, and Society, 10:253–258, 2018.

67 [28] Tory J. Farney and Grady P. Dixon. Variability of tornado climatology across the continental United States. 3006(November 2014):2993–3006, 2015.

[29] Bernold Feuerstein, Nikolai Dotzek, and J´’urgen Grieser. Assessing a tornado climatology from global tornado intensity distributions. JCLI, 18:585–596, 2005.

[30] Christopher M. Fuhrmann, Charles E. Konrad II, Margaret M. Kovach, Jordan T. McLeod, William G. Schmitz, and Grady P. Dixon. Ranking of tornado outbreaks across the United States and their climatological characteristics. Weather and Forecasting, 29:684–701, 2014.

[31] Theodore T. Fujita. Proposed Characterization of Tornadoes and Hurricanes by Area and In- tensity. Chicago. University. Satellite and Mesometeorology Research Project. SMRP research paper. University of Chicago, 1971.

[32] Theodore T. Fujita. Tornadoes around the world. Weatherwise, 26(2):56–83, 1973.

[33] John P. Gagan, Alan Gerard, and John Gordon. A historical and statistical comparison of “Tornado Alley” to “Dixie Alley”. National Weather Digest, 34:145–155, 2010.

[34] Miquel Gay`a.Tornadoes and severe storms in Spain. Atmospheric Research, 100(4):334–343, 2011.

[35] R. Cecil Gentry. Genesis of tornadoes associated with hurricanes. Monthly Weather Review, 111:1793–1805, 1983.

[36] Dario B. Giaiotti, Mauro Giovannoni, Arturo Pucillo, and Fulvio Stel. The climatology of tornadoes and waterspouts in Italy. Atmospheric Research, 83(2-4 SPEC. ISS.):534–541, 2007.

[37] A. M. Goliger and R. V. Milford. A review of worldwide occurrence of tornadoes. Journal of Wind Engineering and Industrial Aerodynamics, 74-76:111–121, 1998.

[38] Basil Gomez and John Paul Jones II, editors. Research Methods in Geography. Wiley-Blackwell, 2010. 279-314.

[39] Richard W. Gray. The tornado of April 5, 1925, near Miami, Fla. Monthly Weather Review, 53(4):145–146, 1925.

[40] Bartlett C. Hagemeyer. Peninsular Florida tornado outbreaks. Weather and Forecasting, 12:399–427, 1997.

[41] Bartlett C. Hagemeyer and Gary K. Schmocker. Characteristics of east-central Florida tornado environments. Weather and Forecasting, 6:499–514, 1991.

68 [42] James I. Hendricks, Judith A. Curry, and Carlos D. Hoyos. Variability in tornado frequency associated with U.S. landfalling tropical cyclones. Geophysical Research Letters, 36(17):1–5, 2009.

[43] James A. Henry, Kenneth M. Portier, and Jan Coyne. The Climate and Weather of Florida. Pineapple Press, Inc, 1994. 141-167.

[44] Charles A. Doswell III and Donald W. Burgess. Tornadoes and tornadic storms: A review of conceptual models. Geophysical Monograph Series, pages 161–172, 1993.

[45] Kevin E. Kelleher, Kelvin K. Droegemeier, Jason J. Levit, Carl Sinclair, David E. Jahn, Scott D. Hill, Lora Mueller, Grant Qualley, Tim D. Crum, Steven D. Smith, S. Lakshmivarahan Stephen A. Del Greco, Linda Miller, Mohan Ramamurthy, Ben Domenico, and David W. Fulker. Project craft: A real-time delivery system for NEXRAD level II data via the internet. Bulletin of the American Meteorological Society, 88(7):1045–1057, 2007.

[46] Todd B. Kimberlain. Tropical cyclone report: Tropical Cyclone Debby. https://www.nhc.noaa.gov/data/tcr/AL042012 Debby.pdf, January 7 2013.

[47] Cameron C. Lee. Utilizing synoptic climatological methods to assess the impacts of climate change on future tornado-favorable environments. Natural Hazards, 62(2):325–343, 2012.

[48] Michael Meuser. Florida coastline. http://www.mapcruzin.com/free-united-states- shapefiles/free-florida-arcgis-maps-shapefiles.htm.

[49] Cathryn L. Meyer, Harold E. Brooks, and Michael P. Kay. A hazard model for tornado occurrence in the United States. American Meteorological Society, 16th Conf. on Probability and Statistics, pages J88–J95, Orlando, FL, 2002.

[50] NCEI. U.S. billion-dollar weather and climate disasters. https://www.ncdc.noaa.gov/billions/, 2017.

[51] NCEI. U.S. tornado climatology. https://www.ncdc.noaa.gov/climate-information/extreme- events/us-tornado-climatology, 2018.

[52] NOAA and National Weather Service. The May 31, 2013 El Reno, OK tornado. https://www.weather.gov/oun/events-20130531-elreno, Oct 2016.

[53] Sean Potter. Fine-tuning Fujita. Weatherwise, 60:64–71, 2007.

[54] Southeast Regional Climate Center (SERCC). Tornado climatology of the Southeast U.S. https://www.sercc.com/tornado climatology, 2007.

[55] Kevin M. Simmons and Daniel Sutter. Economic and Societal Impacts of Tornadoes. American Meteorological Society, Boston, 2011. 282.

69 [56] Michalis V. Sioutas. A tornado and waterspout climatology for Greece. Atmospheric Research, 100(4):344–356, 2011.

[57] American Meteorological Society. Glossary of Meteorology. American Meteorological Society, 1959. 585.

[58] Stephen M. Strader, Thomas J. Pingel, and Walker S. Ashley. A Monte Carlo model for estimating tornado impacts. Meteorological Applications, 23(2):269–281, 2016.

[59] Stephen M. Strader, Walker S.Ashley, Thomas J.Pingel, and Andrew J. Krmenec. Projected 21st century changes in tornado exposure, risk, and disaster potential. Climatic Change, 2017.

[60] Florida Department of Transportation. Florida transportation trends and conditions. www.dot.state.ﬂ.us/planning/trends, June, 2014.

[61] William M. Trochim. Inferential statistics. http://www.socialresearchmethods.net/kb/, 2006.

[62] Stephanie M. Verbout, Harold E. Brooks, Lance M. Leslie, and David M. Schultz. Evolution of the U.S. tornado database: 1954-2003. Weather and Forecasting, 21:86–93, 2006.

[63] Kathrin Wapler, Thomas Hengstebeck, and Pieter Groenemeijer. Mesocyclones in Central Europe as seen by radar. Atmospheric Research, 168:112–120, 2016.

[64] Morton D. Winsberg, James O’Brien, David Zierden, and Melissa Griﬃn. Florida Weather, second edition. University Press of Florida, 2003. 90-93.

[65] Joshua Wurman, Curtis Alexander, Paul Robinson, and Yvette Richardson. Low-level winds in tornadoes and potential catastrophic tornado impacts in urban areas. Bulletin of the American Meteorological Society, 88(1):31–46, 2007.

[66] Zillow. United States home prices & home values. https://www.zillow.com/home-values/, 2018.

70 BIOGRAPHICAL SKETCH

The author, Emily Ryan, grew up on a farm in a small town in Minnesota. From an early age she obtained a fascination with weather and storms. She has always had a love for the environment and understanding how it changes over time. After graduating from high school in 2012, she chose to attend the University of Oklahoma in Norman. During her time at OU, she participated in many organizations including playing clarinet in the Pride of Oklahoma marching band, holding leadership positions for the Geography and Environmental Sustainability Club, and working part-time at the One University Store as a Product Specialist. In 2016, she earned her Bachelor’s of Science degree in Environmental Sustainability with a focus on Science and Natural Resources and a minor in Meteorology. She decided to attend Florida State University to earn her Master’s of Science degree in Ge- ography. She has worked as a teaching assistant for Environmental Science and Introduction to GIS, while researching and attending class. Upon graduation, her future plan is to find a job that will help her grow and expand her professional skill set. She is interested in working in Emergency Management or in a GIS position, where she can help people and make a difference. In her free time, the author enjoys being around family and friends, hiking, fishing, dancing, listening to music, and continuously learning about the world. She is looking forward to someday traveling the world and seeing where life takes her.