UNIVERSITY OF CALGARY
Mitigation of the Impact of Tornadoes in the Canadian Prairies
by
Samanthi Walawe Durage
A THESIS
SUBMITTED TO THE FACULTY OF GRADUATE STUDIES
IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE
DEGREE OF DOCTOR OF PHILOSOPHY
DEPARTMENT OF CIVIL ENGINEERING
CALGARY, ALBERTA
APRIL, 2014
© Samanthi Walawe Durage 2014
Abstract
Tornadoes are a destructive form of the extreme weather associated with thunderstorms. Canada
gets more tornadoes than any other country with the exception of the United States. This thesis is
the first study centered on the mitigation of the impact of tornadoes in the Canadian Prairies.
Initially, a regression based analysis of the Prairie tornado database is conducted to obtain the
trend for the number of tornadoes reported in each year. Given that the population is an
influencing factor; both population and time are considered as important parameters in the model
development. Based on this analysis, a wave form for the time trend with a period of around 65
years is recognized.
The detection, warning and communication stages at the pre-touchdown phase of a tornado are
analyzed using network modelling and simulation methods. The simulation results of the
network developed for the City of Calgary illustrating the role of collaborating partners provide a
probabilistic representation of the overall time consumption from tornado detection point to the
warning completion point. Furthermore, a qualitative comparison of the Canadian tornado
detection, warning and communication system with the US system supports the recognition of
key areas that need to be improved.
The evacuation response behaviour of households and drivers during a tornado is analyzed
through a stated preference survey. Probit models are developed to examine the factors
influencing the evacuation behaviour of households and drivers. The behavioural responses that
ii
emerge from the survey provide important factors to be considered in mitigating the impact of
tornadoes at the individual level as well as the community level.
The total time consumption from the warning issuance point to the evacuation completion point
is analyzed by combining the time distribution of the network and the evacuation time
distribution developed based on the survey data. The comparison of the probabilistic time
estimates obtained from the resultant distribution with the warning lead time of around 10
minutes indicates why it is imperative to improve the warning lead time.
The forecasters‟ warning decision-making and the public‟s decision to respond to a warning is
analyzed using the decision tree approach. A logical basis for the warning decision-making is
developed and a fundamental inequality of decision making for issuing tornado warnings is
identified. False warning and missed event probabilities are also analyzed using the data from the
Canadian Prairies. Further, the underlying factors for false warnings and missed events are
analyzed in detail. The overall analysis provides important suggestions to improve the warning
performance.
The research contributes to a deeper understanding of the tornado detection, warning and
communication system by providing an overall analysis that spans across different areas under
the general umbrella of tornado disaster mitigation. A set of recommendations are offered as
guidelines for consideration and possible adoption by stakeholders who are involved at different
stages of the tornado detection, warning, communication and evacuation process.
iii
Acknowledgements
It is a pleasure to thank the many people who made this thesis possible. I express my sincere
gratitude to my research supervisor Dr. S.C. Wirasinghe for his guidance and inspiration that
nurtured this research. I have learnt many professional values of a mentor from you. The
privilege to work with you is the most memorable gift I received in my academic journey. I am
grateful to my co-supervisor Dr. Janaka Ruwanpura for his advice and insightful comments that
have influenced my way of thinking about the subject of this thesis. I wish to thank: Dr. Lina
Kattan for her great support to design the survey and analyze data, as well as Dr. Shawn
Marshall for his valuable input as a climate change professional; Dr. Ian Winchester and Dr.
Niru Nirupama for their time and willingness to engage with my work.
I am also grateful to: Calgary Emergency Management Agency (CEMA), particularly Director
of CEMA and Fire Chief Bruce Burrell, for providing financial and in-kind support to conduct
this research; Chief Tom Sampson, Chief Len MacCharles and Emergency Manager Gregory
Solecki for their support to collect Calgary-based data; the meteorologists of Environment
Canada particularly Patrick McCarthy (Heard- Prairie and Arctic Storm Prediction Centre), Bill
McMurtry and Shannon Bestland for their support in providing data and information; the
meteorologists in the US Storm Prediction Center for their assistance during my 3-day visit.
Thanks also goes to: Mr. Peter Peller (Library –University of Calgary) for his support to collect
population data; the staff in the Department of Civil Engineering for their assistance since the
start of my studies in 2010; all the survey participants for their responses that helped to produce a
iv
unique data set; my senior colleague Dr. Sanjeewa Wickramaratne for his support and feedback
and all my friends for their cherished company.
This journey would not have been possible without the emotional support from my family. I am
grateful to my husband for his love and support. Study is hard work, but it has been a wonderful
journey together. I am also grateful to my parents-in-law for their blessings. Most importantly, I
owe my deepest gratitude to my parents for everything they have done for me.
v
Dedication
As a token of love and respect, I dedicate this thesis to my parents.
vi
Table of Contents
Abstract ............................................................................................................................... ii
Acknowledgements ............................................................................................................ iv
Dedication .......................................................................................................................... vi
Table of Contents .............................................................................................................. vii
List of Tables ..................................................................................................................... xi
List of Figures and Illustrations ........................................................................................ xii
List of Abbreviations ....................................................................................................... xiv
CHAPTER ONE: INTRODUCTION ..................................................................................1
1.1 Research Questions ....................................................................................................3
1.2 Research Objective and Key Steps ............................................................................5
1.3 Outline of the Thesis ..................................................................................................7
CHAPTER TWO: REVIEW OF THE DISASTER-TORNADO ........................................8
2.1 Tornado Definitions ...................................................................................................8
2.2 Tornadoes in Canada .................................................................................................9
2.2.1 Top Ten Worst Tornadoes in Canada ..............................................................12
2.2.2 Tornadoes in the Canadian Prairies .................................................................13
2.3 Tornado Formation ..................................................................................................13
2.4 Climate Change and Tornadoes ...............................................................................15
2.5 Tornado Prediction / Detection ................................................................................17
2.6 Tornado Warning and Communication ...................................................................19
2.7 Tornado Damage and Intensity Rating ....................................................................20
2.7.1 Fujita Scale (F-Scale) ......................................................................................22
2.7.2 Enhanced Fujita Scale (EF-Scale) ...................................................................23
Source: (EC, 2013b) ......................................................................................................25
2.8 Evacuation Strategies for Tornadoes .......................................................................25
2.9 Conclusion ...............................................................................................................28
CHAPTER THREE: ANALYSIS OF THE HISTORICAL TORNADO DATABASE IN
THE CANADIAN PRAIRIES..................................................................................29
3.1 Related Literature ....................................................................................................29
3.2 Data ..........................................................................................................................30
3.3 Tornado Frequency Variability ................................................................................31
3.4 Model Development ................................................................................................33
3.5 Final Model Selection ..............................................................................................41
3.6 Time Series Analysis ...............................................................................................43
3.7 Discussion ................................................................................................................53
CHAPTER FOUR: ANALYSIS OF THE TORNADO DETECTION, WARNING AND
COMMUNICATION (TDWC) SYSTEM IN CANADA ........................................55
4.1 Related Literature ....................................................................................................55
4.2 Research Questions ..................................................................................................56
4.3 Network Modelling Approach .................................................................................57
4.3.1 Monte Carlo Simulation ..................................................................................59
vii
4.3.2 Triangular Distribution ....................................................................................60
4.4 Collaborating Partners .............................................................................................60
4.4.1 Environment Canada –Storm Prediction Centres ............................................60
4.4.2 Spotter Network...............................................................................................63
4.4.3 Alberta Emergency Alert .................................................................................64
4.4.4 Broadcasting Media .........................................................................................65
4.4.5 Calgary Emergency Management Agency (CEMA) .......................................66
4.4.6 Police and 911 .................................................................................................67
4.4.7 Public ...............................................................................................................67
4.5 Activity Network Development ...............................................................................68
4.6 Network Simulation .................................................................................................71
4.6.1 Overall Network ..............................................................................................71
4.6.2 Network with SPC Activity Sequence ............................................................72
4.6.3 Network with Local Level Activity Sequence ................................................74
4.7 Analysis of the Total Time for Warning, Communication and Evacuation ............75
4.7.1 Curve Fitting ....................................................................................................76
4.7.2 Time from the Warning Issuance to the Warning Receipt Point.....................77
4.7.3 Time from the Warning Receipt Point to the Evacuation Completion Point ..78
4.8 Behaviour of the Network under Different Distributions for Activity Duration .....82
4.9 Conclusion ...............................................................................................................84
CHAPTER FIVE: COMPARISON OF THE CANADIAN AND US TORNADO
DETECTION AND WARNING AND COMMUNICATION SYSTEMS..............85
5.1 Introduction ..............................................................................................................85
5.2 US Tornado Detection, Warning and Communication System ...............................85
5.2.1 Tornado Warning Stages .................................................................................87
5.2.2 Tornado warnings by WFO .............................................................................89
5.2.3 Warning Communication ................................................................................90
5.2.4 Role of SKYWARN Spotters ..........................................................................93
5.2.5 Emergency Manager and First Responders .....................................................94
5.3 Comparative Analysis of the US and Canadian Systems ........................................94
5.3.1 Prediction/Detection Capabilities ....................................................................95
5.3.2 Warning Provision and Emergency Preparedness ...........................................97
5.3.3 Warning Area ..................................................................................................98
5.3.4 Warning Dissemination Methods ....................................................................99
5.3.5 Spotters‟ Role ................................................................................................100
5.4 Conclusion .............................................................................................................101
CHAPTER SIX: EVACUATION BEHAVIOUR OF HOUSEHOLDS AND DRIVERS
DURING A TORNADO - ANALYSIS BASED ON A STATED PREFERENCE
SURVEY.................................................................................................................102
6.1 Introduction ............................................................................................................102
6.2 Disaster Evacuation ...............................................................................................102
6.3 Stated Preference Survey Method..........................................................................103
6.4 Regression Methods ...............................................................................................104
6.5 Survey Design ........................................................................................................105
6.6 Tornado Knowledge, Preparedness and Previous Disaster Experience ................108
viii
6.7 Tornado Warning Sources of Information .............................................................111
6.8 Pre-evacuation Actions ..........................................................................................116
6.9 Evacuation Wait Time ...........................................................................................118
6.10 Evacuation Actions ..............................................................................................119
6.11 Partially Generalized Ordered Probit Model for Household Evacuation ............124
6.12 Multinomial Probit Model for Evacuation of Drivers .........................................131
6.13 Conclusion ...........................................................................................................136
CHAPTER SEVEN: IMPACT OF FALSE WARNINGS AND MISSED EVENTS ON
TORNADO WARNING PERFORMANCE ..........................................................138
7.1 Warning Decisions .................................................................................................138
7.2 True Warnings, False Warnings and Missed Events .............................................139
7.2.1 True Warning.................................................................................................139
7.2.2 False Warning ................................................................................................140
7.2.3 Missed Event .................................................................................................143
7.2.4 Status Quo .....................................................................................................144
7.3 Warning Spectrum and the Rationale of a Warning ..............................................144
7.4 Factors behind False Warnings and Missed Events...............................................146
7.5 Household Decision Tree.......................................................................................149
7.6 Warning Decision Tree ..........................................................................................152
7.6.1 Application of the Inequality for the City of Calgary ...................................156
7.6.2 A Detailed Illustration of the Decision Tree .................................................159
7.7 Analysis of Tornado Warnings in the Canadian Prairies .......................................160
7.7.1 True Warning, False warning, Detection Probabilities given a Severe Weather
Bulletin...........................................................................................................163
7.7.2 Bayes‟ Theorem based Inferences .................................................................165
7.7.3 True Warning and Missed Event Probabilities based on Tornado Intensity .166
7.8 Conclusion .............................................................................................................167
CHAPTER EIGHT: SUGGESTIONS FOR AN IMPROVED TORNADO MITIGATION
SYSTEM .................................................................................................................168
8.1 Storm Prediction Centre (SPC) ..............................................................................168
8.1.1 Improved Technology ...................................................................................168
8.1.2 Sufficient Forecasters ....................................................................................168
8.1.3 Spotters ..........................................................................................................169
8.1.4 Communication between the Storm Prediction Centre and Local Emergency
Management Agencies ...................................................................................169
8.1.5 Weatheradio Canada ......................................................................................170
8.1.6 Alternative Dissemination Methods ..............................................................170
8.1.7 Information Content of Warnings .................................................................170
8.1.8 False Warning Reduction ..............................................................................171
8.1.9 Missed Event Reduction ................................................................................172
8.1.10 Increased Lead Time ...................................................................................172
8.2 Calgary Emergency Management Agency ............................................................172
8.2.1 Community Awareness and Preparedness.....................................................172
8.2.2 Alberta Emergency Alert System ..................................................................173
8.2.3 Communication Sources ................................................................................173
ix
8.2.4 Community Evacuation Actions....................................................................174
8.2.5 Improved Communication between the Storm Prediction Centre and the Calgary
Emergency Management Agency ..................................................................174
8.3 Alberta Emergency Management Agency .............................................................174
8.4 Spotter Network .....................................................................................................175
8.5 Schools and Children‟s Activity Centres ...............................................................175
8.5.1 Centres‟ Preparedness ...................................................................................175
8.5.2 Parents‟ Evacuation Actions..........................................................................176
8.6 Calgary‟s Road Operations Centre (ROC) ............................................................176
8.6.1 Variable Message Signs ................................................................................176
8.6.2 Peak Period Traffic Preparedness ..................................................................176
8.7 Media .....................................................................................................................177
8.8 Emergency Services ...............................................................................................177
8.9 The Public ..............................................................................................................178
8.9.1 Watches and Warnings ..................................................................................178
8.9.2 Environmental Cues ......................................................................................179
8.9.3 Heeding Warnings .........................................................................................179
8.9.4 One Reliable Source ......................................................................................180
8.9.5 Personal and Family Preparedness ................................................................180
8.9.6 Safest Actions ................................................................................................180
8.9.7 Road Evacuation ............................................................................................181
8.9.8 False Warnings / Missed Events....................................................................181
8.10 Crucial Factors for each Stakeholder ...................................................................182
CHAPTER NINE: CONCLUSION .................................................................................184
9.1 Research Summary and Conclusions .....................................................................184
9.1.1 Tornado Trend in the Canadian Prairies ........................................................184
9.1.2 Application of Network Modelling for Tornado Detection, Warning and
Communication (TDWC) System..................................................................185
9.1.3 Comparison of the US and Canadian TDWC Systems .................................186
9.1.4 Evacuation Behaviour Analysis – Stated Preference Survey ........................187
9.1.5 Analysis of False Warnings and Missed Events............................................187
9.2 Research Contributions ..........................................................................................187
9.3 Research Deliverables............................................................................................189
9.4 Summary and Recommendations for Future Research..........................................191
REFERENCES ................................................................................................................192
APPENDIX A: DATASET USED FOR REGRESSION MODELS ..............................214
APPENDIX B: SIMULATION OF REGRESSION MODELS ......................................215
APPENDIX C: STATED PREFERENCE SURVEY......................................................216
APPENDIX D: COPYRIGHT PERMISSIONS ..............................................................228
x
List of Tables
Table 2-1 : Fujita Scale ................................................................................................................. 22
Table 2-2: EF-Scale in Canada ..................................................................................................... 25
Table 3-1: Correlations among Variables used in the Model ....................................................... 32
Table 3-2: T-test Statistics ............................................................................................................ 36
Table 3-3: Model Comparison ...................................................................................................... 38
Table 3-4: Analysis for Model 4 ................................................................................................... 41
Table 3-5: Analysis for Model 5 ................................................................................................... 42
Table 3-6: Summary of Polynomial Trends for [T observed - T vs t,P] ....................................... 48
Table 4-1: Probability Density Functions ..................................................................................... 76
Table 5-1: Hierarchical Process of Tornado Warnings in the US ................................................ 88
Table 6-1: Profile of Survey Respondents .................................................................................. 107
Table 6-2: Tornado Evacuation Actions ..................................................................................... 120
Table 6-3: Pair-wise Comparisons for Household Actions ........................................................ 122
Table 6-4: Variable Description of the Probit Model for Household Evacuation ...................... 127
Table 6-5: Results of the Partially Constrained Ordered Probit Model ...................................... 128
Table 6-6: Variable Description of the Probit Model for Driver Evacuation ............................. 132
Table 6-7: Results of the Multinomial Probit Regression Model ............................................... 133
Table 7-1: 2x2 Contingency Table ............................................................................................. 139
Table 7-2: Average Number of Summer Severe Weather Reports (1984-2006)........................ 160
Table 7-3: Tornado Warning and Occurrence Records in the Canadian Prairies ....................... 162
Table 7-4: Probabilities related to Tornado Occurrence and Warning Records (2003-2012) .... 164
Table 7-5: True Warning and Missed Event Probabilities based on Intensity (2003-2012) ...... 166
Table 8-1: Recommendations for Stakeholders .......................................................................... 183
xi
List of Figures and Illustrations
Figure 2-1: Picture of a Tornado ..................................................................................................... 9
Figure 2-2: Tornadoes Expected to Occur in about a 4-Year Period ............................................ 10
Figure 2-3: Tornado Distribution in North America..................................................................... 11
Figure 2-4: Tornado Distribution in Canada ................................................................................. 11
Figure 2-5: Hook Echo.................................................................................................................. 18
Figure 2-6: Tornadoes and the Transport System ......................................................................... 28
Figure 3-1: Number of Tornadoes Observed in the Canadian Prairies ......................................... 31
Figure 3-2: Linear Relationship of Population with Time ............................................................ 34
Figure 3-3: Contour Plot ............................................................................................................... 35
Figure 3-4: Model Fit .................................................................................................................... 36
Figure 3-5: Model Output Results ................................................................................................ 44
Figure 3-6: Polynomial for [T observed - T vs t,P] ...................................................................... 44
Figure 3-7: Residual Series ........................................................................................................... 45
Figure 3-8: Sample Autocorrelation Function .............................................................................. 46
Figure 3-9: Overall Trend for the Observed Number of Tornadoes ............................................. 47
Figure 3-10: Tornado Trend in Recent Decades ........................................................................... 53
Figure 4-1: DSSS Template –Model Window .............................................................................. 58
Figure 4-2: Triangular Distribution .............................................................................................. 60
Figure 4-3: An Official Tornado Warning Bulletin ...................................................................... 61
Figure 4-4: Public Warnings in EC Website................................................................................. 62
Figure 4-5: AEA -Critical Alert for a Possible Tornado............................................................... 64
Figure 4-6: Tornado Detection, Warning and Communication Network ..................................... 70
Figure 4-7: CDF Curve for the TDWC Network .......................................................................... 71
Figure 4-8: CDF Curve – Network with SPC Activity Sequence ................................................ 73
xii
Figure 4-9: CDF Curve with Local Level Activity Sequence ...................................................... 75
Figure 4-10: Time distribution from Warning Issuance to Warning Receipt ............................... 78
Figure 4-11: PDF for Household Evacuation Time ...................................................................... 79
Figure 4-12: PDF for Driver Evacuation Time ............................................................................. 79
Figure 4-13: PDF for Household Scenario ................................................................................... 80
Figure 4-14: CDF for Household Scenario ................................................................................... 80
Figure 4-15: PDF for Driving Scenario ........................................................................................ 81
Figure 4-16: CDF for Driving Scenario ........................................................................................ 81
Figure 4-17: Variation with Different Distributions ..................................................................... 83
Figure 5-1: US Tornado Detection, Warning and Communication Network ............................... 86
Figure 5-2: A Tornado Warning issued by a WFO in the US ...................................................... 92
Figure 6-1: Response Percentages to Tornado Knowledge Questions ....................................... 109
Figure 6-2: Tornado Warning Sources ....................................................................................... 112
Figure 6-3: Pre-evacuation Actions ............................................................................................ 117
Figure 6-4: Evacuation Wait Time ............................................................................................. 118
Figure 7-1: Conceptual Model for Warning Accuracy ............................................................... 145
Figure 7-2: Household Decision Tree ......................................................................................... 149
Figure 7-3: Warning Decision Tree ............................................................................................ 153
Figure 7-4: Map of the Maximum U2/U3 Ratio .......................................................................... 158
Figure 7-5: Warning Decision Tree: Detailed Illustration .......................................................... 159
Figure 7-6: Venn Diagram of Tornado Warning and Occurrence Records ................................ 163
xiii
List of Abbreviations
Symbol
Definition
AEA
AEMA
AMS
CANWARN
CDF
CEMA
CMC
DI
DOD
DSSS
EAS
EC
EM
EMS
EPWS
IIA
IPCC
ITS
MSC
NAAD
Alberta Emergency Alert
Alberta Emergency Management Agency
American Meteorological Society
Canadian Weather Amateur Radio Network
Cumulative Distribution Function
Calgary Emergency Management Agency
Canadian Meteorological Centre
Damage Indicator
Degree of Damage
Decision Support Simulation System
Emergency Alert System
Environment Canada
Emergency Manager
Emergency Medical Service
Emergency Public Warning System
Independence of Irrelevant Alternatives
Intergovernmental Panel on Climate Change
Intelligent Transportation Systems
Meteorological Service Canada
National Alert Aggregation and Dissemination
System
National Warning System
National Oceanographic and Atmospheric
Administration
National Severe Storms Laboratory
National Weather Service
NOAA Weather Wire Service
Prairie and Arctic Storm Prediction Centre
Probability Density Function
Probability of Detection
Public Safety Canada
Rich Site Summary
Specific Area Message Encoding
Network of Severe Weather Spotters for NOAA
Stated Preference
Storm Prediction Centre
Tornado Detection Warning and Communication
Texas Tech University
Storm Prediction Center –United States
Variable Message Signs
Weather Forecast Office
NAWAS
NOAA
NSSL
NWS
NWWS
PASPC
PDF
POD
PSC
RSS
SAME
SKYWARN
SP
SPC
TDWC
TTU
USSPC
VMS
WFO
xiv
Chapter One: Introduction
Canada is prone to a wide range of natural disasters. These disasters fall into two main
categories: geophysical disasters, such as avalanches, earthquakes and landslides; and, hydrometeorological disasters, such as hurricanes, floods, hailstorms and tornadoes. The frequent
occurrence and high intensity of natural disasters can impose irreversible negative effects on
people. Taking mitigation actions well in advance can avoid or significantly reduce the impacts
of disasters. Moreover, early recognition of disasters and proper communication of warnings in
the pre-disaster phase help the public to become ready and to respond appropriately and
effectively. In this regard, “… anticipation of natural hazards through warnings, forecasts and
scenarios …” (McBean, 2005, p.358) is an important aspect of disaster management.
Tornado hazards are destructive forms of extreme weather phenomena associated with severe
thunderstorms that threaten life and property. A tornado has a high potential to lead to disaster
when it touches down, causing injury, death or property damage to the vulnerable population
along the path of destruction.
Canada gets more tornadoes than any other country with the exception of the United States
(US).Although the tornado risk is there, the significance of the risk is generally not adequately
appreciated by disaster management authorities and the public. This may be due to reasons such
as the tornado season being limited to summer months and the probability of occurrence being
relatively low (Dore, 2003).
1
The attention and readiness for such an event by weather officials, disaster management
authorities or the public is also not sufficient to respond to a real tornado hazard. For example,
timely detection of isolated tornadoes is a challenging task for forecasters, as these events can
easily go undetected in radar observations. Dissemination of warnings is also a limiting factor
(Etkin et al., 2002), and some people do not hear or heed the warnings before a tornado strikes.
Even if timely warnings are received, some people become nervous and do not have a clear
understanding of how to respond quickly (Dotto et al., 2010).
According to Murphy et al. (2005), until an effective early warning communication model is
implemented, Canadian citizens will be subject to higher levels of risk. Even if the emergency
managers respond immediately in the post-disaster stage, the preparedness by emergency
managers is not adequate to address the pre-disaster stage. To address these issues effectively,
there is a need to improve the awareness of the risk and to introduce risk mitigation strategies at
the local level (Durage et al., 2013a).
Considering the risk and uncertainties associated with tornadoes in Canada, McBean (2005)
concludes that is important to alter Canadian disaster management strategies to account for
greater risk of disaster impact in the future. In disaster management, “a well planned and
practised emergency response program ... smoothly put into action at the appropriate time”
(Comstock and Archer, 2004, p.199) plays a major role in reducing the number of casualties and
the amount of property damage. The proactive phase of a disaster involves both hazard
mitigation and emergency preparedness activities that are also generally perceived as mitigation.
2
Although the hazard potential cannot be reduced, actions can be taken to mitigate the overall
impact by reducing vulnerability to and increasing the capacity to cope with tornado disasters.
Despite various barriers to the implementation of mitigation approaches in disaster management
(Henstra and McBean, 2005), Canada is gradually shifting from the ways that governments have
historically approached disasters, through response and recovery methods, to mitigation
strategies (Emergency Management Act c.15, 2007). Recognizing research as an essential
component, Canada‟s National Disaster Mitigation Strategy (PSC, 2010, p.3) highlights the need
to “apply and promote scientific and engineering best practises in order to build a knowledge
base for sustainable, cost-effective mitigation decisions that contribute to community resiliency”.
Calgary is a city in southern Alberta that is located in the Canadian Prairies bordering the
foothills of the Canadian Rockies. The Calgary Emergency Management Agency (CEMA),
which is the local level partner in emergency management, has been concerned about the hazard
potential of tornadoes and is looking into mitigation measures. How local residents and
emergency managers in Calgary receive tornado warnings and how they should react to them are
the major issues being considered. This research analyzes the key issues associated with the
present tornado detection, warning and communication (TDWC) system in detail and focusses
on how to develop a more efficient system to mitigate the impact of tornadoes, using the City of
Calgary as a case study.
1.1 Research Questions
The main research questions are:
3
1. What are the problems associated with the present TDWC system?
2. How should the overall efficiency of the system be improved to mitigate the impact of
tornadoes?
In particular, this research investigates the area where the City of Calgary is located. The
following are specific questions that this study intends to answer:

What is the tornado frequency trend over the Canadian Prairies?

How do different institutional partners collaborate with each other to detect, warn and
communicate the tornado threat to the public?

How can the network modelling concept be used to model and analyze the TDWC system
in Canada, in order to obtain the overall time for the activity sequence from tornado
detection to warning dissemination?

What is the total time from the tornado warning issuance to the point where the public
has completed their evacuation actions?

What is the behaviour of the activity network when a different distribution is used to
represent the uncertainty of activity durations?

How is the Canadian TDWC system different from the US system? What are the lessons
that can be learned when compared with the US system?

How will Calgary households and drivers behave during the evacuation phase of a
tornado emergency?

How should the household decision-making to evacuate or not in response to a tornado
warning be analyzed?
4

How should the decision-making process of forecasters to issue or not to issue a warning
in relation to the disutility of the public be analyzed?

What are the issues associated with false warnings and missed events in the Canadian
Prairies and how should they be addressed to provide reliable, timely warnings to the
public?
These questions provide the basis for the development of the conceptual framework of this
research that outlines the objective and the key steps.
1.2 Research Objective and Key Steps
The main objectives of this research are the study, analysis, modelling, simulation and proposal
of improvements to plans and systems to mitigate the impacts of tornadoes in the Canadian
Prairies. The key steps of the research are as follows:

Review of the meteorological phenomenon of tornadoes: how they occur, why they
occur, where they occur, their properties, the types of damage they cause andthe linkage
between climate change and frequency of tornadoes.

Statistical analysis of the historical tornado database in the Canadian Prairies and
development of a model to represent the tornado frequency trend; statistical testing of the
theories that their frequency may be increasing with time. Regression and time series
analysis methods are used to develop models and analyze the database.

Review of the TDWC system in the Canadian Prairies and development of an activity
network, using the City of Calgary as a case study; use of triangular distribution to
represent activity durations and simulate the network using the DSSS template of
5
Simphony software. Monte Carlo simulation methods are used to determine the overall
time durations of the network.

Analysis of the behaviour of the network when a different distribution is used to represent
the time variation of each activity in the TDWC network.

Analysis of the total time duration from the warning issuance point to the evacuation
completion point.

Review of the US tornado detection, warning and communication system and a detailed
comparison with the Canadian system. A qualitative analysis is conducted and
improvements are proposed to the Canadian system.

Stated preference survey to collect data about the evacuation behaviour of Calgarians
during a tornado emergency. An online survey is conducted to analyze the evacuation
behaviour of households and drivers. Data is analyzed using probit regression methods,
and the conclusions are derived based on the output results.

Analysis of false warnings and missed events probabilities in the Canadian Prairies. False
warning and missed event records are analyzed using conditional probability methods.
The expected disutility minimization concept is applied to develop a logical basis for the
warning decision-making process.

Conclusions and recommendations based on the overall research.
These key steps are explored and the results discussed in detail in the remaining chapters of the
thesis. This is applied research with practical significance, as deficiencies are identified and
suggestions are provided to improve the TDWC system in the Canadian Prairies. Overall, the
6
output results lead to the establishment of a set of guidelines for emergency response plans to
mitigate the impact of tornadoes.
1.3 Outline of the Thesis
The rest of the thesis is organized as follows. Chapter Two deals with reviewing tornadoes,
focussing on factors such as formation, detection, damage and linkage with climate change.
Reviewing the literature is important to obtaining a comprehensive understanding of the research
theme. Considering the diverse nature of the topics addressed under the umbrella of this tornado
research, the related literature is discussed prior to the analysis in each chapter.
Chapter Three provides a statistical analysis of the historical tornado database in the Canadian
Prairies. Chapter Four reviews the TDWC system in Canada and analyzes the use of the network
modelling approach. A comparison of the Canadian and US tornado detection and warning
systems is provided in Chapter Five. Chapter Six describes the stated preference survey
conducted to analyze the evacuation behaviour of households and drivers during a tornado.
The impact of false warnings and missed events on tornado warning performance is investigated
in Chapter Seven. A set of recommendations for each stakeholder is provided in Chapter Eight.
Chapter Nine summarizes the overall research work and discusses research contributions and
deliverables.
7
Chapter Two: Review of the Disaster-Tornado
The aim of this chapter is the review of the meteorological phenomenon of tornadoes in order to
lay a foundation for this research. The term „tornado‟ is briefly described, followed by a
discussion on tornadoes in Canada. Aspects such as tornado formation, detection, warning,
communication and evacuation strategies are discussed in detail in the subsequent sections.
2.1 Tornado Definitions
Tornadoes are nature‟s most violent storms with huge destructive forces that can destroy an
entire community (NOAA, 1995). According to the Glossary of Meteorology- American
Meteorological Society (AMS) , a tornado (Figure 2-1) can be defined as “a violently rotating
column of air, in contact with the surface, pendant from a cumuliform cloud, and often (but not
always) visible as a funnel cloud” (AMS, 2013). According to Grazulis (2001, p.11), the
phenomenon of a tornado is a “...naturally occurring atmospheric vortex whose circulation
extends from the ground at least to the base of a convective cloud”. The Oxford dictionaries
website defines a tornado as “a mobile, destructive vortex of violently rotating winds having the
appearance of a funnel-shaped cloud and advancing beneath a large storm system” (Oxford,
2013).
These definitions indicate the common nature of a tornado as a violently rotating column of air
extending from a thunderstorm cloud to the ground. The term „waterspout‟ is used to define any
tornado over a body of water. Landspout is also a term used for a tornado that is not associated
with a mesocyclone (or supercell) of a thunderstorm. A gustnado is “a small, whirlwind which
8
forms as an eddy in thunderstorm outflows” (NOAA, 2013); however, it is not considered to be a
tornado, as it is not associated with a storm-scale rotation.
Figure 2-1: Picture of a Tornado
Picture Courtesy: Environment Canada
2.2 Tornadoes in Canada
There have been reports of tornadoes on all the continents, except Antarctica. A map of tornado
occurrences expected in about a 4-year period reproduced by Goliger and Milford (1998) based
on early work done by Fujita (1973), shows that tornadoes occur on both hemispheres between
the latitudes of 20ο and 60ο(Figure 2-2). However, the frequency of occurrence is highest in
North America, especially in the US (Goliger and Milford, 1998).
9
Figure 2-2: Tornadoes Expected to Occur in about a 4-Year Period
Source: Goliger and Milford (1998)Copyright (2013), with Permission from Elsevier
The geography of North America creates favourable conditions for the development of tornadoes
(Figure 2-3). Cold dry air from the Rocky Mountains meets the warm, moist air from the Gulf of
Mexico, creating atmospheric instabilities that induce severe thunderstorms and tornadoes. North
America has the most tornado-prone region in the world, which is called Tornado Alley.
Historical records of tornadoes show that there are two main clusters of tornado-prone regions in
Canada: the central part of Canada, including Southern Ontario and Quebec; and, the Canadian
Prairie region, including the provinces of Alberta, Saskatchewan and Manitoba (Figure 2-4).
According to Environment Canada (EC), which is the authority responsible for tornado detection
and warning, an average of 43 tornadoes per year occur across the prairies provinces and about
10
17 occur across Ontario and Quebec (EC, 2013a).This regional pattern results from a
combination of climatology and the population distribution pattern (Etkin, 1995).
Figure 2-3: Tornado Distribution in North America
Source: Grosvenor et al.(1998)
Figure 2-4: Tornado Distribution in Canada
Source: Environment Canada
11
Tornado climatology can be better explained with the prevalent storm formation mechanisms
across the country. Moreover, formation of air boundaries is a main factor that explains the
localized tornado incidence patterns (Etkin et al., 2001). It seems that this distribution pattern is
associated with the bordering regions in the US, where Tornado Alley is located.
As noted from the tornado distribution map, tornado occurrences reported are not geographically
uniform over a region, and there is a general decline in the number of tornadoes as one travels
northward. Several researchers have also identified the compatibility of the number of tornadoes
reported in Canada with the population distribution pattern (Newark, 1984; Etkin et al., 2001).
Tornadoes that occur in unpopulated areas are not generally reported; thus, there can be a
population bias.
2.2.1 Top Ten Worst Tornadoes in Canada
Historical tornado records provide evidence that many densely populated regions in the Canadian
Prairies, such as Edmonton and Regina, have had devastating tornadoes. Environment Canada
has ranked the ten worst tornadoes in the recorded history based on death count as:

Regina, Saskatchewan - June 30, 1912 - 28 dead, hundreds injured

Edmonton, Alberta - July 31, 1987 - 27 dead, hundreds injured

Windsor, Ontario - June 17, 1946 - 17 dead, hundreds injured

Pine Lake, Alberta - July 14, 2000 - 12 dead, 140 injured

Valleyfield, Quebec - August 16, 1888 - 9 dead, 14 injured

Windsor, Ontario - April 3, 1974 - 9 dead, 30 injured

Barrie, Ontario - May 31, 1985 - 8 dead, 155 injured

Sudbury, Ontario - August 20, 1970 - 6 dead, 200 injured

St-Rose, Quebec - June 14, 1892 - 6 dead, 26 injured
12

Buctouche, New Brunswick - August 6, 1879 - 5 dead, 10 injured
The Regina tornado in 1912 ranks as Canada‟s deadliest tornado disaster with a record of 28
deaths and hundreds of injuries. Although the above ranking is based on fatality counts, all these
tornadoes also had devastating impacts on property.
2.2.2 Tornadoes in the Canadian Prairies
The Canadian Prairie region, namely the three provinces of Alberta, Saskatchewan and
Manitoba, is one of the most active areas in Canada for severe summer thunderstorms.
Meteorological conditions that produce severe thunderstorms can arise in any part of the
Canadian Prairies (Paul, 1982). Four aspects of thunderstorms that are potentially hazardous are
hail, heavy rain, lightning and strong winds (with occasional tornadoes).
Summer rainfall in the prairie provinces brings severe weather conditions that expose the region
to natural hazards. Tornadoes are the most vigorous winds associated with thunderstorms. Higher
frequencies of tornado occurrences can be observed in the months of June, July and August,
which have significant amounts of summer rainfall associated with severe thunderstorms. The
winter season has the lowest probability of tornado occurrence.
2.3 Tornado Formation
Tornadoes are complex events that have small-scale trigger mechanisms for formation (Murphy
et al., 2005). Meteorological conditions that contribute to the development of tornadoes can be
explained by both the synoptic scale and the local level environmental conditions (Etkin et al.,
13
2001). Meteorological elements such as variations in moisture, wind and temperature that lead to
tornadic storms are generally similar to conditions required for the formation of severe storms.
Temperature is a main factor that contributes to suitable conditions for thunderstorm
development. A high temperature in the lower atmosphere leads to increased moisture content
through evaporative processes. The formation of clouds in this moisture-rich atmosphere and the
associated turbulent wind fields induce severe convective storms. There is a possibility for a
tornado when a continuously rotating updraft of air develops within a supercell storm. A rotating
updraft within a supercell can be present as much as 20 to 60 minutes before a tornado forms
(NSSL, 2009).
Movement of the storm and its complex interaction with the surrounding wind shear
environment can develop instabilities and strengthen supercells, thereby forming funnel clouds.
When a funnel cloud made up of water droplets descends from the parent thunderstorm
ultimately reaching the ground with a dangerous suction force and winds that can lift up objects
to the air, it is called a tornado.
Although many devastating tornadoes are associated with supercells, not all supercells generate
tornadoes. It is also noteworthy to mention that not all the tornadoes are induced from supercell
thunderstorms. There are non-supercell tornadoes that do not form from an organized stormscale rotation. In addition, air boundaries at the local level also create favourable environments
for tornado formation (Etkin et al., 2001) without a supercell. Such a tornado can develop when
14
an updraft moves over and stretches a vertically spinning parcel of air already occurring near the
ground due to wind shear (NSSL, 2009).
2.4 Climate Change and Tornadoes
Climate change has been generally manifested in the increase of the average global temperature
in recent decades. There are supportive arguments in climatology-related research about changes
in tornado frequency due to climate change (Etkin, 1995; Dore, 2003; McBean, 2005; Dotto et
al., 2010). The analysis of historical data on tornadoes in Canada by Etkin (1995) has indicated
that tornado frequency increases with the positive mean monthly temperature anomalies that
represent a gross measure of climate. Analyzing the tornado risk in Canada in the context of a
changing climate, McBean (2005) argued that there is no evidence that the number of tornadoes
will decrease in the future and highlighted the need for risk management strategies to assume
more frequent tornado events.
A warmer atmosphere can bring more moisture to the atmosphere through feedback processes.
Holding more water vapour can add more energy (i.e. convective available potential energy,
which varies with the atmospheric water vapour content) to the atmosphere, fuelling the
development of thunderstorms. Increase in the sea surface temperature due to global warning
also provides warm humid air to inland areas; and, supercell formation occurs when a warm
moist air flow and a cold, dry air flow collide with each other. This indicates that a warming
climate may have a positive influence on the formation of supercells.
15
The presence of a strong jet stream is also an important component in severe weather
development. The position of the jet stream helps determine which regions are more likely to
experience severe weather. A warmer atmosphere can change the jet stream position, impacting
the local level meteorological conditions that lead to severe weather. In a warmer atmosphere,
the wind shear behaviour is, however, questionable, due to the decrease of the temperature
difference required for forming atmospheric instabilities between warm, moist and cold, dry air
masses. This indicates that a warming climate can influence the elements for tornado formation
in different ways.
Even if a warming climate has a positive influence on the formation of tornadoes at the local
level, its impact on the tornado frequency is still debatable. A recent article about extreme
weather events highlights the issue by asking “Can violent hurricanes, floods and droughts be
pinned on climate change?” and answering “Scientists are beginning to say yes” (Schiermeier,
2011, p.148). Moreover, Schiermeier (2012, p.125) says, “Climate scientists believe that the
frequency and severity of extreme weather events will increase as temperatures continue to rise.”
Nevertheless, “there is insufficient evidence to determine whether trends exists in ...small-scale
phenomena such as tornadoes, hail, lightning and dust storms” (IPCC, 2007, p.33). The US
National Oceanic and Atmospheric Administration (NOAA,2011a, para.9) also pinpoints that “a
change in the mean climate properties that are believed to be particularly relevant to major
destructive tornado events has... not been detected...”.
The insufficient evidence to detect an increasing trend in tornado frequency does not mean that
there is no linkage between tornadoes and climate change. This relationship is very complex, and
16
further investigations are required to derive a conclusion. However, the hazard potential
associated with the climate factor should not be underrated.
2.5 Tornado Prediction / Detection
Development of tornadoes is a complex process, and these small-scale localized events are hard
to detect and forecast (Murphy et al., 2005; Cao and Cai, 2011). There is still no accepted
methodology for precisely predicting tornadoes (Stensrud et al., 2009). However, Doppler radars
provide information on wind speeds that can be used to detect rotations, in order to infer tornado
activities and their approximate locations and issue warnings. There is the possibility for a
tornado occurrence when a continuously rotating updraft of air within a supercell is detected by
Doppler radar.
The Doppler radar (which is an acronym for radio detection and ranging) is a life-saving
discovery developed based on the Doppler effect, which is named after the Austrian physicist
Christian Doppler (1803-1853). According to the NOAA,
Doppler radar uses radio waves to create pictures showing the location and
intensity of precipitation. Doppler radar allowed scientists to measure motion
inside storms for the very first time, providing valuable clues into the
development of severe weather. Using this radar, scientists also discovered that
when a tornado begins to form, its winds blow raindrops in a way that appears as
a distinguishing pattern or signature on the radar screen. (NOAA, 2012a, para. 1)
17
Doppler radar detects tornadoes based on radar reflectivity data. Based on the pattern in radar
reflectivity data, a hook echo, which looks like a hook extending from the radar echo, can be
recognized (Figure 2-5). The hook echo is often associated with a deeply rotating updraft of air
called a mesocyclone. The presence of a mesocyclone is an indication of a supercell. A detection
algorithm called a tornado vortex signature can detect supercells, alerting forecasters about
tornado formation. With recent advancements in dual-polarized technology, radars send out both
horizontally and vertically polarized waves, which can confirm that tornadoes are on the ground
causing damage (NSSL, 2013a).
Figure 2-5: Hook Echo
Source: (NSSL, 2013b)
The use of Doppler radars improves the tornado warning performance, increasing the probability
of detection and mean warning lead time, while reducing the probability of false warnings
(Simmons and Sutter, 2006). Satellite images and numerical weather prediction models are also
used to analyze severe weather information and detect tornado activities.
18
Although there are various methods to detect the tornado potential, important factors, such as the
timing, location and intensity of these events, cannot be precisely predicted. In addition to the
systematic ways for the monitoring and detection of tornadoes, people can identify incoming
tornadoes through environmental cues, such as a dark or greenish sky, large hail, thunder and
lightning, funnel clouds and rumbling sounds. People who live in tornado-prone regions need to
be alert to severe weather to detect these cues.
Although a supercell may be recognized, still there is not a verified method to clearly identify the
tornadogenesis of a supercell. McGovern et al. (2014, p. 29, 30) are researching “the
development of novel spatiotemporal data mining techniques for discriminating between
supercell storms that produce tornadoes and those that do not”. They apply these novel
techniques to numerical models in order to “reveal more clues about the processes that lead to
tornado formation within some supercell thunderstorms but not within others”. The results of this
research will likely have a significant contribution to the tornado detection and warning process.
2.6 Tornado Warning and Communication
“Warnings are the culmination of a sequence of actions taken by forecasters that act to alert the
public to a heightened probability of high-impact weather minutes, hours, or even days in
advance” (Stensrud et al., 2009, p.1487). Having a fully integrated warning decision and
dissemination system plays a major role in saving lives and reducing the number of injuries
during tornado occurrences (McCarthy, 2001).
19
Issuing timely and rapid watches and warnings are very important in warning the public about
the threat in advance. The warning lead time of a tornado is very small compared to other
disaster warnings: sometimes, it is even zero or negative (Brotzge and Erickson, 2009). In the
US, the average warning lead time is 14 minutes (NOAA, 2011b). In Canada, a warning can be
issued with a lead time of around 10 minutes, if the tornado is within the coverage of a Doppler
radar installation (MSC, 2003).
Outside of the Doppler radar coverage area, warnings are issued based on eyewitness reports of
funnel clouds or tornadoes in the area. Since the warning lead time is very small, even a delay of
a minute in the information flow can have severe impacts. It is also very difficult to precisely
predict a tornado‟s touchdown point, its path and the size of the forecasted region (McBean,
2005): a single tornado can have multiple touchdowns. Due to these reasons, generally warnings
are issued for a large area, although the impacts are localized.
Communication of tornado warnings should, therefore, be a rapid process that requires giving
information to the public without any avoidable delays. There are various methods that a
forecaster can use to communicate a warning to the public, such as local media (TV and radio),
the Internet and social media, so that they can seek shelter immediately.
2.7 Tornado Damage and Intensity Rating
The lift created by rotating updrafts is the major force that raises objects and structural elements
into air. A rotational wind field within the diameter of a tornado is spread over a large area in
comparison with the dimensions of structures (Murray and McDonald, 1993), exposing them to
20
approximately unidirectional winds. Houses and other structures that are not properly anchored
to their foundations have little capability in withstanding strong winds. The damage path
trajectory of a tornado is influenced by factors such as building strength, orientation, number and
type of openings, roof type, degree of shielding and impact by neighbouring objects. For
example, openings on the windward side of a building can increase the internal wind pressure,
causing additional updraft on the roof (Marshall, 1993).
Tornadoes that strike some structures can even trigger secondary disasters. For example, missiles
generated due to airborne debris can damage structures and safety-related infrastructure or
expose equipment of nuclear power plants, thereby inducing a leak of radiation (Ravindra, 1993).
A tornado that goes through an industrial area can be a precursor to environmental incidents.
Therefore, tornado wind loads become critical in designing structures, such as nuclear power
stations, transmission lines and other industrial facilities that are exposed to severe weather
(Goliger and Milford, 1998).
Generally, tornado damage is highly localized; and, the damage area is small compared to other
major disasters. However, damage per unit area on the path of a powerful tornado is not inferior
to any other disaster. Careful examination can reveal how the damage occurred and provide
insight into the wind speeds associated with such damage. Ground observations, aerial
photographs and satellite images are the major methods used in damage estimation (Yuan et al.,
2002).
21
2.7.1 Fujita Scale (F-Scale)
The intensity of a tornado can be measured by wind speed and the damage done due to a tornado
and is given a rating on the Fujita scale (F-scale). The F-scale rating for tornadoes based on the
damage was first introduced by Tetsuya Fujita in 1971. This is the first scale (Table 2-1)
introduced to differentiate one tornado disaster from another (Fujita, 1971).
Table 2-1 : Fujita Scale
F-Scale
Wind
Speed Damage Level
(km/h)
F0
64-116
F1
117180
F2
181253
F3
254332
F4
333418
F5
419512
Light Damage
Some damage to chimneys and TV antennae; breaks twigs off trees; pushes
over shallow rooted trees
Moderate Damage
Peels surface off roofs; windows broken; light trailer houses pushed or
overturned; some trees uprooted or snapped; moving automobiles pushed off
the road
Considerable Damage
Roofs torn off frame houses leaving strong upright walls; weak buildings in
rural areas demolished; trailer houses destroyed; large trees snapped or
uprooted; railroad boxcars pushed over; light object missiles generated; cars
blown off highway
Severe Damage
Roofs and some walls torn off frame houses; rural buildings completely
demolished; trains overturned; steel framed hangar/warehouse type
structures torn; cars lifted off the ground; most trees in a forest uprooted,
snapped, or levelled
Devastating damage
Whole frame houses levelled, leaving piles of debris; steel structures badly
damaged; trees debarked by small flying debris; cars and trains thrown some
distances or rolled considerable distances; large missiles generated
Incredible Damage
Whole frame houses tossed off foundations; steel-reinforced concrete
structures badly damaged; automobile-sized missiles generated; incredible
phenomena can occur
Source: (Fujita, 1971)
22
The F-scale gives wind speeds associated with damage caused by a tornado. A higher F-scale
rating is associated with a higher wind speed. The assignment of an F-scale rating based on the
degree of damage to structures is a subjective visual procedure (Marshall, 1993). A proper Fscale estimate requires re-evaluation of initial estimates on the basis of detailed information
about structural integrity. In addition to this subjective bias, the F-scale can be dependent on
observer training and experience. Therefore, a rating can have ±1 F number (Edwards et al.,
2013).
Furthermore, in this traditional F-scale, the winds that are associated with each level of damage
are not very well correlated. To overcome these limitations a new scale called the Enhanced
Fujita scale (EF-scale) began operational use in the US and Canada in 2007 and 2013,
respectively.
2.7.2 Enhanced Fujita Scale (EF-Scale)
The EF-scale in Canada has been derived based on minor revisions made to the EF-scale in the
US. According to Environment Canada,
The EF-scale employs a large number of damage indicators, ranging from
residential housing to office towers to trees. Wind speeds are more accurately
related to wind damage thanks to an expert elicitation process that involved the
meteorological, engineering and architecture communities. The damage ratings
are also backwards compatible with the original F-Scale; only the associated wind
speeds have undergone major changes.(EC, 2013b, para.4)
23
Environment Canada uses 31 damage indicators (DI) to evaluate the damage. For each DI, there
are several degrees of damage (DOD). Wind speeds associated with each DOD are obtained
through an expert elicitation process. The recommended approach for assigning an EF-scale
rating for a tornado involves the following steps (TTU, 2006):

Conduct an aerial survey of the damage path to identify possible DIs and define the
extent of the damage path;

Select several DIs that tend to indicate the highest wind speed within the damage path;

Locate those DIs within the damage path;

Conduct a ground survey and carefully examine the DIs of interest;

Follow the steps outlined for assigning an EF-scale rating to individual DIs and document
the results;

Consider the estimated wind speeds of several DIs, if available, and arrive at an EF-scale
rating for the tornado event;

Rate the tornado intensity by applying the highest rated DI, provided there is supporting
evidence of similar damage intensity immediately surrounding the DI;

Record the basis for assigning an EF-scale rating to the tornado event; and,

Record other pertinent data relating to the tornado event.
Wind speeds associated with the EF-scale in Canada are summarized in Table 2-2.This is an
improved scale with more DIs, which account for construction variability. It is noteworthy that
the use of the EF-scale in Canada is in a very preliminary stage and requires strengthening of the
institutional capacity to conduct ground level investigations.
24
Table 2-2: EF-Scale in Canada
EF-Scale Rating
EF-Scale Wind Speed (Rounded to 5km/h)
0
90-130
1
135-175
2
180-220
3
225-265
4
270-310
5
315 or more
Source: (EC, 2013b)
2.8 Evacuation Strategies for Tornadoes
Evacuation strategies for tornadoes are somewhat different from other disaster evacuation
methods. The recommended proactive action is to shelter in place. In the absence of a nearby
safe location, people have to find evacuation routes to move to a safer place immediately. The
Government of Canada gives tornado evacuation instructions to the public who are involved in
various activities through the website http://www.getprepared.gc.ca (GP, 2013) as follows:

If you are in a house

Go to the basement or take shelter in a small interior ground floor room, such as a
bathroom, closet or hallway.

If you have no basement, protect yourself by taking shelter under a heavy table or
desk.

In all cases, stay away from windows, outside walls and doors.
25

If you live on a farm

Livestock hear and sense impending tornadoes. If your family or home is at risk, the
livestock will be a non-issue. If your personal safety is not an issue, you may only
have time to open routes of escape for your livestock. Open the gate, if you must, and
then exit the area in a tangent direction away from the expected path of the twister.

If you are in an office or apartment building

Take shelter in an inner hallway or room, ideally in the basement or on the ground
floor.


Do not use the elevator.

Stay away from windows.
If you are in a gymnasium, church or auditorium

Large buildings with wide-span roofs may collapse if a tornado hits.

If possible, find shelter in another building.

If you are in one of these buildings and cannot leave, take cover under a sturdy
structure such as a table or desk.

Avoid cars and mobile homes

More than half of all deaths from tornadoes happen in mobile homes.

Find shelter elsewhere, preferably in a building with a strong foundation.

If no shelter is available, lie down in a ditch away from the car or mobile home.
Beware of flooding from downpours and be prepared to move.
26

If you are driving

If you spot a tornado in the distance, go to the nearest solid shelter.

If the tornado is close, get out of your car and take cover in a low-lying area, such as
a ditch.

In all cases

Get as close to the ground as possible, protect your head and watch for flying debris.

Do not chase tornadoes – they are unpredictable and can change course abruptly.

A tornado is deceptive. It may appear to be standing still but is, in fact, moving
toward you.
This website also emphasizes the importance of having a family emergency plan to ensure
families are safe in emergency situations.
At the community level, evacuation actions in a timely manner are a greater challenge. For
example, evacuation of people from a large outdoor activity within a very short time can create
many problems and result in panic, causing injuries with or without any tornado strike.
Transportation is also critical to evacuation operations. The area of the road system directly
impacted by a tornado can be limited to a single interchange and the lead time can only be
several minutes (Figure 2-6). People on roads have to find safe locations immediately. Selfevacuation by vehicle can cause many disruptions, such as traffic jams, aggressive and disrupting
driving behaviours, such as faster speeds than usual, and frantic cell phone calls.
27
Figure 2-6: Tornadoes and the Transport System
Source: (ITS, 2007)
It is important to have well-designed severe weather plans, especially a well-publicized shelter
plan, in order to reduce tornado- and panic-induced causalities (Edwards and Lemon, 2002).
Attention should also be paid on upgrading the local level emergency preparedness plans, so that
people on the roads are safe during a tornado emergency. Moreover, the public should always
remain alert to severe weather watches and warnings, as well as signs of approaching tornadoes,
in order to seek shelter if threatening conditions exist.
2.9 Conclusion
This chapter discusses descriptions of and various aspects related to tornadoes. In particular, this
chapter presents issues that are important in analyzing the current tornado detection, warning and
communication and evacuation system.
28
Chapter Three: Analysis of the Historical Tornado Database in the Canadian Prairies
This chapter describes the methodology adapted to analyze the historical tornado database in the
Canadian Prairies. A novel model developed using regression and time series analysis methods
is discussed.
3.1 Related Literature
Extreme weather events are often the most important aspect of a climate system that is closely
associated with man and the environment (Timbal et al., 2010). In Canada, approximately 80%
of disasters are due to extreme weather events such as tornadoes, hurricanes, hail storms etc
(Hwacha, 2005). There is a growing concern about the frequent occurrence of convective
phenomena such as tornadoes and severe thunderstorms in recent decades.
Assessing the climatology of tornado events is important in various ways to help plan to mitigate
the impacts of tornadoes. A scientific approach can be used for analyzing the time trend, with a
good set of geographically and yearly distributed tornado records (Verbout et al., 2006; Holden
and Wright, 2004).
Apart from the meteorological causes, various factors that have influenced the number of
tornadoes in each year such as population, rainfall, seasonal-scale impacts such as El Niño–
Southern Oscillation (ENSO) events can be analyzed using a historical tornado data set (King,
1997; Etkin et al., 2001; Shepherd et al., 2009). Tornado researches in Canada have highlighted
the population bias as a significant factor influencing the number of tornadoes reported in each
29
year (Newark, 1984; Paruk and Blackwell, 1994; Etkin et al., 2001). This is also evident by the
point map of tornadoes in Canada (Figure 2-4) that shows the distribution concentrated
surrounding the large population centres in the country.
There have been a few studies done to analyze the tornado time trend in Canada. Analyzing
Canadian Prairie windstorms for spatial variations and time trends Hage (2003) says that the
recent increase in observed tornado counts are due to improved identification of weak tornadoes.
A study conducted for the data in Ontario by Cao and Cai (2011) shows the existence of an
upward trend in tornado frequency. Their study has used linear regression, Mann-Kendall test
and Monte Carlo simulation methods to verify the upward trend. Nevertheless, this study lacks
correction for population which is apparent in tornado data sets in Canada. This chapter
addresses the research question of what is the trend of the number of tornadoes observed in the
Prairies in the presence of a population bias. To answer this question, regression based modelling
was conducted using both tornado and population data.
3.2 Data
There have been number of authors who studied tornadoes and contributed to the development of
a historical tornado database in Canada (e.g. Lowe and McKay, 1962 as cited in Etkin et al.,
2001; Newark, 1984). During the last three decades, severe weather data collected by local
weather offices have also been used to improve the dataset. This study uses records of observed
number of tornadoes obtained from the Prairie and Northern Region Severe Weather Database
(Ed. McCarthy, 2011). Population data in the three Prairie Provinces for this period was obtained
30
from the records of Statistics Canada (CANSIM, 2011). The dataset used for the analysis is
shown in Appendix A.
3.3 Tornado Frequency Variability
With the idea of causality in mind, a regression model was constructed for number of tornadoes
observed versus time. A simple fit of a linear trend line for the data set indicates an increasing
trend of tornadoes over time. The T test statistics for this regression shows that the upward trend
(Figure 3-1) is statistically significant at the 95% confidence level.
90
No of Tornadoes Observed
80
70
60
50
Tornadoes
40
Linear
(Tornadoes)
30
20
R² = 0.2651
10
0
1915
1935
1955
1975
1995
2015
Year (1921-2011)
Figure 3-1: Number of Tornadoes Observed in the Canadian Prairies
The upward trend is supportive with the notion that there is an increase in the number of
tornadoes reported with time. Nevertheless, the trend is not purely attributed to the
31
meteorological phenomena; there are several factors hidden behind this trend. One such
important factor is the population or the increased number of eyes to detect and report tornadoes.
Hence, the known fact is that the observed number of tornadoes per year is increasing with
population. The intention on this analysis is to check whether the annual number of tornadoes is
increasing with time. Given that the population is an influencing factor, the model development
needs to consider both population (P) and time (t) as important parameters in the analysis.
Multiple regression method is a useful way of conducting the analysis in the presence of two
independent variables; population and time. However, a simple correlation matrix (Table 3-1)
showed that these two independent variables are highly correlated with each other indicating that
the multi-collinearity is a problem emerged when conducting a multiple regression analysis.
Table 3-1: Correlations among Variables used in the Model
Time (t)
Time (t)
Population (P)
Tornadoes
Population (P)
Pearson Correlation
1
Sig. (2-tailed)
Pearson Correlation
.981
Sig. (2-tailed)
Tornadoes
**
.515**
.000
.000
1
.525**
.981
**
.000
Pearson Correlation
.515
Sig. (2-tailed)
**
.000
.000
.525
**
1
.000
**. Correlation is significant at the 0.01 level (2-tailed).
Attention was paid to find an alternative way to analyze the tornado trend in the presence of both
population and time impact. The following section shows a novel method developed to remove
the impact of the population and analyze the remaining trend over time.
32
3.4 Model Development
The analysis focussed on developing several models and choosing the best model that satisfies
the requirements. Here, the key steps of the model development is explained taking the model
which was chosen later for further analysis as an example.
I.
A model to represent the influence of both population and time is recognized as
To(t) =TA eβP^k tα .
II.
(3.1)
where To(t) = observed number of tornadoes
t= time (in years)
P= population (in millions)
TA , β, k,α = constants
Due the non-linear nature, this model cannot be estimated with linear regression analysis
as written. A log transformation of the equation can convert it into a
„log-linear‟
relationship that facilitates linear regression. Taking the log of the equation (3.1) gives,
ln To(t)= ln TA+ β Pk +αln t.
III.
(3.2)
Now, the objective is to obtain an expression in terms of time (t), replacing the
population related term with a meaningful expression based on t. A simple linear
regression of the population versus time for the data from 1921 to 2011 showed that there
is a strong linear relationship with a coefficient of determination (R2 value) of 0.9632 as
shown in Figure 3-2.
33
7
Population (In Millions)
6
5
4
3
Population
Linear (Population)
2
y = 0.044x + 1.508
R² = 0.963
1
0
1921 1931 1941 1951 1961 1971 1981 1991 2001 2011
Year (1921-2011)
Figure 3-2: Linear Relationship of Population with Time
This linear relationship can be written as
P= A+Bt ;
(3.3)
Substituting P= A+Bt to equation (3.2) gives,
ln To(t)= ln TA+ β Ak(1+Ct)k +α ln t
where C=B/A= 0.0449/1.5087=0.03 (for the data from 1921-2011);
ln To(t)= ln TA+α [ln t +(β/α) Ak (1+Ct)k];
ln To(t)= K1+K2 [ln t +d (1+Ct)k]
where K1= ln TA K2=α
IV.
(3.4)
and d= (β Ak )/α .
Running linear regressions for a range of d and k values using Matlab software gave the
opportunity to find the regression equation with maximum R2 value (Appendix B). A
contour plot of R2 for these d and k values can be obtained as shown in Figure 3-3.
34
Figure 3-3: Contour Plot
The maximum R2 region is shown in dark brown colour contour line of the lower right
quadrant. (The above ranges for d and k values were selected after conducting several
trial and error processes to see the maximum region.) The maximum R2 value of 0.3333
is relevant to the point (d,k) = (4.2,-5). R2 value shows that the model accounts for 33%
of the variance of the independent variable. The corresponding regression equation gives
parameter estimates K1= -0.6302 and K2 = 0.9836. The constants TA, α and β can be
calculated as TA=0.53, α=0.9836 and β=32.3. The resulting linear model can be written as
ln To(t)= -0.6302+0.9836 [ln t +4.2(1+0.03t)-5] .
35
(3.5)
V.
The significance of the linear model is verified through a T test (Table 3-2). A p value
which is less than 0.05 indicates the significance of the X variable.
Table 3-2: T-test Statistics
VI.
Coefficients
Std. Error
T Statistics
p-value
Intercept
-0.63023
0.578543
-1.08934
0.278944
X variable
0.983642
0.147462
6.670484
2.09* 10-9
Substituting TA, α and β values for the equation (3.1), the final model can be written as
To(t) =0.53 e32.3P^-5 t0.9836 .
VII.
(3.6)
The above model (T vs t,P) can be graphically represented as shown in Figure 3-4.
90
80
70
Tornadoes
60
50
40
T vs t,P
30
Observed Tornadoes
20
10
0
1921 1931 1941 1951 1961 1971 1981 1991 2001 2011
Year (1921-2011)
Figure 3-4: Model Fit
36
The same procedure was applied to derive a set of models. A summary of the fitted models is
shown in Table 3-3. These models were compared and rank ordered to select the best model.
Among these seven models, models 6 and 7 were removed due to practical difficulties in
analyzing the model as commented in Table 3-3. The behaviours of the first five models were
evaluated focussing on several factors: model‟s comparability with the observed trend, behaviour
of the model with the increase of the parameters t and P. Models 1 and 2 have big deviations
from the observed tornado trend and they were ranked as least favourable models. The model 3
shows a trend that tally with the observed tornado trend. However, there is a rapid decrease at the
initial stage of the data set that puts the models to a lower ranking order. The models 4 and 5
show comparability with the observed tornado trend. The behaviour of the model 4 with the
increase of the parameters t and P are similar to the other lower ranked models. In contrast to
this, model 5 shows the opposite behaviour.
In this model analysis, the known fact is that the number of tornadoes is increasing with
population. At the same time, given the conditions such as climate change and global warming, it
is desirable that the model should behave in a way that the number of tornadoes is increasing
with time as well. Therefore, further analysis is required to select the most preferred model to
represent the tornado trend.
37
Table 3-3: Model Comparison
Model
1. To(t) = TA e
βt^k
Regression Equation
αP
-0.5
(1-1/e )
ln To(t) +1= 3.2587 +0.0164 (t +62t )
(p<0.05) model is a significant fit to the data.
R2
Comments
0.3046
To vs t,P is slightly increasing. However,
there is a big deviation from the observed
tornado trend.
To(t) = 14.92 e1.0168(t^-0.5)(1-1/e0.3667P)
Increasing the value of t leads to decrease
To(t) whereas increasing the value of P
leads to increase To(t).
Tornadoes
100
A. T observed Vs time
from Model 1
80
60
C. T observed vs t , P
actual from model 1
40
20
T observed - from
database
0
1
11
21
31
41
51
61
Year (1921-2011)
71
2. To(t) = TAtβ (1-1/eαP)
81
91
Rank =4
ln To(t)+1 = 4.0885+ +0.0196 (t-11.2 lnt)
(p<0.05) model is a significant fit to the data.
0.3016
To vs t,P is slightly increasing , However
there is a big deviation from the observed
tornado trend.
To(t) = 30.71 t -0.2195 (1-1/e0.438 P)
Increasing the value of t leads to decrease
To(t) whereas increasing the value of P
leads to increase To(t).
Tornadoes
100
80
T observed - from
database
60
40
A. T observed Vs time
from Model 2
20
0
1
11
21
31
41
51
61
71
81
91
C. T observed vs t, P
actual from Model 2
Year (1921-2011)
Rank=4
38
Model
3. To(t) =TA e
βt^k
P
α
To(t)= 1.5867 e-1.5528t^0.5
P10.4
Regression Equation
R2
Comments
ln To(t)= 4.7813
-1.5528( t0.5 -6.7ln (1+0.03t))
(p<0.05) model is a significant fit to the data.
0.3301
To vs t,P increasing. However, there is a
rapid decrease at the beginning.
Increasing the value of t leads to decrease
To(t) whereas increasing the value of P
leads to increase To(t).
400
Tornadoes
350
300
250
T observed from database
200
150
T observed vs time model 3
100
Rank=3
50
0
T observed vs t,P model 3
1
11
21
31
41
51
61
71
81
91
Year (1921-2011)
4. To(t) =TA tβ Pα
Ln To(t) =3.3396-0.5283(ln t-4.1ln (1+0.03t))
(p<0.05) model is a significant fit to the data.
To(t) =11.47 t-0.5283 P2.166
Tornadoes
( t value starts from 1)
0.3189
To vs t,P increasing
The model is comparable with the
observed tornado trend
Increasing the value of t leads to decrease
To(t) whereas increasing the value of P
leads to increase To(t).
90
80
70
60
50
40
30
20
10
0
T observed from database
T observed vs time Model 4
Rank=1 or 2
T observed vs t,P model 4
1
11
21
31
41
51
61
71
81
91
Year (1921-2011)
39
Model
5. To(t) =TA eβP^k tα
Regression Equation
R2
Comments
ln To(t)= -0.6302+0.9836 [ln t +4.2(1+0.03t)-5]
(p<0.05) model is a significant fit to the data.
0.3333
To vs t,P increasing.
The trend is comparable with the
observed tornado trend
To(t) =0.53 e32.3P^-5 t0.9836
Increasing the value of t leads to increase
To(t) whereas increasing the value of P
leads to decrease To(t).
90
80
70
60
50
40
30
20
10
0
A. T observed vs time model
5
B. T observed vs t, P model 5
Rank=1 or 2
Tornadoes
1
11
21
31
41
51
61
71
81
91
Year (1921-2011)
6. To(t) =TA eβt^k eαP^m
ln To(t)= K1+K2 [tk +d (1+0.03t)m]
(p<0.05) model is a significant fit to the data.
7. To(t) =TA +βt+αP
Need to vary 3 unknowns( k,d,m) to get
R2 max
Cannot visualize a contour map of R2 .
Collinearity of the two parameters t and
P is a problem in this multiple linear
regression model.
40
3.5 Final Model Selection
The next step of the analysis focusses on how the model was selected based on a comparative
analysis. It is important to note that these model behaviours are not fixed for all the time periods.
The same model can have different behaviours in different time periods. The model fit for the
models 4 and 5 was evaluated. Separate model fittings for each time period were also conducted
as indicated by the data range of the last columns of Table 3-4 and Table 3-5. The analysis
focussed on how one unit increase in population and time influence the each component of the
model (e.g. eβP^k and tα in model 5). The intention was to select a data range for the model to be
consistent in terms of the behaviour of To(t) based on the changes of t and P. An increase of a
component is shown by the „U‟ sign where as a decrement is shown by the „D‟ sign.
Table 3-4: Analysis for Model 4
Model 4
To(t) =TA tβ Pα
To(t) =11.47 t-0.5283 P2.17
tβ
β
Pα
α
R2
-0.5283
D
2.17
U
0.3190
To(t) =6.369 t-0.5192 P2.49
-0.5192
D
2.49
U
0.3508
3.5.1
0.1576
U
0.74
U
0.3764
To(t) =14.68t0.3994 P-0.36
0.3994
U
-0.36
D
0.2803
To(t) =96.05 t0.4871 P-1.66
0.4871
U
-1.66
D
0.1420
To(t) =30905.77 t0.9297 P-5.95
0.9297
U
-5.95
D
0.4382
To(t) =222498.3t0.2987 P-5.79
0.2987
U
-5.79
D
0.5442
To(t) =6.14 t0.1576 P0.74
41
Data
Range
Last
91Y
Last
80Y
Last
70Y
Last
60Y
Last
50Y
Last
40Y
Last
30Y
Table 3-5: Analysis for Model 5
Model 5
To(t) =TA eβP^k tα
To(t) =0.53 e32.3P^-5 t0.9836
β
eβP^k
α
tα
R2
32.3
D
0.9836
U
0.3337
To(t) =154941 e-16.71P^-1.2 t-1.402
-16.71
U
-1.402
D
0.4058
To(t) =1987.2e-22.6P^-2.1 t-0.796
-22.6
U
-0.796
D
0.4244
To(t) =7135.23e-179.8P^-3.6 t-1.23
-179.8
U
-1.23
D
0.3777
To(t) =169973.8e-3881.6P^-5.4 t-2.16
-3881.6
U
-2.16
D
0.3852
To(t) =20.05e-2.36*(10^-4)P^5.1 t0.539
-2.36*10-4
D
0.539
U
0.4931
To(t) =62.66e-1.424*(10^-7)P^9 t-0.0018
-1.424*10-7 D
-0.0018
D
0.5687
Data
Range
Last
91Y
Last
80Y
Last
70Y
Last
60Y
Last
50Y
Last
40Y
Last
30Y
In Table 3-4, a consistency of an upward trend for the time component and a downward trend for
the population component can be observed for the models with last 30, 40, 50 and 60 years of
data. In Table 3-5, a consistency of an upward trend for the population component and a
downward trend for the time component is seen for models with last 50 years of data to last 80
years of data.
Even if there is a consistent range in Model 4, the downward trend of the population component
does not tally with the known fact that the number of tornadoes observed are increasing with
population. On the basis of this discrepancy, the model 4 has to be ranked lower compared to the
model 5 that satisfies the above mentioned condition. In model 5, increasing the value of t leads
to decrease To(t). Further analysis is required to evaluate this behaviour.
42
Although the model 4 is slightly lower ranked compared to the model 5, it is interesting to see
that the model 4 with the last 70 years of data has upward trends for both population and time
components. This is a condition that we eagerly expect from a model selected based on this
analysis. To analyze this condition further, a separate analysis needs to be conducted.
The next step of the analysis is to see what is going on with the unexplained portion of the trend
over time, after deducting the corresponding values of model 5 from the observed number of
tornadoes. The same analysis was conducted for the model 4 with last 70 years of data in order
to compare the results of models 4 and 5 corresponding to the 70-year period. Time series
analysis methods were used to analyze these tornado time trends.
3.6 Time Series Analysis
Time series analysis methods can be used to extract meaningful statistics and other
characteristics of a data set collected at equally spaced time intervals. Here, the analysis was
conducted for the four models highlighted in Table 3-5 and for the model 4 with last 70 years of
data shown in Table 3-4. The key steps of this time series analysis process is explained using a
selected model with last 60 years of data as discussed below.
I.
The model output results corresponding to the number of tornadoes in each year was
calculated substituting the t, P values in each year. The output of the selected model
To(t) =7135.23e-179.8P^-3.6 t-1.23 (1952 - 2011) is shown in Figure 3-5 as T vs t, P. Here „T
observed‟, represents the total number of tornadoes observed in each year. The lower R2
value of 0.19 indicates the need to further improve the model to explain the variation of
the observed number of tornadoes.
43
Tornadoes
90
80
70
60
50
40
30
20
10
0
T vs t,P
T observed
1952
1962
1972
1982
1992
2002
Year (1952-2011)
Figure 3-5: Model Output Results
II.
The difference [T observed - T vs t,P] was obtained and the trend for this dataset was
obtained using polynomial regression in ITSM2000 software (Brockwell and Davis,
2002). The trend curve is shown in the following Figure 3-6.
40.
[T observed - T vs t,P]
30.
20.
10.
0.
-10.
-20.
0
10
20
30
Year (1952-2011)
Figure 3-6: Polynomial for [T observed - T vs t,P]
44
40
50
60
The polynomial trend M(t) of the dataset of [T observed - T vs t,P] can be written as
M(t) =2.67 - 1.54 t + 0.098t2 - 0.0013 t3
(3.7)
which has a R2 value of 0.301.
III.
It is important to check whether the polynomial is a good representative of the trend.
After fitting the polynomial trend, the residual series can be obtained as shown in Figure
3-7.
Rescaled Residuals
3.
2.
Residuals
1.
0.
-1.
-2.
0
10
20
30
40
50
60
Year (1952-2011)
Figure 3-7: Residual Series
Figure 3-8 shows the corresponding sample autocorrelation function at lags 0,1,….40.
Approximately 95% of the sample autocorrelations fall between the bounds ±
1.96
𝑛
where
n is the sample size. (Since 1.96 is the 0.975 quartile of the standard normal distribution.)
Therefore, roughly 40×0.05 =2 values would be expected to fall outside the bounds as
shown in broken lines. The sample autocorrelation function (ACF) of residuals satisfies
45
this condition indicating the randomness of the residual series and there is no cause to
reject the model on the basis of these autocorrelations.
Residual ACF
ACF
1.00
Residual PACF
1.00
.80
.80
.60
.60
.40
.40
.20
.20
.00
.00
-.20
-.20
-.40
-.40
-.60
-.60
-.80
-.80
-1.00
-1.00
0
5
10
15
20
25
30
35
40
0
Lag
5
10
15
20
25
30
35
40
Lag
Figure 3-8: Sample Autocorrelation Function
IV.
The next step is to add the unexplained portion of the model represented by the
polynomial trend to the original model so that the model is a better representation of the
trend for observed number of tornadoes. The resulting graph is shown in Figure 3-9. The
R2 value of 0.576 shows how a greater amount of variation is explained by the overall
trend.
46
Tornadoes
90
80
70
60
50
40
30
20
10
0
T vs t,P
T observed
Overall Trend
1952
1962
1972
1982
1992
2002
Year (1952-2011)
Figure 3-9: Overall Trend for the Observed Number of Tornadoes
The same approach was used in determining trends for [T actual - T vs t,P] and obtaining the
overall trends in other periods as well. The summary of the trend curves are shown in Table 3-6.
The downward directions of all the overall trend curves provide evidence that number of
tornadoes are decreasing in recent decades.
Comparison of overall trends for the last 70 years of data in both model 4 and 5 indicates the
presence of a cyclic nature for the observed tornado frequency with a period of around 65 years.
This wave pattern shows the presence of a decreasing trend of the number of tornadoes observed
in last two decades. This downward trend may be preceded by an upward trend of the number of
tornadoes in coming years. However, there is no way of confirming it in the absence of data.
Temporal extrapolation to assess the future tornado frequency is also not desirable as there are
number of uncertainties involved with the observed trend. It is noteworthy that the intention of
this analysis was to provide a description of the data set. The analysis did not explicitly focus on
understanding the mechanism generating this overall trend pattern.
47
Table 3-6: Summary of Polynomial Trends for [T observed - T vs t,P]
Model 5: To(t) =TA eβP^k tα
I.
To(t) =154941 e-16.71P^-1.2 t-1.402 (Last 80 years)
M(t) = - 83.65 + 4.623 t - 0.05 t2
0.
-100.
-200.
-300.
-400.
0
10
20
30
40
50
60
70
80
120
100
Tornadoes
80
60
Overall Trend
40
T vs t,P
20
T observed
0
-20
-40
1934
1944
1954
1964
1974
1984
Year (1932-2011)
(After removing the edge effect in 1932 and 1933)
48
1994
2004
II.
To(t) =1987.2e-22.6P^-2.1 t-0.796 ( Last 70 years)
M(t) = 0.1935 – 1.267t + 0.074t2 - 0.00087t3
40.
30.
20.
10.
0.
-10.
-20.
-30.
-40.
0
10
20
30
40
50
60
70
90
80
Tornadoes
70
60
50
T vs t, P
40
30
T actual
20
Overall Trend
10
0
1942
1952
1962
1972
1982
Year (1942-2011)
R2 of the overall trend = 0.616
49
1992
2002
To(t) =7135.23e-179.8P^-3.6 t-1.23
(Last 60 years)
M(t) = 2.673 - 1.538 t + 0.098t2 - 0.0013 t3
40.
30.
20.
10.
0.
-10.
-20.
0
10
20
30
40
50
60
90
80
70
60
Tornadoes
III.
50
T vs t,P
40
T observed
30
Overall Trend
20
10
0
1952
1962
1972
1982
1992
Year (1952-2011)
50
2002
IV.
To(t) =169973.8e-3881.6P^-5.4 t-2.16 (Last 50 years)
M(t) = - 61.71 + 8.117 t -0.279 t2 +0.0028t3
60.
40.
20.
0.
-20.
-40.
-60.
-80.
-100.
-120.
-140.
-160.
0
10
20
30
40
50
90
80
70
Tornadoes
60
50
T vs t,P
40
T actual
30
Overall Trend
20
10
0
-10 1963
1973
1983
1993
Year (1962-2011)
(After removing the edge effect in 1962)
51
2003
Model 4: To(t) =TA tβ Pα
I.
To(t) =6.14 t0.1576 P0.74 (Last 70 years)
M(t) = 10.247 - 2.468t + 0.115 t2 - 0.001 t3
50.
40.
30.
20.
10.
0.
-10.
-20.
-30.
0
10
20
30
40
50
60
70
90
80
Tornadoes
70
60
50
T vs t, P
40
T observed
30
Overall Trend
20
10
0
1942
1952
1962
1972
1982
Year(1942-2011)
R2 of the overall trend = 0.591
52
1992
2002
3.7 Discussion
Even if the analysis does not focus on understanding the underlying factors behind this trend,
some possible explanations can be discussed based on visual evidence.
It is interesting to note that there are a large number of tornadoes reported in the late 80‟s. It is
worthy to note that since 1984 the weather centres started hiring students who are responsible for
collecting reports from people, calling around in the vicinity of storms or doing follow-up calls.
This may result in more tornadoes being reported in the 80‟s. Further, and most importantly, the
1987 Edmonton tornado became an eye opener to weather forecasters as well as the public in the
Prairie region. Perhaps, after this disaster, people had a heightened alert for tornadoes and
therefore more were reported, or anything even resembling a tornado may have been reported
because people were so hyper-aware (Figure 3-10). The misrecognition of non-tornadic severe
weather events as tornadoes could have influenced the higher number of tornadoes to be reported
(Personal Communication, 2012).
Figure 3-10: Tornado Trend in Recent Decades
53
The Edmonton tornado also triggered the implementation of a Doppler radar network in Canada
in early 90‟s. Even if more number of events can be detected by increased number of eyes
(population increase), the advancement of the Doppler radar technology can distinguish
tornadoes from other non-tornadic events based on a scientific approach. Forecasters can sort the
actual number of tornadoes observed in each year from all the events reported. Further, it is
possible that people are more educated about tornadoes and therefore don‟t fear them as much
and may not report everything they see. They can be plausible explanations for the downward
tornado time trend observed in the recent decades.
54
Chapter Four: Analysis of the Tornado Detection, Warning and Communication (TDWC)
System in Canada
This chapter intends to apply the network modelling approach for critically analyzing the TDWC
system recognizing the sequence of activities from the point a tornado is detected to the point
where the public receive warnings. The network modelling approach in the Decision Support
Simulation System (DSSS) template in Simphony Legacy software was used to model and
simulate the activity network developed taking the City of Calgary as a case study.
4.1 Related Literature
Investigation of the pre-touchdown phase of a tornado is well covered in literature in the US.
Golden and Adams (2000) described the state of knowledge for prediction, forecast, warning and
response actions. Their study highlighted the warning partnership of the National Weather
Service (NWS), local media, forecasters and local emergency managers, as well as the NWS‟s
approach to the warning process. Moreover, it recognized the needs in research, policy and
programs in order to improve the warning communication and coordination process. Collins and
Kapucu (2008) investigated how local governments should provide early warning for an
impending tornado danger. League et al. (2010) discussed the communication process that
emergency managers employ during a tornado outbreak, focussing on four main topics, namely:
acquisition, interpretation and verification of weather information and communication of
warnings to the public. According to Schumacher et al. (2010, p.1413), “Understanding the flow
of information among decision makers and the public, and how warnings are interpreted, are the
key first steps toward maximizing the effectiveness of warnings”. Their study highlighted the
55
importance of performing researches that integrate meteorology, climatology and the social
sciences.
In Canada, although there are studies that have discussed the climatology related issues, no work
has appraised or evaluated the process from detection to warning communication which is an
important aspect in tornado disaster mitigation. The main focus of this chapter is to study,
analyze, model, simulate and propose improvements to plans and systems to mitigate the impacts
of tornadoes in Canada considering the process from tornado prediction (or detection) to warning
communication.
4.2 Research Questions
This section intends to answer the following questions.

How are tornadoes predicted and verified, or detected?

How are tornado warnings communicated to the public?

What are the roles of organizations and different levels of government in disseminating
tornado warning messages to the public?

What are the activities within the tornado detection, warning decision-making,
communication and evacuation process and their interrelationships?

What are the durations of the activities in this process?

How can this process be modelled?

What is the overall time consumption of the TDWC network?
56
The network modelling approach was used to model and simulate the network. The overall time
consumption obtained through this analysis is a good indicator for the viability of the existing
TDWC system.
4.3 Network Modelling Approach
Application of network modelling in emergency situations is an emerging approach that can be
applied to model and analyze disaster scenarios. Such an approach for tsunami mitigation
network analysis has been reported by Fernando et al. (2008), Ruwanpura et al. (2009) and
Wickramaratne (2010). There are vital differences between tsunami and tornado disasters with
respect to factors such as origin, nature, impact area, warning lead times, and evacuation
strategies. Although, the approach for network modelling and simulations are similar, the
development of the TDWC network is solely based on the analysis of the pre-disaster stage of a
tornado in the considered region and it is expected to be widely different from other applications.
The „network‟ represents the sequence of activities with their interrelationships. The
development of such a network gives an overall picture about the TDWC system. Unlike general
activity networks in construction management, the overall time consumption from the start of the
first activity to the finish of the last activity has a very short duration which can generally be
represented in minutes. The duration of the whole process is determined by the time
consumption of individual activities. Using the network modelling approach, the activity
network can be modelled and simulated to get the total time consumption.
57
In reality, „one set of variables in – one answer out‟ is not appropriate for determining total time
consumption. The activities are subjected to a wide variety of fluctuations and interruptions.
Therefore, activity durations become subject to random variations. This „stochastic‟ nature
requires representation of the time consumption of each activity using a distribution function.
Decision Support Simulation System (DSSS) template in Simphony software
(Figure 4-1)
provides an ideal environment to construct the network illustrating the sequence and
interrelationships of activities that have stochastic time durations (Moussa, 2007,
Wickramaratne, 2010).
Figure 4-1: DSSS Template –Model Window
This simulation tool utilizes a number of modelling elements (as shown below the term SMDSS in the Figure 4-1) and logical relationships to construct the network. The modelling element
58
„link‟ is capable of representing various activity relationships such as SS (Start to Start), Finish
to Start (FS) and Finish to Finish (FF). In this TDWC network, FS activity relationship was used,
with no time lag from any activity to its successor. The „OR‟ relationship which represents the
realization of any of the preceding activities to realize succeeding activity, and the „AND‟
relationship that requires realization of all preceding activities for a succeeding activity to be
realized were used as logical relationships when necessary. Furthermore, a modelling element
called „Hammock‟ was used to identify the time consumption focussing on a desired part of the
network. Simulating the network using Monte Carlo Simulation methods provided the output
variable: the overall time consumption.
4.3.1 Monte Carlo Simulation
Simphony software uses the monte carlo algorithem to simulate the network. According to
Mooney (1997, p.3), “the principle behind Monte Carlo Simulation is that the behaviour of a
statistic in random samples can be assessed by the empirical process of actually drawing lots of
random samples and observing this behaviour”. The system generates a large number of random
numbers in the range of zero to unity and assigns them in the relevant probability distribution of
each activity. The resulting random variates are used to calculate the overall time consumption of
the network. Due to the presence of stochastic time durations for activities, the network
simulation does not result in the same time consumption value in each run. Therefore, the overall
time consumption needs to be obtained as a Probability Density Function (PDF) or a Cumulative
Distribution Function (CDF) after simulating the network with a large number of runs (generally
1000 or more). The probability that the process can be completed within a given time or less can
be obtained from the CDF curve.
59
4.3.2 Triangular Distribution
It was assumed in this study that the activity durations follow Triangular probability distributions
(Figure 4-2). The triangular distribution has been widely used in construction management
applications (Chau, 1995, Lee and Shi, 2004) to analyze activity networks.
Figure 4-2: Triangular Distribution
In the absence of detailed data sets, the triangular distribution can effectively be used to describe
the variation of time estimates. The durations (in minutes) associated with individual activities
can be defined by the three-point estimate: the minimum time (the fastest possible response), the
maximum time (the slowest possible response) and the mode (most likely response time). The
rest of this chapter presents the analysis of the TDWC system and the application of the network
modelling approach to the activity network developed taking the City of Calgary as a case study.
4.4 Collaborating Partners
4.4.1 Environment Canada –Storm Prediction Centres
Warnings, watches and special statements for severe weather provide public and emergency
services with the level of preparedness required to manage an actual or pending severe weather
emergency. Environment Canada (EC) is the official source of weather warnings in Canada.
60
Through the Storm Prediction Centres (SPC), EC monitors weather conditions and provides
weather forecasts and severe weather warnings (Figure 4-3 and Figure 4-4). The Canadian
Meteorological Centre (CMC) also provides forecast guidance to the SPCs by preparing weather
and atmospheric data used in local forecasts. Presently, the CMC uses new versions of NWP
models, such as GEM MesoGlobal, in predicting severe weather. High impact weather that can
result in significant impacts on safety, property and/or socioeconomic activity are recognized
based on model output results (Sills, 2009).
Figure 4-3: An Official Tornado Warning Bulletin
61
Figure 4-4: Public Warnings in EC Website
(Warning area is shown in red colour.)
Prairie and Arctic Storm Prediction Centre (PASPC) is the responsible authority for providing
round-the-clock forecast support to Canadian Prairie Provinces and Arctic region. By
continuously monitoring thunderstorms, PASPC issues bulletins conveying the status of the
severe weather condition time to time. Within a storm, numerous bulletins can be in effect based
on the storm monitoring results. When there is a severe weather warning in effect, forecasters
62
look for conditions that are favourable for tornado formation. Especially the severe
thunderstorms are closely monitored and tornado watches are issued when strong supercells are
detected. Tornado warnings are issued when it is likely that a tornado would develop soon in the
area, when a tornado is occurring in a nearby area and may soon move into the area, or when a
tornado is already occurring in the area. In many cases, watches and warnings are preceded by
bulletins issued for severe weather watches and warnings. Sometimes, the sudden appearance of
a tornado or report of a tornado leads to bypass the watch stage and issue a warning. There are
„public forecast regions‟ which consist of groups of municipalities for the purpose of issuing
warnings. These regions also can be sub-divided into „summer severe weather warning regions‟.
Watches and warnings are issued for the relevant region only. Generally, warnings have to be
issued for a large area although the impacts are localized.
4.4.2 Spotter Network
Storm spotters are a strong partner group that support the process of tornado warning and
communication. They provide real-time reports of tornado sightings that occur where they are.
CANWARN (Canadian Weather Amateur Radio Network) is the spotter network in Canada that
consists of personnel trained to recognize severe weather. According to Central Alberta
Armature Radio Club, “CANWARN is not about storm chasing, it is about putting trained eyes
at the local level to confirm what is happening under severe weather and communicating that
information to the Meteorological Service of Canada (MSC)” (CANWARN, 2011, para.2). Even
if the PASPC can issue warnings based on radar and satellite indications, spotter reports provide
strong evidence about what is happening on the ground.
63
4.4.3 Alberta Emergency Alert
Alberta Emergency Management Agency (AEMA) is the provincial partner responsible and
accountable for emergency management under the Government of Alberta. It launched a digital
public warning system called Alberta Emergency Alert (AEA) in June 2011 to alert the public to
hazards, emergencies or disasters in the province. This is actually the upgraded version of the
Emergency Public Warning System (EPWS) which was established by the Government of
Alberta after the major F-4 tornado-1987 in Edmonton.
Figure 4-5: AEA -Critical Alert for a Possible Tornado
Source: (AEA, 2011)
There are two types of AEAs that can be sent to the public, namely; critical alert and information
alert. A Critical Alert is sent when there is an imminent life threatening danger. An Information
Alert provides less critical information to the public to help them to prepare for an emergency.
64
EC is an authorized user of the AEA system and PASPC can issue alerts to give tornado warning
or watch information. “PASPC has three standard forms by which the AEA can be activated;
these include Scripts for tornadoes, dangerous supercells and possible tornadoes (from rotating
thunderstorm)” (PASPC, 2011, p.5). Authorized provincial and municipal level emergency
management officials also can activate the AEA to broadcast warning messages quickly to the
public. It is the responsibility of the AEA users that reliable information is disseminated through
this alert system. AEA has made a step forward from a traditional stage by designating a warning
area that can reduce the location uncertainty of the threat.
A tornado warning is a critical alert whereas a tornado watch can be a critical alert or an
information alert depending on the severe weather situation. The AEA system disseminates these
alerts through various media such as internet, RSS feed, radio, television and social media.
Turning to social media such as Facebook and Twitter to disseminate alerts provides a new
mobile version of the warning communication through internet-enabled smart phones. People on
roads can also receive these alerts so that they can be warned about imminent dangers. Other
than radio broadcasts this is the main dissemination method in which road users can be given
official alerts and warnings.
4.4.4 Broadcasting Media
PASPC has several options to broadcast tornado warnings to the public. Its „Weatheradio‟
service sends out alerts to the relevant area notifying that a warning is being issued. In addition,
radio and television networks, mobile phone alerts, EC website, private meteorological
companies as well as social media broadcast tornado watch or warning information directly to
65
the public. There is an alert system called NAAD (National Alert Aggregation and
Dissemination System) that collects and makes available for distribution public safety alerts from
municipal, provincial, territorial and federal governments and agencies. Environment Canada can
access the NAAD system to make the tornado warnings available to distribution media. This free
information system gives easy access to local media networks so that they can broadcast the
warning messages (EC, 2013c). Furthermore, warning information from Environment Canada is
available through Weather Media Site which is a free weather web service dedicated to the
media. When there is a tornado warning, local radio and television stations should take
immediate action to air the warning by interrupting regular programming. Private meteorological
companies send out tornado watches and warnings messages obtained from PASPC. The
province-wide Alberta emergency alert (AEA) system is also used by PASPC to warn the public
in the Province of Alberta.
4.4.5 Calgary Emergency Management Agency (CEMA)
In Calgary, CEMA receives tornado sighting information directly from the public via 911. When
there is an imminent danger of a tornado or a funnel cloud appearing in the Calgary area,
residents can call 911 to inform it to the emergency officials. Authorized AEA users in the
CEMA can evaluate and validate information and activate the AEA to warn the public in the
threat area. If necessary, CEMA informs other emergency respondents such as local police
stations and Emergency Medical Service (EMS) regarding this threat and activates its action plan
to respond to the emergency.
66
4.4.6 Police and 911
Emergency Services play a major role in ensuring the safety of the public during disaster
situations. They provide warning and cautioning information to the public. When an emergency
call about a tornado sighting in the locality is received by the 911 public safety communication
system, CEMA validates the threat and the police receive the message immediately. Upon
receiving timely information, emergency services can be ready to respond to possible
emergency. As a responsible authority for public safety, the police also give sighting information
to the local emergency management officials when a tornado is observed by them.
4.4.7 Public
Though there are various warning communication methods, people cannot entirely depend on
official warnings. Sometimes, the relevant agencies are unable to issue warnings until a tornado
touches down. In addition to the systematic ways of monitoring and detection of tornadoes,
people can identify cues regarding tornadoes in advance. In the absence of an official warning,
the best warning is based on what people can see by themselves in the surrounding environment.
People should remain alert to severe weather watches and warnings as well as environmental
cues of approaching tornadoes. If people can safely do so, they can inform the 911 about what is
happening on the ground. This information helps to issue a tornado warning immediately so that
all the people in the area are warned about the incoming disaster.
The above discussion of the role and responsibility of each level of government and other
collaborating partners in communicating tornado warnings to the public provides the basis to
develop a network showing the sequence of activities, their interrelationships and the time
67
consumption. The next section describes how the network was developed based on the data
obtained from various sources and simulations were carried out in the DSSS template of the
Simphony software.
4.5 Activity Network Development
Data required for developing the activity network (Durage et al., 2013a) representing the tornado
detection, decision-making, and communication was collected using various methods such as
direct discussions, telephone and email conversations with collaborating partners and secondary
information sources.
A series of discussions and brainstorming sessions at the CEMA provided useful information
about their involvement in communicating tornado warnings to the public. A poster of the draft
activity network with assumed time distributions was presented to the CEMA and EMS officials
in June 2011. This presentation and the following discussion provided useful information about
their involvement. Furthermore, the importance of social media in emergency warning
communication was emerged during this discussion. Time distribution of each activity was
collected via a consensus based on original personal values of participants represented as
minimum, maximum, mode and mean.
The AEA workshop held in Calgary in July 2011 also provided useful information on how an
alert can be issued to communicate a tornado threat to the public. Furthermore, content analysis
of journal articles, websites of collaborating partners and online news articles also provided
important information to develop this network. A site visit to the damaged area immediately
68
following the F-0 tornado touchdown in Calgary on 13th July 2011 was helpful to get real
information regarding the TDWC system in Calgary. Based on the comments made by local
residents, some modifications to the current network were recognized as necessary.
It is noteworthy that the City of Calgary has only experienced minor tornado touchdowns in its
history. The network was developed based on the information from past tornado experiences as
well as other possible systems and procedures that will work. The reliability of several links is
subject to some level of uncertainty; however, they represent what the collaborating partners
ought to do when there is a need to communicate a tornado warning to residents.
The sequence of activities and their interrelationships were determined and the Canada-AlbertaCalgary-Household tornado detection and warning network was developed (Figure 4-6). Time
durations of the activities from tornado detection to warning communication phase was included
in the network using a triangular distribution. The parameters of the activity durations that follow
triangular distributions are shown as: (minimum, mode, maximum) in each box. In the network,
the activity name is shown in the upper part of the box and the collaborating partner of that
activity and the duration is shown at the bottom of the box. Activities have been colour coded in
order to recognize the information flow by each collaborating partner. Issuance of a watch or a
warning is a milestone of this information flow as indicated by broken lines A-E. The possibility
for tornadoes is checked when active severe weather is occurring; thus the activity network that
ultimately leads to communicate warnings to the public starts at this stage. It is assumed that a
severe thunderstorm warning and a tornado watch have already been issued and active severe
weather is taking place.
69
Figure 4-6: Tornado Detection, Warning and Communication Network
70
4.6 Network Simulation
The TDWC network was modelled with key contributions from EC-SPC, Spotters, local media,
and AEA while private meteorologists and the social media play supportive roles. Simulation of
the network gives probability estimates of a successful completion of detection, warning and
warning communication in a given time. Delay points as well as faster links in the information
flow were recognized and suggestions were provided to improve its quality and the timeliness.
4.6.1 Overall Network
Based on the simulation results, cumulative probabilities associated with time duration can be
recognized (Figure 4-7).
Figure 4-7: CDF Curve for the TDWC Network
According to the CDF curve, there is a 50% chance that tornado detection, warning and
communication can be completed within 20.5 minutes or less. The maximum time predicted
71
through simulation is about 28 minutes. The earliest time that a detection, warning and
communication can be completed is around 14 minutes from the triggering point of severe
weather that lead to a tornado occurrence. Furthermore, criticality index reports of the activities
reveal that the SPC based activities are more critical. This indicates the importance of paying
attention on these activities to avoid any delays in the overall duration. It is important to note
that the numerical figures obtained from monte carlo simulation are based on the time
consumption used for individual activity durations.
This TDWC network is a complex network that consists of a large number of interactions among
collaborating partners. To analyze the network with clarity, simulations were also conducted
separately considering the activity sequence as follows.
1. The sequence of activities focussing on the SPC‟s role in issuing warnings and
disseminating them to the public
2. The sequence of activities by local level partners such as public, emergency services
(Police and 911), AEA users and CEMA in communicating a tornado warning to the
public based on the environmental cues observed
These two networks were discussed and compared focussing on factors such as the reliability of
each branch or each activity, the chance of a branch failing and how to make the network better.
4.6.2 Network with SPC Activity Sequence
The SPC‟s activity sequence starts with the activity “visualization, analysis and detection of
supercells”. Tornado warning issuance is the key activity resulting from the decision making
based on information received through radar and satellite systems as well as ground truth
72
information from spotters. Imminent threat of a tornado is recognized when it is detected by the
radar, satellite system or the reliable information is received from experienced spotters as shown
by the „OR‟ relationship for the activity „recognition of tornadoes‟. It is enough to realize at least
one of these two paths to make decision and proceed to the next level issuing a tornado warning.
As shown in Figure 4-8, the average time consumption for the activity sequence is 22.75 minutes
which is slightly higher than the time consumption of the overall network.
Figure 4-8: CDF Curve – Network with SPC Activity Sequence
The detection of a supercell hints SPC forecasters to recognize the threat. Moreover, the
experienced spotters can detect the tornado based on environmental cues. This ground truth
information further support SPC forecasters‟ decision making process. Nevertheless, local level
severe weather spotting is not very active in the Calgary area as shown by broken arrows in the
73
TDWC network (Figure 4-6). Even if there are volunteers who provide reports of funnel clouds
occasionally, there is not an organized group to act during a severe weather event. It is important
to strengthen the local level detection by spotter networks in order to link the radar, satellite
observations with local level information. This can improve the reliability and efficiency of the
warning decision making process. Furthermore, sufficiency and efficiency of the technology to
detect tornadoes should also be evaluated. Communicating tornado warnings to the public via
weathradio is also a low reliable link of the network. SPC needs to pay attention on these
imperfect links to make the TDWC system more efficient.
4.6.3 Network with Local Level Activity Sequence
Appearance of funnel clouds is the one of the main cues of tornado formation. Sighting of funnel
clouds by local residents, emergency services as well as AEA representatives can lead the
information flow toward an AEA-critical alert of a tornado. Simulation of this information flow
results in the CDF curve as shown in Figure 4-9. This CDF curve has slightly shifted rightward
from the CDF curve obtained for the SPC activity sequence. It is reasonable due to the fact that
the local level detection of a tornado from active severe weather stage takes more time than
advanced detections by radar and satellite based technology. „Hammocks‟ that placed in the
network model also revealed that local level detection to the AEA activation point consumes
more time (18 minutes) than radar detection to the warning issuance point at SPC (15 minutes).
Even if a slightly higher time is consumed for the TDWC network, much of this time is allocated
for tornado detection and verification. Once a tornado threat is verified by CEMA or other
authorized AEA user, the message can be disseminated immediately to the public through AEA
system.
74
Figure 4-9: CDF Curve with Local Level Activity Sequence
It is important to understand that the SPC detection and early warning to the public is not always
feasible due to the chaotic nature of supercell development. A sudden appearance of a tornadic
rotation may not be detected by the radar system. Furthermore, with the isolated nature of
tornado developments that cannot be detected easily, not all the events are preceded by early
warnings. In this regard, it is absolutely imperative to strengthen the local level detection of
tornadoes. This can be done by actions such as improving the public awareness on tornado
safety, utilizing properly trained spotters and developing close interactions among the local
emergency management officials, other AEA users and the public.
4.7 Analysis of the Total Time for Warning, Communication and Evacuation
Once a warning is issued, how much time it takes for households and drivers to complete the
evacuation is the key parameter analyzed in this section. This time can be compared with the
75
tornado warning lead time (warning issuance to the touchdown point) to analyze the performance
of the SPC in giving advance warnings to the public.
4.7.1 Curve Fitting
Curve fitting procedures required for this analysis uses the software called “EasyFit” that mainly
utilizes Maximum likelihood estimation methods (Wickramaratne, 2010) for fitting different
distribution functions (Mathwave, 2013). The data sets obtained from interviews or surveys often
resemble PDFs with a higher number of parameters. The types of distributions and there PDFs
used in this analysis are shown in Table 4-1.
Table 4-1: Probability Density Functions
Distribution
Parameters
Probability Density Function
Normal
σ, µ
1 x−µ
exp − 2
σ
𝑓(𝑥) =
σ 2π
Burr
k,α,β,γ
x − γ α−1
β
𝑓(𝑥) =
x − γ α k+1
β 1+
β
Generalized
k, σ, µ
Extreme Value
2
αk
1
𝑓 𝑥 = exp −(1 + kz)−1/k (1 + kz)−1−1/k
σ
1
𝑓 𝑥 = exp −z − exp⁡
(−z)
k=0
σ
Frechet
α,β,γ
α β
f(x) =
β x−γ
76
α+1
β
exp −
x−γ
k≠0
α
It can be seen that different probability distributions with higher number of parameters such as
Burr, Generalized extreme value and Frechet are resulting from these datasets when a continuous
data domain is assumed.
The total time consumption from the point of warning issuance to the evacuation completion can
be divided into two main parts as warning issuance to the warning receipt point and the warning
receipt point to the evacuation completion point. The time distributions of these two parts were
analyzed to get a combined probability distribution function for the total consumption.
4.7.2 Time from the Warning Issuance to the Warning Receipt Point
The required time distribution begins from the activity „Tornado Warning Issued‟ to the activity
„Public Receive Warning‟ in Figure 4-6. However, as there are linkages that connect only with
the last activity, it is difficult to analyze and get the time distribution using „Hammock‟
placement. To overcome this problem, the network with the SPC activity sequence was
considered.
The network was simulated in Simphony after placing a hammock between the above two
activities and the resulting PDF and CDF was obtained to further analyze the distribution. The
PDF was reconstructed in EasyFit. It was revealed that the time is normally distributed (Figure
4-10) function with a mean of 7.94 and a standard deviation of 1.178.
77
Probability Density Function
0.32
0.28
Probability
0.24
0.2
0.16
0.12
0.08
0.04
0
4
6
8
10
12
Time (In Minutes)
Normal (1.1785; 7.9469)
Figure 4-10: Time distribution from Warning Issuance to Warning Receipt
The PDF is based on the analysis of the general activity network and it can be used as the time
distribution for warning communication phase for both household and driver scenarios as will
discussed later in this analysis.
4.7.3 Time from the Warning Receipt Point to the Evacuation Completion Point
This time consumption needs to be discussed separately for household and drivers. The data
required for fitting distributions were obtained from the stated preference survey that will be
discussed in Section 6-9. The corresponding PDF curves for the CDFs shown in Figure 6-4 (in
Section 6-9) were obtained as shown in Figures 4-11 and 4-12. The distributions for the total
time required for households and drivers can be obtained by combining each distribution
separately with the normal distribution shown in Figure 4-10. As the distributions are
78
independent of each other, the set of values obtained by simple adding random numbers
generated for each distribution represents the resulting PDF.
Probability Density Function
0.16
0.14
Probability
0.12
0.1
0.08
0.06
0.04
0.02
0
0
5
10
15
20
25
30
Time (In Minutes)
Burr (6.1194; 1.1754; 22.59)
Figure 4-11: PDF for Household Evacuation Time
Probability Density Function
0.16
0.14
Probability
0.12
0.1
0.08
0.06
0.04
0.02
0
0
5
10
15
Time (In Minutes)
Gen. Extreme Value (0.22644; 2.3991; 2.4421)
Figure 4-12: PDF for Driver Evacuation Time
79
35
The resulting PDF and CDF for household scenario are shown in Figure 4-13 and 4-14.
Probability Density Function
0.13
0.12
0.11
0.1
Probability
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
10
20
30
40
Time (In Minutes)
Burr (0.20519; 55.712; 48.818; -39.996)
Figure 4-13: PDF for Household Scenario
Cumulative Distribution Function
1
0.9
0.8
0.7
F(x)
0.6
0.5
0.4
0.3
0.2
0.1
0
10
20
30
x
Burr (0.20519; 55.712; 48.818; -39.996)
Figure 4-14: CDF for Household Scenario
80
40
The PDF and CDF for driving scenario are shown in Figure 4-15 and 4-16.
Probability Density Function
0.13
0.12
0.11
0.1
Probability
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
5
10
15
20
25
30
25
30
Time (In Minutes)
Frechet (6.6377; 18.946; -8.6212)
Figure 4-15: PDF for Driving Scenario
Cumulative Distribution Function
1
0.9
0.8
0.7
F(x)
0.6
0.5
0.4
0.3
0.2
0.1
0
5
10
15
20
x
Frechet (6.6377; 18.946; -8.6212)
Figure 4-16: CDF for Driving Scenario
81
Figures 4-14 and 4-16 provide important information to analyze the results comparatively with
the SPC‟s warning lead time of around 10 minutes. According to Figure 4-14, there is around
25% chance that the evacuation can be completed by a household within 10 minutes from the
warning issuance point. Increasing the time by 5 minutes gives a 75% chance for a household to
complete evacuation. Figure 4-15 says that there is nearly a 33% chance that a driver can
complete evacuation within 10 minutes from the warning issuance point. Further, if the lead time
is 15 minutes, there is an 80% percent confidence that a driver can complete evacuation prior to a
tornado touchdown. These estimates are based on the intended responses from the individuals
and they are situational dependent. For example, during a peak traffic situation, the total time
taken by a driver to complete evacuation depends on other drivers‟ behaviour. However, these
estimates indicate the importance of improving the warning lead time.
4.8 Behaviour of the Network under Different Distributions for Activity Duration
In the previous section, triangular distribution was used to represent activity durations.
Triangular distribution which is an approximate distribution based on a three-point estimate of
the endpoints is used in many applications to simulate activity networks. Here, the overall
TDWC network was simulated separately fixing time durations as a normal distribution so that
the resulting CDF could be compared with that of the triangular distribution. Mean and variance
are the parameters required to fit a normal distribution for the activity duration. Mean (µN) was
taken as the average value of the minimum (a), mode (m) and maximum (b) time estimates used
for the triangular distribution as follows
µ𝑁 =
(a+m+b)
3
.
Standard Deviation (σN) for the triangular distribution was calculated as
82
(4.1)
σ𝑁 =
(𝑎 2 +𝑚 2 +𝑏 2 –𝑎𝑚 −𝑎𝑏 −𝑚𝑏 )
(4.2)
18
and it was used as the standard deviation when fitting a normal distribution. The two CDF
graphs were overlapped to illustrate the comparison (Figure 4-17).
The maximum gap of the two graphs that can be observed at 10% probability
is around one
minute. As shown by green arrows, the difference between the average time consumptions in
these CDFs is less than one minute. This indicates only a minor variation even if a more general
normal distribution is used. This shows that the triangular distribution can safely be used to
interpret activity durations. Triangular distribution provides the opportunity to collect data easily
by directly asking questions about minimum, most likely and maximum time estimates.
However, a normal distribution cannot be fitted from the values directly taken from the public.
Even if the mean value of an activity can be obtained the variance (or the standard deviation)
cannot be obtained directly.
Figure 4-17: Variation with Different Distributions
83
Comparison of the two CDFs of triangular and normal distribution based networks shows that
the triangular distribution can safely be used to interpret activity durations.
4.9 Conclusion
This chapter has presented the analysis of the TDWC system in Calgary, Canada. It is important
to note that the time consumption and activity scenario of this TDWC network has been
developed linking the possible activity sequence and their time consumption obtained from
various sources. This activity sequence and time consumption can have variation from one
tornado warning situation to the other. In this regard, the network is flexible and can have
variations based on the availability of information. To better understand the present status of the
Canadian TDWC system, it was compared with the US system as discussed in chapter 5.
84
Chapter Five: Comparison of the Canadian and US Tornado Detection and Warning and
Communication Systems
5.1 Introduction
The status of the present TDWC system in Canada can be better understood when compared with
the well-established US system. The objective in this chapter is the comparison of Canada‟s
system with the US TDWC system, paying particular attention to activities in the pre-touchdown
phase. The US TDWC system was analyzed, and a separate network was developed. The
comparative analysis was carried out, focussing on key issues such as prediction/detection
capabilities, warning provision, emergency preparedness, institutional partnerships, warning
area, warning dissemination methods and the importance of spotter networks. Based on the
comparison results, the key areas that need to be considered in improving the TDWC system in
Canada were recognized.
5.2 US Tornado Detection, Warning and Communication System
The National Weather Service (NWS) of the National Oceanic and Atmospheric Administration
(NOAA) is the organization responsible for issuing weather, hydrologic and climate forecasts
and warnings for the US. The Storm Prediction Center (USSPC), located in Norman, Oklahoma,
is one of the national centers for environmental prediction that provides severe weather and
tornado watches. Tornado warnings are issued by local weather forecast offices (WFO).
Presently, there are 122 WFOs covering the whole US. Each forecast office has an area of
responsibility of between 20 to 50 counties. The development of the activity network of the US
TDWC system (Figure 5-1) is a result of a search process to identify the possible links in the
network.
85
Figure 5-1: US Tornado Detection, Warning and Communication Network
86
Information for the US TDWC network was collected by means of reviewing extensive literature
and websites of collaborating partners of the warning communication network. Moreover, a site
visit to the USSPC, WFO-Norman and the National Severe Storm Laboratory (NSSL) provided
the latest information regarding the TDWC system.
In Figure 5-1, each box in the network represents an activity. The activity name is shown in the
upper part of the box and the collaborating partner of that activity is shown at the bottom of the
box. In this network, finish-to-start (FS) activity relationship with no time lag from any activity
to its successor was used to connect the activity sequence. When there is more than one
predecessor, two logical relationships; „OR‟ and „AND‟ were used as necessary. Situations with
„OR‟ relationship is shown in the relevant boxes of succeeding activities. The rest of the
activities with more than one predecessor have the „AND‟ relationship. The following
subsections provide a review of the main stages of the activity network.
5.2.1 Tornado Warning Stages
There are four main stages in the tornado detection and warning process, as shown in the Table
5-1. They are outlook, mesoscale discussion, watch and warning. Issuing of bulletins at these
stages is based on the level of threat. This hierarchical process increases the confidence that an
event is occurring. The USSPC issues severe weather outlooks, mesoscale discussions and
tornado watches, whereas the WFOs are responsible for issuing tornado warnings. The USSPC
uses Doppler radar data, satellite data, numerical weather prediction models and ground level
atmospheric observations to identify severe weather conditions. When predicting severe weather
several days in advance, forecasters continuously look for conditions favourable for the
87
development of severe weather situations and issue various bulletins based on their best estimate
of the probability of an event occurring .
Table 5-1: Hierarchical Process of Tornado Warnings in the US
Bulletin
Type
Typical
Time
Outlook
1 to 3days
Mesoscale
Discussion
Tornado
Watch
Tornado
Warning
Lead
Probability of Confidence
Forecaster confidence is 10 percent
or greater .
1-3 hours prior to Forecaster confidence is generally 30
tornado watch
to 50 percent. (When thunderstorm
development is ongoing or expected
soon.)
2 to 8 hours
Forecaster confidence is generally
greater than 50 percent.
10 to 30 minutes
Forecaster confidence is generally
greater than 80 percent.
Issued By
USSPC
USSPC
USSPC (updates,
cancellations by
WFO)
Local WFO
Source: (NWS, 2010a)
The USSPC follows the concept of a forecast funnel to understand the development of large
scale weather systems several days in advance to the development of tornadoes within several
minutes (Personal Communication, 2011). According to the forecast funnel concept, the tornado
forecasting and warning process analyzes the weather situation on several levels. The initial level
of the analysis is at the hemispheric and synoptic level. At this level, satellite imagery results are
used to analyze the weather situation and track changes in the environment that can trigger
potentially severe weather conditions. Severe weather outlooks are issued based on these results.
The next level is the regional level analysis. Mesoscale discussions generally focus on the
regional level weather conditions, such as the development of cold fronts and rotations. At this
level, Doppler radar, lightning and wind profile analyzes are used to predict the development of
88
cold fronts. USSPC forecasters also use numerical weather prediction models to determine if
atmospheric conditions, temperature and wind flow patterns create favourable conditions for
developing severe weather situations.
At the local level, the forecasting process is mainly associated with the advanced recognition of
supercells. Thunderstorm movements are closely observed for wind structures that can spawn
supercells. When there is a high possibility that supercell thunderstorms will occur based on realtime observations, severe weather watches (by the USSPC) and severe weather warnings (by
WFOs) are issued. Tornado watches are issued by the USSPC for the areas with higher potential
for an incoming tornado from these supercell thunderstorms. This information alerts WFOs,
emergency managers and the public about the potential for tornadoes.
5.2.2 Tornado warnings by WFO
Local WFOs monitor the weather around the clock and provide warnings, forecasts, and services
to the public in their area of responsibility. They have technologies for observing and forecasting
severe weather conditions and resources for issuing warnings. WFOs mainly use Doppler radar
data that help to determine whether a watch or warning is needed. Although Doppler radar data
show the possibility of developing tornadoes, it cannot predict tornado occurrences directly.
Therefore, forecasters get the support of spotters to closely monitor the ground conditions and
conduct in-house analyzes of weather situations before making tornado warning decisions. The
USSPC and neighbouring WFOs provide information to the local WFO. According to the NWS
reports, issuance of a tornado warning must satisfy at least one of the following conditions:

Atmospheric conditions support the tornado hazard,
89

Radar signatures indicate a tornado, and/or

A credible severe weather spotter reports a tornado hazard.
In the activity network, this situation is represented by the use of „OR‟ logical relationship for
the activity „Tornado Warning Issued‟. The forecaster‟s decision to issue a tornado warning is
more subjective, and they display a large range of confidence thresholds in practise. However,
they have around 80 % confidence for issuing warnings for more easily detectable large-scale
outbreaks.
The warning message often involves the locations of the warnings, a valid time period,
recommended evacuation or sheltering methods, as well as supplementary information provided
by the spotters on the ground (Figure 5-2). According to the NOAA records, the warning lead
time is generally 14 minutes (NOAA, 2011b). However, in some cases, when the environment is
very favourable for tornadoes, or a long-track tornado is being observed, the warning lead time
may reach or exceed 60 min.
5.2.3 Warning Communication
With the goal of maximizing the dissemination of warning information, the NWS uses a
multilayered approach. Local radio and television stations are the primary conduits. Many of the
local television stations maintain their own radar systems and spotter networks. When a WFO
issues a tornado warning, local media can show a map of the area affected by the warning. This
helps the public to have a general understanding of the impacted area. They may also show live
radar images from the NWS or from their own local radar, along with pictures and audio from
storm spotters employed by the station. This audio/visual delivery is thought to help the public
90
understand the location and severity of a tornado threat and encourage them to take protective
action. The NOAA Weather Radio service, which is the voice of NOAA‟s National Weather
Service (NWS, 2010a), provides tornado watch and warning information. NOAA Weather Wire
Service (NWWS) disseminates high-priority watch and warning products to users within ten
seconds of issuance, through specific satellites or the Internet.
There is a national system that consists of a network of telephone circuits connecting state and
federal warning points, called the National Warning System (NAWAS). During weather
emergencies, the NAWAS can be activated by a WFO for the dissemination of watches and
warnings to relevant recipients, such as emergency officials. This is also used as a
communication means among emergency managers in nearby counties for sharing information,
such as tornado sightings, speed, and direction.
The national level emergency alert system (EAS) can also be used to broadcast tornado warning
messages to the public. In the NWS, the NOAA Weather Radio network is the entry point to the
EAS. Authorized users at the local level can activate the EAS to send messages. In addition to
these methods, local emergency managers can activate an outdoor warning siren system to alert
people who are outdoors that they need to go indoors and seek further information. Presently,
many tornado-prone regions have siren system coverage.
91
Figure 5-2: A Tornado Warning issued by a WFO in the US
Source: (NWS, 2010b)
92
Sending warning information through mobile phones is a relatively new application. Mobile
phone users can get warning information as mobile text alerts. In addition to this, Internetenabled mobile phone users receive warnings through social media, such as Facebook and
Twitter. WFOs are also experimenting with communicating directly with the public via NWS
accounts on Facebook and Twitter. Weather warnings are also available on the official NWS
website (www.weather.gov). With these various warning dissemination methods, there can be
several ways in which people receive warnings. This is shown in the network by the number of
predecessors for the activity „Public Receives Warning‟.
5.2.4 Role of SKYWARN Spotters
The NWS has trained a volunteer network of storm spotters that often operate under the label of
SKYWARN. These volunteers help to obtain critical weather information in the field.
SKYWARN spotters identify signs of a developing tornado and relay this information to the
WFO in their locality (SKYWARN, 2011). Many SKYWARN spotter groups communicate
directly with one or more WFOs via amateur radio. Activation of the network is done by the
WFO or the emergency managers in the area, such as the local police, fire department or an
emergency management group. Spotters communicate among themselves and with network
controllers, as well as volunteers working in the WFO. For WFOs, spotters provide timely
information, which is very important in tornado warning decision-making (McCarthy, 2001). For
those emergency managers who activate sirens and local broadcast systems during severe
weather, SKYWARN spotters provide critical information that may prompt immediate action.
93
5.2.5 Emergency Manager and First Responders
As can be seen in the network, emergency managers (EM) play a major role in communicating
the threat to the public. Their collaboration with the WFOs, spotters, media and the public
ensures that local emergency managers are ready to respond should a disaster occur. During the
early preparedness phase, local emergency managers activate their emergency operation center.
They assess severe weather threats on county or city basis and support disseminating warnings to
the public through several ways: outdoor siren activation, notifications to school and hospitals,
EAS activation and cable television interruption (League et al., 2010). As first responders, local
police also support the operation of the warning communication and activation of sirens.
The US tornado detection and warning network represents the information flow from the point of
tornado forecast, detection to providing warning communications to the public. It also represents
the interactions among various authorities, volunteers and the public in the tornado detection,
warning and communication process. As can be seen in the network, the issuance of a tornado
warning depends on input from several sources. The warning communication system is also
multimodal, in order to distribute tornado warnings rapidly and efficiently.
5.3 Comparative Analysis of the US and Canadian Systems
The US and Canadian TDWC systems were compared, focussing on key issues such as
prediction/detection capabilities, warning providers, warning area, warning dissemination
method, and the spotters‟ role in tornado detection and warning.
94
5.3.1 Prediction/Detection Capabilities
In the US, advances in research and forecasting technology have increased the potential for
recognizing severe weather several days in advance. Observations, numerical forecast models,
Doppler radar and satellite data, coupled with the best available technology, are used to generate
forecasts. A number of weather forecasters at the USSPC look at the weather situation at
different stages of severe weather development. Local WFOs also continuously check local
severe weather conditions.
The US is gradually enhancing the predictive or early detective capacity for tornadoes. The use
of a probabilistic approach to inform the public on the occurrence of a severe weather outbreak
gives information about the level of risk to be expected. This alerts the public to prepare for
tornado threats; therefore, the issuance of a tornado warning does not become an unexpected
event during a severe weather outbreak.
Although the US is gradually moving toward a warning on forecast approach (NSSL, 2013c),
Canada mainly relies on a warning on detection approach, i.e., detection based tornado warnings,
which gives a very brief window of opportunity for evacuation. In Canada, tornado warnings are
issued based mainly on observations and related analyzes (Cao and Cai, 2008). The five SPCs
have the responsibility for weather forecasting for the entire country, and they pursue a more
centralized approach to forecasting. This is a result of the restructuring of the MSC in 2003, due
to financial pressures, reducing the number of local offices from 14 to 5 and increasing the
automation of forecasts through numerical weather prediction (Sills, 2009). Although this
automated prediction is good at the initial recognition of environmental conditions supporting
95
tornado development, such as heavy rainfall, its tornado prediction capability is not as good. It is
important to have a human forecaster to recognize these localized and short-lived events in
advance. Furthermore, having a large area of responsibility (more than 1,000,000 km2) impedes
the efficiency of a forecaster in recognizing high impact weather, especially when it comes to
tornado forecasting through analysis of storm-scale processes.
There are several reasons for the disparities in tornado prediction between the US and Canada.
Among them, resource allocation is a major issue. In the US, a network of 161 Doppler radars
provides data to analyze severe weather development, whereas Canada has only 31 Doppler
radars located in highly populated regions. There are a number of factors to be considered in
developing a Doppler radar network such as tornado climatology, Doppler radar coverage issues,
local site considerations as well as other financial and technical capacity issues (Newark and
McCulloch, 1992). Tornado prediction/detection ability is highly influenced by the resolution
and the sensitivity of radar as well. Sufficiency and efficiency of the present radar network needs
to be examined to answer why Canada is lagging in Doppler radar based prediction/detection
capabilities.
In the US, a number of observers, including USSPC forecasters, WFO forecasters, broadcast
meteorologists at television and radio stations, storm spotters and emergency managers at the
local level, look at severe weather data. Due to these reasons, it is unlikely that an approaching
tornado will go undetected in the US. However, in Canada, SPC forecasters are the observers of
the Doppler radar data for the detection of tornadoes. In the absence of a large number of eyes
96
looking at severe weather that mainly develops as isolated storms, tornadoes have a high
possibility of going undetected.
When considering the tornado occurrence rate in Canada, it may not be feasible to institute local
level analyzes to detect the low number of isolated events. Furthermore, forecasters at SPCs may
not be under much pressure for prediction. In the US, due to the high tornado occurrence rate and
the outbreak pattern impacting a large area, the demand and the pressure for prediction and early
detection is much greater. This is even manifested by the high false warning probability in the
US TDWC system.
5.3.2 Warning Provision and Emergency Preparedness
The US has a decentralized approach for tornado warnings, which are issued by the WFOs
within their locality. The USSPC and WFOs across the US focus on weather in their regions. The
USSPC plays a supporting role by issuing tornado watches in advance and supplying information
to the WFOs prior to issuing tornado warnings. In addition to the WFOs, emergency managers
can also issue alerts to the public regarding an imminent tornado danger, without waiting until an
official warning is issued by a WFO.
In Canada, only the SPCs are responsible for issuing official tornado warnings. Even if a tornado
is sited in a locality, it should be reported to an SPC to issue a tornado warning. However in
Alberta, with the establishment of the AEA system, authorized users, including the local
emergency managers, can issue critical alerts to warn the public about tornadoes. Such local
systems are very important for the public to be aware of the potential for tornadoes, as the early
97
detection and warning capability of the SPC is not always feasible. Although a very good
alerting system is available, having a source of reliable information to activate the AEA at the
local level is a critical issue. Relying on information from outside sources create delays.
Information from the public generally comes at the last movement. The 911 centres are generally
overloaded with emergency calls. These factors can delay the immediate activation of the AEA.
SPCs as warning providers do not have direct communication with the local emergency
managers. They receive SPC watch or warning information broadcasted through the media or
RSS (Rich Site Summary) feeds only and do not receive any direct alerts about incoming
dangers to their area other than media; thus they have a very limited time span to launch
preparedness activities.
5.3.3 Warning Area
The US has recently moved to a new paradigm of tornado warnings based on storms, i.e., storm
based warnings, the aim of which is improvement in the accuracy and quality of the warnings.
With this new system, the warning area is more specific, covering only the population under
highest threat. This approach avoids unnecessary warnings to the public. Knowing the area under
threat, first responders can better pinpoint the locations of severe weather and can safely move
resources to the areas that need assistance.
In Canada, tornado warnings by SPC are issued for a whole county or a regional municipality,
mainly based on tornado sighting reports. This puts the whole population of the region under
warnings, even if the tornado prone area for a particular thunderstorm is much less than the area
98
of the warned region. The location uncertainty creates problems related to how the public heeds
the warnings. This uncertainty also impedes the emergency managers‟ and first responders‟
responses in the earliest stage. No first responder will go out to a field before a tornado
touchdown. In Alberta, AEA has made a step forward in designating a warning area in their
alerts even if it is not precisely delineated based on storm level analysis.
5.3.4 Warning Dissemination Methods
Having a number of reliable warning dissemination methods ensures that the message is received
by the maximum number of people among the target population. Both the US and Canada use
broadcasting media, the Internet and social media to issue tornado warnings to the public.
Weather radio is an automated warning system to disseminate warnings. This application is
especially useful when the public are not using many of the warning dissemination methods,
such as radio, television, mobile phones or the Internet, especially during the night. Although
weather radio is a common application used in the US, it is not a popular warning dissemination
method in Canada.
Canadians seek warning information mainly from broadcast media. Tornado warnings have been
recognized as urgent, and their immediate airing is required. However, these tornado warnings
are not supplemented with radar images, storm spotter reports or real-time communication with
the mobile media crews, as in the US. The use of Internet and mobile text alerts, as well as social
media, are much easier ways of disseminating warnings to the working population as well as
people on roads. Canada is gradually moving toward these new warning dissemination systems.
99
The activation of sirens by emergency managers is also an effective way of warning the public,
especially when the power and telephone lines fail and cellular phones malfunction during severe
weather outbreaks. Many tornado prone areas in the US have full siren system coverage.
However, Canada does not have such systems for local level emergencies. Implementation of a
siren system is, in fact, costly and not a strong requirement considering the low frequency of
isolated tornado events in Canada; therefore, local authorities in Canada may hesitate to consider
such a warning system.
Although there are various warning communication methods, people cannot entirely depend on
official warnings. Sometimes, authorities are unable to issue tornado warnings until the tornado
touches down. The best warning is based on what people can see for themselves. People should
remain alert to severe weather watches and warnings, as well as signs of approaching tornadoes,
in order to seek shelter if threatening conditions exist.
5.3.5 Spotters’ Role
Ground verification of severe weather and tornado potential, such as the appearance of funnel
clouds, improves forecaster confidence and adds more detail and credibility to warning
messages. Sometimes these field reports are used to supplement warnings already in effect in the
area (McCarthy, 2001). Spotter confirmation also reduces the probability of false tornado
warnings. The WFOs in the US have a close collaboration with the storm spotters in the field.
They provide ground level information to the WFO, as well as to the emergency managers, and
play a major role in the tornado detection and warning system. Some forecasters do not issue a
tornado warning until visual confirmation from spotters is received (Brotzge and Erickson,
100
2009). The spotters‟ role is especially important during the night, at locations far from radar
sites, as well as on marginal tornado days.
In Canada, although there is a network of spotters, they do not have a close collaboration with
the SPCs and local emergency managers. The SPCs contact individual spotters when ground
level information is required. However, large-scale network activation does not happen. Low
tornado frequencies and the isolated nature of severe thunderstorms may be the reasons that have
made self-activation weaker or unavailable. In the US, spotters know the potential of severe
weather several days in advance, based on outlooks issued by the USSPC. Spotters in Canada
probably do not receive such early information; therefore, the efficiency in providing
information in the pre-touchdown phase is lower compared to the spotter network in the US.
5.4 Conclusion
This chapter reviews the US TDWC system and provides a qualitative comparison and
evaluation of the US and Canadian systems. When compared with the US system, there are
several areas that need the attention in improving the TDWC system in Canada as discussed in
this chapter. However, it does not say that simply imitating the US system can solve the problem
areas in the Canadian system. There are several factors such as tornado risk potential, disaster
cost, financial and technical capacity limitations to be considered in making changes. Any
suggestions to optimize the present system should be a result from a proper cost-benefit analysis.
101
Chapter Six: Evacuation Behaviour of Households and Drivers during a Tornado Analysis Based on a Stated Preference Survey
6.1 Introduction
This chapter presents the results of a stated preference (SP) survey conducted to analyze the
evacuation behaviour of households and drivers under a hypothetical tornado warning situation
(Durage et al., 2013b). Using the Probit models, the factors influencing the evacuation decisions
are identified and discussed in detail taking the City of Calgary as a case study.
6.2 Disaster Evacuation
Disaster evacuation involves immediate and rapid removal of persons from the threat area. As
Wickramaratne et al. (2011, p.361) state: “Successful evacuation is the ultimate decisive factor
of the whole warning process since an incomplete evacuation could cause thousands of lives to
be lost”. Proper evacuation actions play an important role in disaster planning and management.
Evacuation decisions are closely associated with the risk perception of individuals (Slovic, 2000,
Nirupama 2012). The hurricane evacuation literature search conducted by Dash and Gladwin
(2007, p.69) indicated that “warnings by themselves do not motivate evacuation – people must
perceive risk”. Perceived risk can be defined as “... a direct function of both the warning
information received and the personal characteristics of the warning recipient” (Mileti and
O'Brien, 1992, p.43). This concept is also applicable for tornado evacuation, despite the fact that
tornado warnings have shorter lead times than hurricane warnings. As Dash and Gladwin (2007,
p.70) explained, “individuals process information through their own social lenses constructed by
102
their particular cultural context, and as a result, different people may well interpret the same
information and messages differently”.
There are several US based studies that have analyzed the evacuation behaviour of people during
a tornado event. Schmidlin et al. (2009) investigated tornado shelter options and shelter-seeking
behaviour among mobile home residents in four tornado-prone US cities. The study identified
several variables that were correlated with seeking shelter by mobile home residents, such as the
belief that they were in the path of the tornado, the belief that they were in danger, knowledge of
a sturdy building available for shelter, male gender and the presence of children in the house.
Although most of the tornado evacuation studies have focussed on the evacuation of households,
drivers are also vulnerable to tornadoes that approach road networks. Schultz et al. (2010)
analyzed how households and drivers receive tornado warnings and subsequently use this
information to make evacuation decisions, using a survey tailored for the risks, local hazards and
geography of Austin, the Capital City of Texas. Although the study had limited variables for
analyzing the factors that impact evacuation decisions, it provides a good reference for
comparing evacuation decisions for this study. However, it is worth noting that the response
behaviour of a city with a low tornado frequency may be different from a city with a relatively
high tornado frequency, such as Austin; thus, generalization may not be possible.
6.3 Stated Preference Survey Method
A SP survey method is a useful way of obtaining responses in order to understand and predict the
behaviour of decision makers (Louviere et al., 2000). It provides the opportunity to collect
103
multiple responses from the same respondent. Matyas et al. (2011) used SP survey methods to
analyze the risk perception and evacuation decisions of Florida tourists under hurricane threats.
They tried to understand the major relationships, but did not establish causal links between
evacuation decisions and various explanatory factors. In contrast, this research is aimed at
estimating the relationships among a set of explanatory variables and likely evacuation decisions.
In this study, a SP survey was designed to assess the tornado evacuation decisions under two
different hypothetical scenarios, i.e. household and driver evacuations, by asking the survey
participants how would they respond to these two situations. Econometric regression methods
were employed to determine how these factors influence the likely evacuation decisions.
6.4 Regression Methods
Regression methods have increasingly been recognized as a way of analyzing the factors
influencing decision-making processes. Mileti and O'Brien (1992) used multiple linear
regression methods to develop a model to analyze the factors influencing different earthquake
emergency response activities. They dummy coded each category on the response variable and
added them to create a scale to use as a continuous response variable. However, their analysis did
not provide information on the factors influencing each response category.
Recognition of the dependent variables in a set of different response categories as multinomial
(categorical) or ordinal (if applicable) is a more informative way to conduct the regression
analysis for discrete outcomes. Logit and probit regression methods can be used to develop better
models for predicting such discrete outcomes (Hosmer et al., 2013). There is extensive literature
on the application of logit/probit regressions in numerous disciplines, such as social sciences,
104
medicine, econometrics and engineering (Kwak and Matthews, 2002; Mahapatra and Kant, 2005;
Tay et al., 2009; Pfarr et al., 2011). The major difference between logit and probit models is the
distribution function. Logit models use a standard logistic probability distribution function, while
probit models assume a cumulative normal distribution in parameter estimation. The shape of the
probit function is very similar to that of the popular logit function, except that a different
distribution is used for parameter estimation. Even the coefficients obtained from these two
models are fairly close. However, the logit model relies on the assumption of independence of
irrelevant alternatives, which is not always desirable (Ben-Akiva and Lerman, 1985). As will be
discussed later in this chapter, the categories for the dependent variable in each model are not
independent and therefore these models violate the IIA (Independence of Irrelevant Alternatives)
assumption.
For example, in the driving scenario, the categories include the same action
(basement) with different evacuation times. In this analysis, the probit model was chosen for the
reason that it does not necessarily respect the IIA assumption.
6.5 Survey Design
A SP survey was designed to analyze the warning and evacuation behaviour of households and
drivers. The respondents were asked to assume that they received a tornado warning; and, their
intended evacuation responses when at home and driving were collected separately. The survey
was developed in SurveyMonkey© software to facilitate completion of the questionnaire online.
It used simple and clear language without technical words, so that any layperson could easily
understand it. Moreover, an optional video of a tornado touching down in an urban area was
included prior to the questions to give a sense of a tornado disaster.
105
The survey focussed on key areas, namely the knowledge of Calgarians about tornadoes and the
present warning system, previous disaster experience, source of warnings, pre-evacuation
actions, evacuation actions and socio-economic variables, such as gender, age, level of education
and presence of school aged children. A pilot survey was initially conducted with a sample of 15
participants, and the questionnaire (Appendix C) was further clarified. The ethics approval for
this survey was obtained from the Conjoint Faculties Research Ethics Board for the University of
Calgary.
The online survey was open for a three-month period from December 1st, 2012 to February 28th,
2013. A number of methods were used to find participants for the survey. The survey was
advertised in two local newspapers: the Calgary Herald and the Calgary Sun. The survey was
also
available
on
two
Calgary-based
websites:
http://www.calgaryarea.com
and
http://calgary.kijiji.ca for easy access to any interested Calgarian. Furthermore, the online survey
link was sent to the authors‟ contacts in Calgary, requesting that they fill out the survey and share
the link with their Calgary-based contacts.
Nearly 500 Calgarians took part in the online survey; and, the records of 421 Calgarians who
completed the survey were used for further analysis. This sample size exceeded the minimum
sample size (n) of 384 obtained using the Cochran‟s formula (Bartlett et al., 2001) that can be
written as;
𝑛=
(𝑡 2 × p × q)
(6.1)
d2
where t= value for selected alpha level of 0.025 in each tail = 1.96
106
d= acceptable margin of error for proportion being estimated = 0.05
p × q = estimate of variance = 0.25.
Thus, the sample was deemed sufficient to infer research findings back to the Calgary
population. The profile of survey respondents is shown in Table 6-1.
Table 6-1: Profile of Survey Respondents
Variable
Response Categories
Percentage
Gender
Male
Female
Below 30
Between 30 and 50
Above 50
Single Detached Dwelling
Other
One
Two
Three or More
Yes
No
66.0
34.0
13.5
52.8
33.7
74.6
25.4
12.8
32.3
54.9
33.5
66.5
Yes
No
Not Answered
Less than $50,000
Between $50,000 and $120,000
Above $120,000
Not Answered
Up to High School
Training after High School
Undergraduate Degree
Postgraduate Degree
Not Answered
6.2
91.7
2.1
9.3
36.8
40.1
13.8
7.1
32.8
38.5
19.0
2.6
Age
Dwelling Type
Household Size
Presence of School
Aged Children
Presence of People with
Reduced Mobility
Household Income
Level of Education
107
The distribution of the demographic and socio-economic variables showed that a diverse group
of respondents participated in the survey. The survey results are analyzed in detail in the
following sections.
6.6 Tornado Knowledge, Preparedness and Previous Disaster Experience
Risk perception is generally associated with knowledge of a disaster and previous experience of
similar situations. The first part of the survey was devoted to testing the respondents‟ knowledge
of tornadoes, analyzing their perception about community preparedness, and checking whether
they had previous disaster experience that required urgent evacuation. The responses for this set
of general questions are shown in the top part of Figure 6-1. The majority of respondents
selected the correct answers for all of these questions.
It is important to note that, given the low frequency of tornado occurrence, there has not been
many awareness programs in Calgary to improve the public‟s knowledge. Respondents‟
awareness is mainly attributed to broadcasts of US tornado related news. There is not much
tornado related news or many tornado warnings issued in Canada. The average number of
tornado warnings received by a Calgarian in a summer season is generally less than five; and,
often there are none that are associated with an actual tornado occurrence. Despite these
limitations, most respondents correctly knew the tornado season, the time of day they occurred,
the official source of warning, the best sheltering place and the lift power of a tornado.
Seventy-six percent of respondents agreed with the fact that precise tornado forecasting is
difficult, due to uncertainties in estimating weather conditions and technological limitations.
108
Around the same percentage of respondents knew the difference between a tornado watch and
tornado warning. However, only 39% of respondents knew that there are various environmental
cues as to when tornadoes are about to occur.
0
20
40
60
80
100
Precise tornado forecasting is difficult due to
uncertainties in estimating weather conditions and…
Tornadoes give warnings through various
environmental cues when they are about to occur.
I know the difference between a “Tornado Watch”
and “Tornado Warning”.
It is important to wait until an official warning is
issued to take evacuation actions.
I will seek more information about an official
warning before evacuating.
Disagree
Neutral
Agree
I know how to take cover during a tornado.
Under a tornado warning, I am not able to
concentrate on the instructions given.
Overall, my residential community‟s preparedness
for a tornado disaster is adequate.
0
20
40
60
80
100
Most tornadoes are born in the turbulence of a
thunderstorm.
Summer is the peak tornado season.
Yes
A tornado can occur at any time of the day or night.
No
The lowest level of a shelter is generally the safest
location during a tornado.
Don't Know
An intense tornado can uplift vehicles away from
roads.
Environment Canada is the official source of tornado
warnings in Canada.
Figure 6-1: Response Percentages to Tornado Knowledge Questions
Twenty-five percent of the population would wait until an official warning is issued to take
evacuation action, and slightly more than half of the population would seek more information
109
about an official warning before evacuating. These results indicate that a considerable percentage
of people make relatively unsafe decisions in waiting for official warnings or seeking more
information. Precise tornado detection is far from perfect; and, even if a warning is issued, the
lead time is very short. Therefore, it is important to pay attention to the information received
through official or unofficial sources and take immediate safe evacuation actions.
It is to be noted that Calgary has not experienced a high intensity (F2 or greater) tornado event in
its recorded history. However, there have been records of several low intensity (F0 and F1)
tornadoes and funnel cloud appearances associated with severe thunderstorms in the summer
months (McCarthy, 2011). Although community impacts have been minimal in these events,
tornado warnings have been given, urging the need for evacuation. In addition to severe
thunderstorms, there are disaster and emergency situations, such as fire, imminent flooding and
wind storms that are common in Calgary.
In response to a question about previous disaster experience, nearly half of the respondents stated
that they had experienced a disaster or an emergency that required an immediate reaction or
evacuation. Nearly 70% of these respondents had experienced fire-related evacuations. Although
it is difficult to link such experiences to the intended behaviour of people during a tornado, they
do provide important lessons on how to activate emergency preparedness plans and get the
public to take immediate action.
Some insights concerning individual and community level preparedness emerged through the
survey answers. The majority of respondents knew how to take safe cover during a tornado
110
(61.8%). This figure is relatively low compared with the 82% of people who said that they were
knowledgeable about safe evacuation actions in the Austin survey (Schultz et al., 2010). Nearly
80% of respondents were able to concentrate on the warning instructions given. These results
show that people are familiar with tornado disasters and know how to prepare for them.
However, when it comes to the community level, the majority of the respondents (61%) believed
that their residential community‟s preparedness for a tornado disaster is not adequate. In this
regard, it is imperative for emergency managers to improve awareness and preparedness at the
community level through training and applied learning activities.
6.7 Tornado Warning Sources of Information
There are a number of official and unofficial sources from which a person can receive warnings.
Warnings sources, such as emergency weather radio and public media, can be recognized as
official sources; whereas warning information coming through other people, unofficial websites,
social networks and visible environmental cues are unofficial sources. The survey listed possible
sources of warnings for households and drivers, and survey respondents were asked to rate the
different sources of information that they would rely on to receive tornado warnings under each
scenario. Rating averages for each warning source are presented in Figure 6-2.
For the household warning scenario, nature‟s way of giving warnings through visible
environmental cues had the highest rating average, followed by television, calls from trusted
persons, mobile text alerts, weather websites, and local radio and social media. For drivers, local
radio was the most likely source of warnings, followed by environmental cues, calls from trusted
persons, variable message signs, mobile text alerts and weather websites.
111
Please rate each likely source of warning to your household in case of a tornado, on a scale
from 1 to 5 (1-Least likely , 5-Most likely).
Call from a trusted person
Visible environmental cues
Weather websites
Mobile text alerts
Television
From neighbours
Social media
Emergency weather radio
Local radio
0
1
2
3
4
5
During your trip, from what sources are you likely to receive the warning? Rate each source of
warning on a scale of 1 to 5 (1-Least Likely, 5-Most Likely).
Visible environmental cues
Variable highway message signs
Mobile text alerts
A call from home or friends
Other road users
Social media
Local radio
0
1
2
3
4
5
Figure 6-2: Tornado Warning Sources
As indicated in the responses, appearance of visible environmental cues, such as funnel clouds,
had a high rating average for giving warnings to both the household and driving populations.
112
However, funnel clouds are not easily seen in all tornado situations due to heavy rain or hail.
Even if a funnel cloud appears, it may be at the last minute. Hence, it is important to be vigilant
about severe weather development.
According to these results, the emergency weather radio (weatheradio) application had the
lowest rating average for households, although it is a major source of disseminating watches and
warnings issued by Environment Canada‟s storm prediction centres directly to the public.
According to Environment Canada, over 90% of Canadians live within the coverage of a
weatheradio transmitter. With Specific Area Message Encoding (SAME) technology, users can
receive only the warnings relevant to their area (EC, 2013d). However, weatheradio is not very
popular in Canada compared to the US, where they are the primary warning systems in tornadoprone US cities.
In recent years, mobile phones have become a primary communication device during
emergencies or disasters. There has been an upsurge of Internet-enabled phones that have the
capability of providing evacuation routes, locations and notifications (Oxendine et al., 2012).
These smart phones allow access to weather websites and social media that provide tornado
warnings issued for the area. Moreover, registered users can get mobile text alerts sent to their
phones.
Analyzing a tornado warning dissemination and response at a university campus, ShermanMorris (2010) showed that cell phones have become the most common means of receiving first
alert messages, especially for students. Recent advancements in mobile technology provide much
113
faster ways to communicate warnings to the public. There are mobile applications (apps) that can
be downloaded to phones to receive warnings. For example, in Canada, The Weather Network is
a private meteorological company that has launched weather apps for phone users. In the US, the
NWS of the NOAA is testing mobile weather warnings, which is a way of relaying tornado
warnings to phones via cell towers. Phones will automatically pick up warnings broadcast by
nearby cell towers (NOAA, 2012b).
Although mobile phones can provide a faster way of receiving warnings, there are several
limitations. For example, access to mobile text alerts is available only to registered users: social
media is limited to certain groups; and, there can be groups outside social networks (Donner et
al., 2012). Using phones while driving for texting, talking or web browsing is quite unsafe and
are restricted under the new distracted driving provision in Alberta‟s Traffic Safety Act (GA,
2011). In addition, wireless and cellular networks may become unreliable in such situations, as
they are prone to failure during severe weather conditions and natural disasters. They may also
become oversaturated as the affected population attempts to reach out to others during such
times.
According to the survey responses, mobile text alerts and calls from home or friends had average
ratings levels, while social media was the least likely source. As can be seen by comparing the
two scenarios (household and driving), the use of these warning sources is relatively low when
driving.
114
Environment Canada‟s PASPC is an authorized user of the AEA system. An alert can be
distributed to the public through various outlets, including radio, television, the Internet, social
media (e.g. Facebook and Twitter) and road signage. Furthermore, Rich Site Summary (RSS)
feeds allow the public to stay informed by capturing the latest alerts though a web browser or as
an e-mail notification.
The use of intelligent transportation systems (ITS) in evacuation is also an emerging technique to
warn road users in real time (Robinson and Khattak, 2011; Ye et al., 2010). In Calgary, there are
variable message signs (VMS) to alert drivers to construction work, accidents, detours,
dangerous weather conditions and more. The Calgary survey showed that this option had a rating
average of around 3, indicating that VMS is a good potential source of warning to drivers. It is
important to have clear communication and coordination between Environment Canada‟s Prairie
Storm Prediction Centre and the City of Calgary‟s Road Operations Centre to present real-time
warning information to motorists using VMS.
Having a multimodal warning communication system is very important to ensure that households
and drivers get reliable warnings through at least one source. With the availability of the latest
technology and different official and unofficial communication methods, it is important for
Calgary emergency managers to coordinate the warning system through a variety of media, in
order to reach a diverse population with different preferences. The public seems to have access to
multiple sources of information for receiving warnings, and some people may seek confirmation
from different sources to reinforce the information prior to making an evacuation decision. This
115
again highlights the need for coordination of warnings through various media and sources, in
addition to traditional channels, such as television and radio.
6.8 Pre-evacuation Actions
Warning issuance itself does not guarantee that people seek shelter immediately. There can be
actions that they may take before evacuating. The survey asked the respondents to rate a set of
pre-evacuation actions on a scale from 1 (very low) to 5 (very high). Respondents were asked to
ignore the question if they definitely would not take any of these actions. The rating averages for
pre-evacuation actions are shown in Figure 6-3.
In both the household and driving scenarios, people tended to communicate with their family
members before evacuating to a safer place. As family members may not be together when a
disaster strikes, it is natural that other family members try to communicate with them. There is
also a tendency to verify or confirm the hazard before taking evacuation actions. However, there
were high perceived risks about tornadoes in both the household and driver scenarios; and, the
respondents would not be likely to take time for taking photos, videos or sharing message
through social media.
116
Household pre-evacuation actions
0
1
2
3
4
Driver pre-evacuation actions
0
1
2
3
5
Verify information with a different
source
4
5
Verify information from different
sources
Go outside to confirm
Inform others on the road
Inform others in the house
Telephone family members/ close
friends
Call family members
Inform neighbours
Plan for photos or videos
Plan for taking photos or videos
Tweet or Facebook to share the
message with others
Tweet, or post on Facebook to
share the message with others
If the children are out of the
house, leave to pick them from the
local school or other place
Drive to pick children from school
or other activity
Figure 6-3: Pre-evacuation Actions
As indicated by rating averages of around 3.5, a significantly high percentage of respondents
stated that they would drive to pick up children from school or other activities before taking a
safe evacuation action. Given the limited lead time associated with tornadoes, going out of the
home or driving to pick up children under a tornado warning situation can be a life-threatening
risk and action. Schools or children‟s activity centres are expected to have the fastest, most
accurate and reliable means of receiving critical weather information and have action plans to
safeguard children against tornadoes. It is important that such information is clearly disseminated
to parents, so they stay calm, seek safe shelter for themselves and avoid taking unsafe actions by
driving to pick up their kids. Currently, Calgary schools do not have tornado evacuation drills.
117
6.9 Evacuation Wait Time
Tornado warnings convey a sense of urgency to the public to safeguard against potential lifethreatening situations that are in progress. The survey asked questions to determine how long
respondents would wait before completing the most likely evacuation action in both the
household and driving scenarios. A set of random numbers that follow a normal distribution
were generated to represent the number of respondents who selected each time range for these
multiple choice questions. A CDF was drawn for each scenario (Figure 6-4) using the generated
data sets.
Driver Evacuation Wait Time
70
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Frequency
60
50
40
30
20
10
1
4
7
10
13
16
19
22
25
28
More
Time (Minutes)
90
80
70
60
50
40
30
20
10
0
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1
4
7
10
13
16
19
22
25
28
More
0
Frequency
Household Evacuation Wait Time
Time (Minutes)
Frequency
Cumulative %
Frequency
Cumulative %
Figure 6-4: Evacuation Wait Time
Based on the CDF curve drawn for household wait times, there is a 50% chance that the
evacuation would be completed approximately within 4 minutes or less. For driver evacuation,
there is a 50% chance that the evacuation would be completed roughly within 3 minutes or less.
In both scenarios, there is an approximately 90% probability that the evacuation would be
118
completed within 10 minutes or less. These figures are plausible time estimates that reflect the
urgent need for evacuation during a tornado.
These time durations are, however, based on the intended time estimates taken before an actual
completion of an evacuation action. The actual evacuation wait time may vary depending on
factors such as mental and physical abilities, surrounding environment, time of the day, and
congestion level in the case of the driving scenario. It is important not to wait until it is too late.
Personal preparedness can assist people greatly in minimizing the wait time in the event of a
tornado.
6.10 Evacuation Actions
Taking proper evacuation actions is key to the minimization of injuries and fatalities from a
tornado. Survey respondents were given a set of evacuation actions and asked to rank them from
the most likely action (Rank 1) to the least likely action. The choice of an option called „other‟
was also included, if a respondent had another action that needed to be mentioned. The ranking
averages of the household and driver evacuation actions are shown in Table 6-2.
The survey results of household evacuation actions show that going to a safer area in the
basement has a ranking average of 1.47, indicating it would be the most likely evacuation action.
Lying down in a bathtub was the second most favourable action by households. A securely
installed bathtub without adjacent mirrors provides a relatively safe place, even if a structural
collapse occurs. As Monfredo (2008) suggested in his paper on the prediction and analysis of an
EF-5 tornado, “if an individual does not have access to underground shelter, protecting oneself in
119
an inner room or bathtub with a mattress pulled over the top of the body and shielding the head
remains good advice, as does keeping as many strong walls as possible between humans and the
storm.”
Table 6-2: Tornado Evacuation Actions
Household Evacuation
Ranking Average (1-6)
Go to a safer area in the basement
1.47
Lie down in a bathtub
3.20
Stay in the house but take no evacuation action
3.79
Go to a safer building within the neighbourhood
3.39
Drive away to avoid the threat
3.70
Other
5.46
Evacuation for Drivers
Ranking Average (1-7)
Drive away from the direction of the tornado
1.71
Drive to the nearest building
2.73
Do nothing and continue driving to my destination
4.85
Stop on the roadside and stay in the vehicle
4.23
Stop on the roadside and hide under the vehicle
5.26
Seek roadside shelters, such as bridges, highway
overpasses and stop underneath
2.74
Other
6.49
120
For household evacuation actions, relatively unsafe decisions, such as no action or driving away,
are in the lower levels of the ranking order. Going to a safer building within the neighbourhood
and driving away to avoid the threat are the options for seeking shelter while at home. The
survey showed that a total of 16.6% respondents preferred these two actions over others. This is
consistent with the Austin study of Schultz et al. (2010), which indicated that 18% of
respondents would choose to leave their homes to get out of the path of a tornado. Moreover, the
survey results show that 4.3% of respondents stated they would stay in the house without taking
any action. This is again consistent with the 3% of respondents who stated that they would take
no action during the Austin survey (Schultz et al., 2010).
The household evacuation actions of lying down in a bathtub (Bathtub), staying in the house but
taking no evacuation action (No Action), going to a safer building within the neighbourhood
(Neighbourhood) and driving away to avoid the threat (Drive Away) showed ranking averages
between 3 and 4. Pairwise statistical comparisons of these four action scores were done to see if
they are significantly different from each other. The paired sample T-test procedure was used to
test the hypothesis of no difference between each two actions at a confidence interval of 95%
(Table 6-3). The non-significance in the pair of No Action and Drive Away indicates that these
two actions were not statistically different. They are both equally unsafe actions. All other pairs
that were partnered with No Action (pairs 1 and 4, Table 6-3) and Drive Away (pairs 3 and 6,
Table 6-3) are significantly different from each other. This shows how the Bathtub and
Neighbourhood actions are relatively safer than Drive Away and No Action.
121
An alarming response was presented in the driving scenario, which indicated that driving away
from the direction of the tornado would be the most likely evacuation action with 58% of
respondents intending to drive away from the direction of the tornado. This was significantly
higher than the 39% of Austin residents in the Schultz et al. (2010) study. The safety of driving
away from a tornado is quite uncertain, given that the tornado may not be visible or the
unpredictability of the direction and speed of a tornado.
Table 6-3: Pair-wise Comparisons for Household Actions
T statistics
Sig.(2-tailed)
Pair 1
Bathtub – No Action
-6.137
0.000
Pair 2
Bathtub – Neighbourhood
-2.003
0.046
Pair 3
Bathtub – Drive Away
-4.31
0.000
Pair 4
No Action – Neighbourhood
4.018
0.000
Pair 5
No Action – Drive Away
0.964
0.335
Pair 6
Neighbourhood – Drive Away
-3.231
0.001
Warnings do not give precise information on a tornado‟s wind speed or intensity prior to an
actual occurrence. Due to these reasons and depending on the strength of the tornado, a vehicle
may also not be a safe place to seek shelter (Schmidlin et al., 1998; Hammer & Schmidlin,
2001). As indicated by lower ranking levels, survey respondents likely recognized the
diminishing safety of vehicle-based actions for evacuation, such as staying inside the vehicle
122
selected by 1.1% of respondents and hiding under the vehicle selected by only 0.24% of
respondents.
The safest decision in a tornado emergency would be to stop driving and seek shelter in the
lowest level of the nearest building, although the survey results show that this action has the
second ranking average. In the absence of a building in the nearby area, people tend to seek
roadside shelters, such as bridges and highway overpasses. Its ranking average of close to that of
the nearest building action indicates that people mistakenly think that a highway overpass or a
bridge provide safety from a tornado.
Leaving a vehicle at the road side may also be problematic, if a driver is in congested traffic,
especially when there are people who want to drive away. Cooperative behaviour instead of
competitive behaviour can be more effective in such a life-threatening emergency. Based on a
literature survey, Drury et al. (2009) highlighted a number of factors – group size, perceived
threat, exit time, leadership, rational contingencies and emotional arousal – as variables that play
a role in the extent of competition versus cooperation during an emergency evacuation. For
example, responding to a tornado warning situation during peak period traffic on a highway full
of vehicles needs a great deal of cooperation, so that everyone can stop their vehicles and find
the nearest shelter. There may, however, be a portion of drivers who perceive a lesser threat and
do not want to evacuate, resulting in competitive behaviour. In the presence of moving vehicles,
those who want to evacuate, may have to wait a long time before reaching a shelter.
123
Road evacuation can be cumbersome, especially in extensive road networks. In reviewing the
methods for improving the efficiency of road evacuation, Abdelgawad and Abdulhai (2009)
proposed the modelling of an emergency evacuation using traffic simulation and optimization
techniques. However, there are some challenging issues, such as the lack of a hazard prediction
method, of an understanding of how to best issue evacuation orders and of destination selection,
which can affect the efficiency of the evacuation modelling process. Given the many
uncertainties and situational dependence, a common evacuation model tailored to tornadoes
requires a comprehensive planning effort.
In the next section, various factors that influence the evacuation actions of households and
drivers are analyzed using regression models. The data analysis is conducted using Stata data
analysis and statistical software. Separate analyzes are conducted for the household and driving
scenarios using probit regression methods.
6.11 Partially Generalized Ordered Probit Model for Household Evacuation
Probit regression uses a transformation of the dichotomous dependent variable into a continuous
variable with the use of the probit link function.
The dependent variable Y is related to the vector of independent variables X and the error term ε
and can be written out in matrix form as:
𝑌 = 𝜙(𝑋𝛽 + 𝜀)
(6.2)
where  is the CDF of the standard normal distribution and β is the vector of coefficients to be
estimated.
124
The inverse transformation of this function gives:
𝜙-1(𝑌) = 𝑋𝛽 + 𝜀
(6.3)
where 𝜙-1(𝑌) is the probit link function. This link function is used for developing the regression
model for household evacuation.
The dependent variable that is related to the safety order of household evacuation actions is
determined by merging the most likely evacuation action and the „wait time before completing
this action‟. It is important to note that Calgary consists mainly of single family houses and that
almost all the houses have basements. In the survey, 73% of respondents mentioned that going to
the basement would be the most likely evacuation action if they were at home.
Considering that this decision was selected by the majority of the respondents and given the wait
time prior to completing the evacuation action, the categories for the dependent variable are
ordered based on the safety level of each category as follows: BasementWithin5min,
Basement5to10min, BasementMoreThan10min and Other. In these dependent variable
categories, BasementWithin5min is the safest option. The safety level decreases when the time it
takes to evacuate to the basement increases. The Other option includes actions such as doing
nothing, driving away to avoid the threat and going out and seeking shelter in a neighbour‟s
house. All of these other actions are not as safe as sheltering in the lowest level of the present
location, which is usually the basement level. Considering these reasons, an ordinal nature of the
dependent variable can be recognized.
125
Parameter estimation procedures for ordered categories with the assumption that coefficients of
the independent variables do not vary between the categories (parallel line assumption) neglect
certain heterogeneous effects of some explanatory factors. To overcome this limitation, a
partially generalized ordinal model is used, where the parallel line assumption is imposed only
for the relevant variables of the model (Williams, 2006). In this study, evacuation options are
analyzed using the generalized ordered probit modelling methodology. The module gologit2 is
used, which is a user written program in Stata (Williams, 2006), originally developed for
generalized ordered logit models. Instead of restriction to a logit model, it also allows the
selection of a more favourable link function. Moreover, its autofit procedure facilitates the
iterative process to categorize the variables that meet the parallel line assumption. This procedure
overcomes the burden of manually setting variables as constrained or unconstrained (Williams,
2006). As explanatory variables, a preselected set of dummy variables are used based on their
significance (Table 6-4).
The model output results are shown in Table 6-5. Based on the results of testing the parallel lines
assumption using a level of significance of 0.05, constraints for parallel lines are imposed for the
following parameters: FromNeighboursH, SmartPhonesH, AgeBelow30, SafestLocationAware,
PickupChildrenH, CallFromaTrustedPersonH and HouseholdSizeThreeOrMore. This is shown
by the identical coefficient being associated with each variable. Constraints for parallel lines are
not imposed for SeekMoreInformation, and it has different coefficients.
126
Table 6-4: Variable Description of the Probit Model for Household Evacuation
Variable Name
SafestLocationAware
Label
1, if an individual knows that the lowest level of the shelter is
generally the safest location during a tornado
SeekMoreInformation
1, if an individual seeks more information about an official
warning before evacuating.
FromNeighboursH
1, if an individual is more likely to receive a warning to
his/her household from neighbours
SmartPhonesH
1, if an individual is more likely to receive awarning to his/her
household from smart phones
CallFromaTrustedPersonH
1, if an individual is more likely to receive a warning through
a call from a trusted person
PickupChildrenH
1, if an individual is more likely to pick up children
AgeBelow30
1, if an individual is younger than 30 years old.
HouseholdSizeThreeOrMore 1, if the household has three or more members
In a gologit2 model, “positive coefficients indicate that higher values on explanatory variable
make it more likely that the respondent will be in a higher category of Y than current one,
whereas negative coefficients indicate that higher values on the explanatory variable increase the
likelihood of being in the current or a lower category” (Williams, 2006).
127
Table 6-5: Results of the Partially Constrained Ordered Probit Model
Household Evacuation
Standard
Coefficient
Options
1
SafestLocationAware
SeekMoreInformation
FromNeighboursH
SmartPhonesH
CallFromaTrustedPersonH
PickupChildrenH
Agebelow30
HouseholdSizeThreeOrMore
Cons
2
SafestLocationAware
SeekMoreInformation
FromNeighboursH
SmartPhonesH
CallFromaTrustedPersonH
PickupChildrenH
Agebelow30
HouseholdSizeThreeOrMore
Cons
3
SafestLocationAware
SeekMoreInformation
FromNeighboursH
SmartPhonesH
CallFromaTrustedPersonH
PickupChildrenH
Agebelow30
HouseholdSizeThreeOrMore
Cons
95% Confidence
Z
P>│z│
Error
Interval
-0.39752
0.481305
0.425351
-0.28255
0.338225
0.319136
0.416343
-0.32902
0.134027
0.202044
0.128273
0.216214
0.123177
0.1339
0.128189
0.166984
0.118895
0.228371
-1.97
3.75
1.97
-2.29
2.53
2.49
2.49
-2.77
0.59
0.049
0
0.049
0.022
0.012
0.013
0.013
0.006
0.557
-0.79352
0.229896
0.001581
-0.52397
0.075786
0.06789
0.089061
-0.56205
-0.31357
-0.001519
0.7327145
0.8491222
-0.041126
0.6006633
0.5703812
0.7436246
-0.095994
0.581626
-0.39752
0.244891
0.425351
-0.28255
0.338225
0.319136
0.416343
-0.32902
-0.1249
0.202044
0.130957
0.216214
0.123177
0.1339
0.128189
0.166984
0.118895
0.229136
-1.97
1.87
1.97
-2.29
2.53
2.49
2.49
-2.77
-0.55
0.049
0.061
0.049
0.022
0.012
0.013
0.013
0.006
0.586
-0.79352
-0.01178
0.001581
-0.52397
0.075786
0.06789
0.089061
-0.56205
-0.574
0.5015628
0.0015192
0.8491222
0.6006633
0.0411268
0.5703812
0.7436246
0.3241993
0.0959943
-0.39752
0.114194
0.425351
-0.28255
0.338225
0.319136
0.416343
-0.32902
-0.36672
0.202044
0.138282
0.216214
0.123177
0.1339
0.128189
0.166984
0.118895
0.23011
-1.97
0.83
1.97
-2.29
2.53
2.49
2.49
-2.77
-1.59
0.049
0.409
0.049
0.022
0.012
0.013
0.013
0.006
0.111
-0.79352
-0.15683
0.001581
-0.52397
0.075786
0.06789
0.089061
-0.56205
-0.81773
0.3852216
0.0015192
0.8491222
0.6006633
0.0411268
0.5703812
0.7436246
0.0842856
0.0959943
As indicated by the negative coefficient associated with the variable SafestLocationAware,
respondents who are aware that the lowest level of shelter is generally the safest location during
a tornado would be more likely to take safer evacuation actions, such as going to the basement
128
within 5 minutes. This is statistically significant with p = 0.05. This indicates the importance of
having an effective public education and information program to raise public awareness about
tornado warnings and to advise citizens of the proper actions that should be taken.
The model gives three coefficients for the variable SeekMoreInformation. The positive
coefficients associated with this variable indicate that those who seek more information about an
official warning before deciding to evacuate are more likely to spend more time before going to
the basement or taking another evacuation action. Having a higher positive value for the first
panel of the variable set indicates that the main impact associated with seeking more information
is the prevention of people going to the basement within a minimum time, such as within 5
minutes (p ≤ 0.05). This again highlights the need for an effective public education program that
informs the public of appropriate and timely evacuation actions.
The positive coefficient of the variable FromNeighboursH indicates that respondents who get
warnings about tornadoes from their neighbours would be more likely to take evacuation actions
that require more time: this is found to be statistically significant with p = 0.05. The variable
CallFromaTrustedPersonH also shows the same effect (p = 0.012). In other words, even if
households get warning information from a phone call from a trusted person, evacuation is not
immediate.
In contrast to these two options, households who get tornado warnings through smart phones
would be more likely to evacuate to the basement quickly, as indicated by a negative coefficient
associated with the variable SmartPhonesH. This is shown to be statistically significant with p =
129
0.022. Smart phones that receive warnings through weather websites or emergency alerts (e.g.
AEA) provide more information on the possibility of the event and urge evacuation actions. This
finding indicates that the direct reception of formal information-rich warnings is more capable of
triggering early evacuation actions than unofficial warnings that come through third parties. This
implies the importance of official warnings in convincing residents to take timely evacuation
actions.
Official warnings should be disseminated through various media sources and take advantage of
the cultural and technological shifts that have occurred in the last decade in issuing warnings
through media such as the Internet, text messages, smart phone apps, social-media alerts such as
Facebook and Twitter, and weather alerts, in addition to traditional media such as television and
radio.
According to the positive coefficient of the variable PickupChildrenH, people who intended to
pick up children from schools or outdoor events are more likely to spend more time before
completing their evacuation actions (p = 0.013). Going out to pick up children under a tornado
warning is a very risky decision, due to a number of uncertainties associated with tornadoes,
such as the unpredictable nature of tornado warning lead times, path direction, touchdown points,
intensity and the risk of staying in a vehicle. Schools themselves need to have programs to
evacuate children to safe places. If people go out of their homes, there is the possibility of more
injuries and fatalities resulting from the oncoming tornado.
130
As shown by the positive coefficient of the variable AgeBelow30, the model shows that
respondents below 30 years of age are more likely to spend more time before taking their
evacuation action (p = 0.013). This indicates that the evacuation decisions and times of this age
group are different from people above 30 years old. The cross tabulation results also show that
the majority of people below 30 years of age are more likely to take other evacuation actions,
such as driving away from home. Special educational efforts should target this age group to
increase their awareness of tornado warnings and safe evacuation actions.
According to the variable HouseholdSizeThreeOrMore, respondents tend to select evacuation
actions with less pre-evacuation time when there are three or more people in the house (p =
0.006). This shows that the households with more members are more concerned and make
quicker decisions than one- or two-person households. The fact is especially important with the
recent increase in households with single-person, single-parent and elderly members. Special
efforts should be made to target these households for increasing their response time to tornado
warnings.
6.12 Multinomial Probit Model for Evacuation of Drivers
The multinomial probit regression is applied to the survey data to analyze the contributing
factors related to driver evacuation actions. The set of evacuation actions of drivers is the
dependent variable. The dependent variable consists of 4 categories, namely DriveAway,
DriveToNearestBuilding, RoadsideShelters and Other. The Other option includes two evacuation
actions: stay inside or under the vehicle, and do nothing. These four categories do not necessarily
be independent. Thus, the use of the less restrictive probit model is more appropriate. The
131
DriveAway option is set as the reference category (base outcome). In these evacuation actions, it
is difficult to recognize a natural order of safety levels. Even if the evacuation action and the
time for each action is merged as in the household analysis, one cannot clearly say that one is
safer than the other. The multinomial probit model, which is a good candidate for an estimator
for situations with unordered categorical dependent variables, is used.
As explanatory variables, a preselected set of dummy variables that are significant in at least one
category of the dependent variable are used, as shown in Table 6-6. The variables which are
significant in each category of the dependent variable can be determined by examining p values
for the Wald z statistics in the regression output (Table 6-7).
Table 6-6: Variable Description of the Probit Model for Driver Evacuation
Variable Name
Label
WatchWarningDifferenceAware 1, if an individual knows the difference between a Tornado
Watch and Tornado Warning
HowToTakeCoverAware
1, if an individual knows how to take cover during a
tornado
LocalradioD
1, if an individual is more likely to get warning from local
radio
OtherRoadUsersD
1, if an individual is more likely to get warning from other
road users
EnvironmentCuesD
1, if an individual is more likely to get warning through
environmental cues
EducationWithDegree
1, if an individual is a degree holder
132
Table 6-7: Results of the Multinomial Probit Regression Model
Driver Evacuation
Standard
Coefficient
Options
1
2
SingleDetachedDwelling
LocalRadioD
EnvironmentCuesD
MobileTextAlertsD
OtherRadioUsersD
_cons
3
SingleDetachedDwelling
LocalRadioD
EnvironmentCuesD
MobileTextAlertsD
OtherRadioUsersD
_cons
4
SingleDetachedDwelling
LocalRadioD
EnvironmentCuesD
MobileTextAlertsD
OtherRadioUsersD
_cons
95% Confidence
Z
P>│z│
Error
Interval
(base outcome)
-0.99884
0.387206
-0.51127
-0.37999
0.574722
-0.4809
0.239857
0.306761
0.22198
0.277701
0.426431
0.327871
-4.16
1.26
-2.3
-1.37
1.35
-1.47
0
0.207
0.021
0.171
0.178
0.142
-1.46895
-0.21403
-0.94634
-0.92428
-0.26107
-1.12352
-0.52873
0.988447
-0.0762
0.164292
1.410512
0.161711
-0.719
0.483251
-0.28999
-0.48186
0.221513
-0.31662
0.216101
0.265554
0.189378
0.243682
0.409381
0.298138
-3.33
1.82
-1.53
-1.98
0.54
-1.06
0.001
0.069
0.126
0.048
0.588
0.288
-1.14255
-0.03723
-0.66116
-0.95947
-0.58086
-0.90096
-0.29545
1.003728
0.081184
-0.00425
1.023885
0.267716
-0.29466
-0.37505
0.540632
-0.46783
1.062533
-1.77463
0.37515
0.368696
0.352649
0.430519
0.514687
0.47852
-0.79
-1.02
1.53
-1.09
2.06
-3.71
0.432
0.309
0.125
0.277
0.039
0
-1.02994
-1.09768
-0.15055
-1.31164
0.053766
-2.71252
0.440621
0.347582
1.231811
0.375968
2.071301
-0.83675
A less stringent significance level of 0.1 is used for the analysis with the intention of grasping
more variables that can influence evacuation decisions. Coefficients with p values greater than
0.1 were considered insignificant and were not included in the final model. Variables with low
statistical significance may be retained in the model if they belong to categorical factors that had
some significant effect in the model Kockelman and Kweon (2002). Thus, some of the variables
that were not significant were still retained in the model as long as at least one of the variables
for the same factor was statistically significant. This approach may reduce the efficiency of the
133
estimates, which was adjusted by using a more liberal confidence level of 90% instead of the
traditional 95%.
The variable SingleDetachedDwelling is significant in categories 2 and 3 (p=0.000 and 0.001)
with negative coefficients. When a tornado warning is given while driving in their residential
area, people who live in single detached dwellings are less likely to take shelter in the nearest
building or seek roadside shelters than selecting the drive away option. These respondents would
be more likely to drive away from the direction of the tornado, probably with the intention of
driving to their homes.
As indicated by its low p value (0.02), the variable EnvironmentCuesD is significant in category
2. The associated negative coefficient says that the odds of seeking shelter in the nearest building
versus driving away when respondents receive warning cues from natural are much lower than
those for people who are not alerted through visible environmental cues. Moreover, they would
be more likely to take other actions, such as stay in or under the vehicle or do nothing. However,
this is found to be only marginally significant with p = 0.092. This may be due to the fact that
visibility of environmental cues, such as funnel clouds, dramatically increases the perceived risk
level, causing them to remain with their vehicle. In the presence of visible environmental cues,
respondents tend to take cover inside or under the vehicle, followed by driving away from the
direction of tornado.
The variable OtherRoadUsersD is found to be significant in category 4 (p = 0.03), with a
coefficient greater than one. This shows that people who receive warnings about tornadoes from
134
other road users are more likely to take shelter inside or under the vehicle than drive away from
the direction of tornado. This may be due to the fact that respondents receive warning messages
at the last moment and may not have much time to think about other options than taking the
quickest vehicle-based evacuation action. Another plausible explanation is that people in this
situation may simply feel safer and more secure in following the crowd behaviour by imitating
other road users‟ reactions. This is different from the results for the household scenario, where
respondents make more independent decisions and take time in seeking shelter when they
receive warnings from a neighbour or trusted person. Drivers take the word of other road users
seriously, and they tend to take shelter immediately.
The significance of the variable MobileTextAlertsD in category 3 indicates that the odds of
seeking roadside shelters versus driving away by people who receive warnings through mobile
text alerts are much lower compared to people who do not get warnings through mobile text
alerts (p = 0.05). Receipt of these mobile text alerts in advance can give extra time for driving
away rather than seeking roadside shelters. However, this type of warning is not always possible.
People may not be able to read their text messages while driving, or there can be alert messages
that get delayed due to issues with mobile networks or e-mail system failure caused by the severe
weather conditions. For these reasons, it is beneficial for drivers to have a more reliable source of
information for warnings instead of relying solely on mobile text alerts.
Local radio is a very good source of warnings. It disseminates warnings issued by Environment
Canada as well other authorized AEA users. According to the model output results, people who
get warnings through local radio are more likely to take roadside shelters compared to the drive
135
away action (marginally significant with p = 0.068). It is difficult to find a reason why the
variable LocalradioD is significant for roadside shelter action, whereas it is not significant for the
nearest building option. One possible reason is that people who receive a warning through radio
may be mainly freeway drivers with long commutes, who are more likely to listen to radio while
driving than other drivers. In the case of a tornado warning, users of a freeway are usually far
from buildings, but closer to roadway shelters located around freeway interchanges.
6.13 Conclusion
This study uses statistical data analysis methods to examine the factors that contribute to the
evacuation behaviour of households and drivers during a tornado emergency. There are number
of benefits associated with this questionnaire survey. The results of this survey are important in
understanding Calgarians‟ stated responses to the present tornado warning, communication and
evacuation system. It is always good for individuals to have a plan of action, in order to be able
to act quickly and efficiently (Monfredo, 2008). Participation in this survey also became a good
exercise for respondents to think about possible warning sources and evacuation methods. As
highlighted by Nirupama and Maula (2013, p.51), community participation can bring about “a
comprehensive and accurate appreciation of people‟s perception regarding hazards, risk,
vulnerability and resilience …”.
The limitations of this survey are mainly associated with how different people perceive risk. The
hypothetical scenarios of tornado warnings to households and drivers do not provide information
on the strength or intensity of the tornado. For this reason, the behavioural responses of
individuals who assume the highest level of risk can be different from those who assume a low
136
level of risks. The results of the survey, however, show the possible responses of Calgarians and
indicate improvements can be made to the existing situation.
It is worth noting that people‟s actual behaviour during a tornado emergency may be different
from their intended behaviour. A rational behaviour may not always be possible under the
panicked nature of a short-notice evacuation. Moreover, there can be a portion of the population
who totally ignore the warning received, assuming that it will not happen to them. However, the
behavioural responses that emerge from the survey do provide direction for efforts in mitigating
the impact of tornadoes at the level of the individual, as well as that of the community.
137
Chapter Seven: Impact of False Warnings and Missed Events on Tornado Warning
Performance
False tornado warnings and missed events bring negative consequences to the public. This
chapter is devoted to analyzing the false tornado warnings and missed events in the Canadian
Prairies to obtain insights on the key performance indicators such as the probability of false
warning, probability of detection and probability of missed events. Furthermore, a logical basis
for the warning decision making is developed based on decision theory and discussed in detail. A
fundamental inequality of decision making for issuing tornado warnings is proposed. False
warning and missed event probabilities are analyzed using the data from the Canadian Prairies.
Bayes‟ theorem [see for example, Jordaan (2005)] based inferences are also made to better
understand the factors behind false warnings and missed events.
7.1 Warning Decisions
“Warnings are the culmination of a sequence of actions ... that act to alert the public to a
heightened probability of high-impact weather, minutes, hours or even days in advance”
(Stensrud et al., 2009, p.1487). Dealing with uncertainties associated with the sequence of
actions in detecting and verifying a tornado has a huge impact on decision-making for warning
issuance and it is inevitable to lead to wrong decisions in some situations. Warning decisionmaking is quite challenging for forecasters especially due to the negative consequences
associated with wrong decisions such as false warnings and missed events. The underlying
factors for false warnings and missed events must be better understood before forecasting
operations can be more clearly focussed to improve the warning performance.
138
7.2 True Warnings, False Warnings and Missed Events
Evaluation of tornado warnings can be done using the 2x2 Contingency Table (Table 7-1). Given
a severe weather event, the joint occurrence probabilities: True Warning (T∩W), Missed Events
(T∩W), False Warning (T∗∩W), can be obtained using the contingency table. Here we define
the „No tornado, no warning‟ scenario as Status Quo (T ∩W). These associated probabilities
p,q,r,s respectively sum to 1.0.
Table 7-1: 2x2 Contingency Table
Tornadoes Observed
Tornadoes Forecasted Yes (W)
(or Warned)
No (W)
Yes (T)
No (T)
p True Warning
r False Warning
q Missed Event
s Status Quo
(Modified from Barnes et al. (2007))
Each category of the contingency table that describes the various status of the warning process is
discussed in the following sections.
7.2.1 True Warning
A true warning for a tornado is a clear communication to the public to evacuate to safer places
prior to an actual occurrence. True warnings minimize deaths, injuries and some property
damage (e.g. to enroute vehicles) that could have happened in the absence of a warning. The
forecaster‟s aim is to issue a warning prior to each and every actual tornado occurrence. Tornado
is a disaster that has a very limited lead-time for warning and the reaction to a tornado warning is
an activity undertaken under time pressure. People should not wait until they get complete
139
information about an approaching tornado. It is sufficient to have a warning from a reliable
source to initiate evacuation actions.
Advice given in a warning, especially its wording (Farley, 2007), has significant impacts on what
people do. Therefore a true warning should be capable of conveying the message clearly and
immediately. The effectiveness of warnings can be improved by increasing the appropriateness
of their information content (Papastavrou and Lehto, 1996). For example, addition of a map
showing the warning area provides easily understood information to the public.
It is noteworthy that even when accurate warnings are issued and properly communicated to the
public, there are people who choose not to heed warnings. This may happen when people
frequently experience tornado warnings that do not result in tornadoes, or they are not aware of
the damage that can be caused by tornadoes, or due to the belief that the tornado will not hit their
house.
7.2.2 False Warning
A false warning can be considered to be a situation when the public is warned by an authorized
service about a tornado and one actually does not occur in the defined area. False warnings
create problems related to the credibility of, and future response to, warnings. High probabilities
of false warnings also can lead officials to refrain from issuing warnings or to delay warnings.
Precise forecasting of a tornado is quite challenging due to the chaotic nature of thunderstorm
development and limitations in the abilities to track that behaviour in advance. Even if a warning
is issued for an actual tornado, verifying the occurrence is also problematic due to several
140
reasons such as reduced resource allocation for verification compared to forecasting and warning
issuance, and sparsity of observational data due to lack of spotters in the field. These problems
associated with verification of event occurrences as well as the uncertainties in forecasting
science and technology can increase the count of the false warnings (Barnes et al., 2007).
Forecasters are aware that the potential consequences of being wrong about a tornado when there
is none (false warning) are much lower than being wrong about a tornado when there is one
(missed event). This notion influences forecasters to issue warnings even when they are
uncertain about an actual tornado occurrence. However, issuing warnings without a real need can
induce negative consequences. For example, analyzing the influence of false tornado warnings
on casualties, Simmons and Sutter (2009) have shown that higher probabilities of false warnings
significantly increase fatalities and injuries from future tornadoes.
There are various arguments about the false warning issue in the literature. According to
Edwards and Lemon (2002, p.234), “false alarms can be costly and can serve to slow or prevent
a future response”. The cost of a false warning includes costs related to issues such as business
closure, road closure, shut down of critical facilities and evacuation of the working population.
Furthermore, it can induce disutility among individuals in terms of inconvenience, fear for self
and loved one‟s, or cause injuries, even deaths during emergency evacuation.
Repeated occurrence of false warnings can lessen public confidence about the warning system
and the immediate response to future warnings. Specially, it can reduce the public‟s compliance
with future tornado warnings possibly causing „cry-wolf‟ syndrome. Although that is the
141
intuition behind the false warning issue, there are also counter-arguments. Mileti and Sorensen
(1990, p.3-3) summarizing the cry-wolf syndrome „myth‟ in a report of the Federal Emergency
Management agency stated that:
“...the effectiveness of people‟s response to warnings is not diminished by what
has come to be labelled as „cry wolf‟ syndrome, if they have been informed of the
reasons for previous „misses‟. Obviously, there should be a negative effect on
subsequent public response if false alarms occurred frequently, if no attempt was
made to explain why there were false alarms, and if the cost of response is high.
Yet, false alarms, if explained, may actually enhance the public‟s awareness of a
hazard and its ability to process risk information in subsequent warning events.
False alarms are better viewed as opportunities for conveying information than as
problems”.
According to Simmons and Sutter (2009), people respond to warnings when there is a positive
value for the net expected utility that can be calculated by deducting the cost of response from
the losses avoided by the response. Higher probabilities of false warnings than true warnings can
cause the net value of responding to a warning to be negative. False warnings can impact the
level of risk that may be accepted by the public in responding to a warning. Assuming the
highest level of risk and taking immediate action is not always desirable for people who
experience disutility associated with the repeated occurrence of false warnings. Therefore,
forecasters need to pay attention to issuing a tornado warning only when there is an imminent
threat.
142
7.2.3 Missed Event
There are situations when no warning is issued prior to an actual tornado occurrence. A missed
event is a situation where a tornado touchdown occurs without an advance warning being issued
for the area of touchdown. The non-issuance of a warning can be due to several reasons such as
lack of information available for a forecaster to issue a warning, inability or wrong judgment in
recognizing intense rotations that could lead to a tornado or the absence of ground level
information support to verify a threat. In addition, there are situations where tornado touchdowns
are reported just outside the warning areas and hence recorded as missed events.
Brotzge and Erickson (2010) have shown that tornado events with characteristics such as weak
(e.g.F0), short-track, singular, nocturnal, far from radar, or the first tornado report in an outbreak
have the greatest chance of not being detected sufficiently early to issue a warning. In fact, these
are the tornado occurrences that the public is least likely to be aware of or observe. They further
indicated that the areas that experience weak, solitary and/or nocturnal tornadoes are likely to
have missed events.
Experiencing a high proportion of missed events is a critical issue to be considered. A missed
event can lead to a catastrophe when a powerful tornado occurs in a highly-populated area. The
intangible costs of a missed event can be higher due to more deaths and injuries being caused
than from an event about which a warning has been issued. The forecaster has to make an effort
to keep the number of missed events, especially of high intensity tornadoes (e.g. ≥ F3) extremely
low and ideally zero.
143
7.2.4 Status Quo
The ultimate wish of warning decision makers, disaster management officials, as well as the
public is to have the „No warning, no tornado‟ scenario for every severe thunderstorm event.
This status quo infers the correct detection of the situation that there is no tornado potential
within a thunderstorm, thus, no warning is required. Here we assume that the forecaster looks
for conditions favourable for the development of tornadoes in every severe weather event. The
status quo situation does not cause major damage, inconvenience or financial cost to the public
except the baseline impact from the thunderstorm. Maintaining an accurate status quo ensures the
clear recognition of the conditions that do not require warning issuance. This is also important to
keep the number of false warnings very low.
7.3 Warning Spectrum and the Rationale of a Warning
Barnes et al. (2007) developed a conceptual model that presents a more general depiction of
warnings for possible events (Figure 7-1). Instead of having a yes-no categorization of warnings,
this spectrum is used to demonstrate a range of accuracy of warnings. This model is especially
good at assessing forecasters‟ performance skills in recognizing events that require warnings to
be issued. With this range of accuracy, a tornado occurrence under a severe weather warning is
an event that occurred but more severe than just a thunderstorm whereas an event like a strong
wind under a tornado warning is recorded as an event that occurred but less severe than a
tornado.
144
Figure 7-1: Conceptual Model for Warning Accuracy
Source: (Barnes et al., 2007)
(c)American Meteorological Society. Reprinted with permission.
Descriptions of warnings that go under false records were analyzed to see what was occurring
within the atmosphere at a time when a warning was issued. According to warning records in the
Prairies, more than 50% of false warnings were based on the Doppler radar recognition of
intense rotation or a possible tornado. Warnings also have been issued for downstream areas,
considering the rapid movement of severe thunderstorms. In all false warnings records, a severe
thunderstorm event has occurred although there is no report of a tornado. There are a number of
cases that the severe thunderstorm has moved a step forward developing funnel clouds or
possible tornadoes. These events less severe than tornadoes are actually the „nearly misses‟ or
the „close calls‟. Recognition of risk levels of these close calls helps to see the underlying basis
145
of warning issuance. Such a warning issuance when in doubt manifests the public alert that it is
better to be safe than sorry.
7.4 Factors behind False Warnings and Missed Events
Tornado detection and warning is much more challenging especially considering the chaotic
nature of thunderstorms and tornado development. There is a trade-off between the efforts to
achieve timeliness and the accuracy of warnings. Generally, every storm begins as a non-severe
storm and some develop to the severe stage leading to tornado formation. The initial stage of a
tornado as a funnel cloud is not a severe condition although it can be violent when it touches
down. The forecasters have to spend much effort in their decision to warn or not. Furthermore,
determining a threshold risk level for tornado warning issuance is mainly situation dependent.
The bottom line is when in doubt, warn the people.
Forecasters‟ aim is to identify the severity of the future storm as early as possible in order to
warn the public with a sufficient lead-time. Considering the urgency and the importance, they
issue warnings in advance so that the public have time to prepare for evacuation. For example,
appearance of a funnel cloud during a period when there is a large gathering for an outdoor event
can be an urgent and important call for a tornado warning. The best available information at the
time of warning issuance may not be sufficient to guarantee an actual tornado event. Forecasters
have to issue a warning for the potential danger from a weather event such as the intense rotation
from a storm observed from Doppler radars. Warned situations with high tornado potential may
also change suddenly and the severe weather environment may recover fast. Despite the
information uncertainty, tornado warnings are issued to avoid possible threats to lives and
146
property. Consequently, some portion of warnings is recorded as false warnings. In contrast to
that, unwarned situations with low tornado potential may lead to tornado outbreaks at a later
stage resulting in missing events.
Besides radar and satellite based observations, and automated guidance products for forecasting,
eyewitness observations play a major role in warning decision making. In the Canadian Prairies,
active spotter or weather watcher involvement is limited only to some populated regions in the
country. Therefore, obtaining local level information about an imminent tornado prior to issuing
a warning is not always possible. Lack of ground truth reports and the sole reliance on Doppler
radar based information in issuing warnings lead to missed events.
There are coverage issues that lead to increase the number of false warnings and missed events.
The Canadian Prairie Region is a very large area in terms of forecasting. However, only “a
single severe weather forecaster is responsible for the provision of warnings for the area
coverage of about ten radars” (Joe et al., 2012, p.49). Paying attention to several radar data
reports simultaneously does not help to provide a deep analysis on the behaviour of a given
storm. Although a tornado potential is detected, and a warning is issued, atmospheric conditions
may change rapidly and the storm may not spawn an actual tornado. According to Joe et al.
(2012), one of the main reasons for missed warnings is that the forecaster is so intent on one
thunderstorm that they forget about the others. This is a loss of situational awareness that has
implications for public safety.
147
From the forecasters‟ point of view, the inability to verify a tornado occurrence is a problem
associated with false warnings. The Canadian Prairie Region has many sparsely populated areas
with population density of less than one person per km2 (Joe et al., 2012). Therefore, it is not
unlikely that a tornado might go undetected to the human eye. Even if a tornado occurs following
a warning, it is counted as a false warning if the occurrence cannot be verified. Discussions
quoted from tornado warning bulletins justify the reasons for warning issuance.
“AT
5:30
PM
DOPPLER
RADAR
INDICATED
AN
INTENSE
THUNDERSTORM 40 KM EAST SOUTHEAST OF PORTAGE LA PRAIRIE,
NEAR ELIE. RADAR INDICATES ROTATION WITH THIS STORM AND
MAY BE CAPABLE OF PRODUCING A TORNADO.”
(TORNADO WARNING ISSUED AT 5:38 PM CDT TUESDAY 13 JULY
2010.)
“AT 5:25 PM CST RADAR INDICATES A POSSIBLE TORNADO AS
STRONG ROTATION IS DETECTED JUST TO THE NORTHWEST OF
MAYFAIR. THIS STORM IS MOVING EASTWARD AT 30 KM/H.THIS IS A
WARNING THAT SEVERE THUNDERSTORMS WITH TORNADOES ARE
IMMINENT OR OCCURRING IN THESE REGIONS. MONITOR WEATHER
CONDITIONS. TAKE IMMEDIATE SAFETY PRECAUTIONS.”
(TORNADO WARNING ISSUED AT 5:26 PM CST MONDAY 18 JULY 2011.)
The possibility that people will question false warnings can be reduced if the tornado warning
bulletins are supplemented with reasonable explanations about the basis for warning issuance.
148
7.5 Household Decision Tree
Given a warning, the decision to be made by a household to respond or not to respond (evacuate
or not to evacuate) and the related disutility can be illustrated using a decision tree as shown in
Figure 7-2.
Figure 7-2: Household Decision Tree
The base disutility U0 =0 caused to a household is associated with the condition where people do
not respond to the warning and a tornado does not occur. The next higher disutility U1 occurs
when people respond to the warning but a tornado does not occur. This disutility is due to
inconvenience and injuries that can occur during evacuation. A much higher disutility U2 occurs
when people respond to the warning and a tornado impacts on the household. Even if properly
responded to the warning, there can be property damage or loss, or personal injuries that bring
disutility to a household. The highest disutility U3 occurs when a household chooses not to
149
respond or is unaware of the warning and a tornado impacts the household. Compared to U2, U3
has a much higher added portion as there can be consequences such as deaths and injuries.
Let the probability of a tornado given that a warning has been issued be P(T/W) or p where T
denotes a tornado and W denotes the conditioning event that a warning is given. Let the
probability of no tornado occurred given that a warning has been issued be P(T′/W) or (1-p)
where T′ denotes an absence of a tornado. The probability that a household responds to a
warning is denoted by r. The probability that a household does not respond to a warning either
intentionally or because it was unaware of the tornado warning is (1-r).
The expected disutility of responding to a warning can be written as
E (R) = p(U2) + (1-p)(U1).
(7.1)
The expected disutility of not responding to a warning is
E (R′) = p(U3).
(7.2)
Applying decision analysis (Raiffa, 1968),
the household will choose to respond if the expected disutility of not evacuating is greater than
the expected disutility of evacuating, i.e.
E (R′) >E(R).
(7.3)
Substituting for E(R) and E(R′) from (7.1) and (7.2) we obtain
p(U3 ) > p(U2) + (1-p)(U1)
(7.4)
which can also be written as
𝑝>
1
1+
U 3 −U 2
U1
.
(7.5)
150
This can be considered as a relationship that shows the underlying factors behind a household‟s
decision to respond. The term (U3-U2) represents the additional disutility or the consequences of
not responding to a tornado warning. When this is higher even a low probability of a true
warning is sufficient to trigger evacuation response. In contrast to this, when the negative
consequences of responding to a false warning denoted by U1 are higher, it is necessary to have a
higher value for p to initiate response actions.
Since U3>U2, we can consider two cases (i) (U3-U2)> U1
U 3 −U 2
U1
and (ii) (U3-U2)< U1. In case (i),
>1 and then the RHS of (7.5) is less than 0.5. Since U1 is small and U3-U2 is high, it is
unlikely that case (ii) will occur. However, in this case,
U 3 −U 2
U1
<1 and then the RHS of (7.5) is
greater than 0.5. Thus evacuation response will occur at a value of p given by (7.5) which is
higher than 0.5. If U3 ≤ U2, the RHS of (7.5) is greater than one and there is no value of p for
which evacuation will occur. Only an irrational household decision maker will take U3 ≤ U2.
More conservative households are likely to make their decision on the basis of
E (R′) + k1σ(R′) >E(R)- k2σ(R)
(7.6)
where σ(R) and σ(R′) are the standard deviations of the disutility and, k1 and k2 are positive
constants with k1 > k2 . It is beyond the scope of the thesis to estimate k1 and k2.
The fraction of households that respond on the basis of (7.5) or (7.6) is taken to be r in Section
7.6.
151
7.6 Warning Decision Tree
Given that a tornado detection and warning system is in place, the decisions to be made by the
forecasters, the probabilities related to the above-described events, the probability of the actions
taken by households (respond or not), and the related disutility can be illustrated in a decision
tree (Figure 7-3). In this tree, r is the percentage of households that will respond (evacuate)
given a warning. In contrast with the Section 7.5, this decision tree has considered the
community (n number of households) response probabilities and the disutility terms for the
whole community. However, n has omitted as it does not appear in the final derivation.
The decision to be made by forecasters is to issue a warning or not to issue a warning at a given
time. The status quo is when the decision is not to issue a warning and a tornado does not occur.
We take that baseline disutility (U0) as zero. The next higher disutility U1 occurs during a false
warning (a warning is given but a tornado does not occur) due to the inconvenience of the
evacuation and any injuries that may occur. Those who intentionally do not evacuate or are
unaware of the warning remain at a disutility of zero. A much higher disutility U2 occurs when
the decision is to issue a warning and the household evacuates (say to the basement). However,
even with the evacuation, property damage will obviously occur as well as a reduced number of
injuries and deaths. The households that choose not to respond or are unaware of the warning
will have a much higher combined disutility U3 as the injuries and deaths will be extensive. The
disutility of a missed event (a tornado occurs though a warning has not been issued) is also U3.
All the above disutility terms are taken as positive values (similar to cost).
152
Figure 7-3: Warning Decision Tree
Let the probability P(T/W) of a tornado given that no warning has been issued be p1 where T
denotes a tornado and W denotes the conditioning event that no warning is given. Let the
probability P(T/W) of a tornado given that a warning has been given be p2 where W denotes the
conditioning event that a warning is given. Even if a warning is issued, the decision to respond to
the warning is decided by the household as discussed in 7.5. The probability that a household
responds to a warning is denoted by r. The probability that a household does not respond to a
warning either intentionally or because it was unaware of the tornado warning is (1-r).
The expected disutility of giving a tornado warning
E (W) = rp2(U2) + r(1-p2)U1 +(1-r )p2 (U3)
(7.7)
can also be written as
153
E (W) = P(T/W) [U3(1-r)+ (U2-U1)r]+rU1 .
(7.8)
The expected disutility of not giving a tornado warning is
E(W)= P(T/W)(U3) =p1(U3).
(7.9)
The forecaster should choose to issue a warning if the expected disutility of not giving a warning
is greater than the expected disutility of giving a warning, i.e.
E(W)> E (W).
(7.10)
Substituting for E(W) and E (W) from (7.8) and (7.9) we obtain
P(T/W)(U3) > P(T/W) [U3(1-r)+ (U2-U1)r]+rU1
(7.11)
which simplifies to
[P(T/W)/ P(T/W)]-1
>r {[(U2/U3)-(U1/U3)-1]+[(U1/(U3 P(T/W))]} .
(7.12)
The disutility value of U1 due to a false warning is very much smaller than the U3 value of a
missed event that damages property and impacts on people through injuries and fatalities on a
large scale. Therefore, by neglecting terms (U1/U3) and [U1/(U3P(T/W))] we obtain
[P(T/W)/ P(T/W)]-1 > r [(U2/U3)-1]
(7.13)
which can also be written as
r >{1-[P(T/W)/ P(T/W)]}/ [1-(U2/U3)].
(7.14)
This can be considered as a fundamental inequality of decision making for tornado warnings.
Here, the term [1-(U2/U3)] is always positive. The term {1-[P(T/W)/ P(T/W)]} varies depending
on the ratio of P(T/W)/ P(T/W). It is positive (but less than one) when P(T/W)/ P(T/W) is
154
smaller than one and negative when P(T/W)/ P(T/W) is greater than one. While one would
expect P(T/W) to be greater than P(T/W), it is not always the case.
If P(T/W)>P(T/W),the right side of (7.14) is always negative since U2<U3. Thus any response
probability r or even a complete non response (r=0) is sufficient to justify a warning. When
P(T/W)< P(T/W), the response probability has to be higher than a positive value to justify a
warning. When P(T/W)/P(T/W)>U2/U3, the necessary value of r is between 0 and 1. Since
U2<U3, this condition will always occur when P(T/W)>P(T/W); However it can also occur for
some situations when P(T/W)<P(T/W).
When P(T/W)<P(T/W), we can consider that the possibility of a wrong decision for warning
issuance is relatively low. Based on (7.14), in such a system tornado warning is effective if the
response probability is greater than a value between 0 and 1. In other words, a reliable system
that fulfils its warning responsibility always expects a high response probability from the public.
Improved technology assists warning decision makers to correctly recognize the situations that
require to issue a warning or not to issue a warning. It also increases the probability P(T/W) that
results from a correct warning decision for a tornado while reducing the probability P(T/W)
associated with a wrong decision for not warning the public for an actual tornado. With the
resulting decrement of the ratio P(T/W)/ P(T/W), the system expects a higher response
probability from the public to make the warning decision more effective. Therefore, application
155
of sophisticated techniques for improving the tornado detection and warning process should
always be coupled with hard and soft measures to enhance the public response.
It is true that the disutility sequence of U1 <U2 <<U3 cannot be generalized for a weak tornado
event that is rarely experienced by a person in the Canadian Prairies. However, the conclusions
derived from the expected disutility concept do provide important aspects to consider when
improving the overall warning system.
7.6.1 Application of the Inequality for the City of Calgary
The two conditional probabilities P(T/W)/ P(T/W) are available from the Prairie data as shown
in Table 7-4 (Section 7.7) . The fraction of responding r is available from the Stated Preference
Survey. Then the ratio U2/U3 can be estimated as an inequality.
P(T/W) =0.107; P(T/W)=0.126; (See Table 7-4)
r = 0.95 for households and r= 0.98 for drivers
Substituting the values to the inequality in (7.14) gives inequalities
U2/U3 <0.84 or U3/U2 >1.19
(7.15)
for households and
U2/U3 <0.85 or U3/U2 >1.18
(7.16)
for drivers.
Based on this, it can be concluded that the forecaster should choose to issue a warning if U2/U3
ratio is lower than 0.84 or if U3/U2 ratio is greater than 1.19 for households and, if U2/U3
ratio is lower than 0.85 or if U3/U2 ratio is greater than 1.18 for drivers.
156
The above disutility terms can be monetized as the cost of injury, cost of death and the cost of
property damage. The cost of injury involves the components such as the cost of medical
expenses, cost due to lost productivity such as temporary work loss or long term disability and
the cost due to the lost quality of life (Cropper and Sahin, 2009). The cost of death which is
equivalent to the value of avoided mortality is much higher than the cost of injury.
For a deadly tornado, having some level of disutility due to the property damage is unavoidable
and the cost is generally covered by the insurance. However, at the community level, the
disutility of deaths and injuries are much higher and, and so do costs. Preventing the deaths and
injuries by any means has a much higher reduction of the disutility. Therefore, the derived
inequalities showing that the disutilities from injuries and fatalities are likely to be at least 1819% higher than only property damage are reasonable.
The inequality (7.14) can also be written as
U2
U3
<1−
1−[𝑃(𝑇/𝑊)/ 𝑃(𝑇/𝑊)]
𝑟
.
(7.17)
It is also known that the U2/U3 ratio is between greater than zero and less than one. Therefore,
the maximum value of the U2/U3 ratio obtained from (7.17) must satisfy the conditions
[P(T/W)/ P(T/W)] + r >1
(7.18)
and
P(T/W)/ P(T/W) < 1.
(7.19)
157
A graph (Figure 7-5) was plotted for the maximum U2/U3 ratio selecting the ranges for the r
value and the P(T/W)/ P(T/W) ratio ( or P ratio) that satisfy (7.18) and (7.19).
Figure 7-4: Map of the Maximum U2/U3 Ratio
This can be considered as a reference to estimate the disutility ratio corresponding to different
values of P ratio and r. Where there is a higher value for the P ratio, warning has to be issued
even when there are high U2/U3 ratios. In other words, a warning system with higher
probabilities of false waning has to issue tornado warnings even when the disutility of deaths and
injuries are minimal.
158
7.6.2 A Detailed Illustration of the Decision Tree
The disutility increases with the F-scale intensity of tornado events. A more detailed decision
tree model with disutility terms associated with different intensity levels can be used to analyze
the warning decision making process. A sample illustration of such a model is shown in Figure
7-5.
Figure 7-5: Warning Decision Tree: Detailed Illustration
Such a breakdown of probability and disutility terms helps to analyze the decision-making
process in detail using the expected disutility minimization concept and develop response
probability ranges to make the warning decision more effective.
159
7.7 Analysis of Tornado Warnings in the Canadian Prairies
The Prairie and Arctic Storm Prediction Centre (PASPC) handles severe weather watches and
warnings for the Prairies. The PASPC uses the „15km and 15 minutes‟ rule to categorize severe
weather events. Under this rule, “several reports of one specific event type that occurs within 15
km and 15 minutes of one another are counted as one event.” (PASPC, 2010,p.3). A severe
weather event is defined when at least one of the following conditions is satisfied:

Wind gusts of 90 km/h or greater, which could cause structural wind damage;

Hail of two centimetres (cm) or larger in diameter;

Convective rainfall of 50 mm or more expected within one hour;

Tornadoes.
According to this criterion for alerting the public, a tornado warning is issued … “when a
tornado has been reported; or when there is evidence based on radar, or from a reliable spotter,
that a tornado is imminent” (EC, 2012, para.18). The warnings area is generally an
administrative area such as city, county or a rural municipality. On average, 203 severe events
are reported over the Prairies during a summer period (Table 7-2). Of these, an average of 36
severe events is reported as tornadoes.
Table 7-2: Average Number of Summer Severe Weather Reports (1984-2006)
Event type
Alberta
Tornado
13
Hail
39
Wind
12
Rain
10
All events
74
%
36.5%
Source: (Taylor et al., 2012)
Saskatchewan
Manitoba
Total
14
33
20
7
74
36.5%
9
25
13
8
55
27%
36
97
45
25
203
100%
160
In this section, we intend to analyze the tornado warning data in the Canadian Prairies. A
database containing the tornado warning records in the Canadian Prairies from 2003 to 2012 was
obtained from EC (EC, 2013e). It contains information regarding tornado warning issuance and
end notices, the areas warned, and discussions of the reasons for tornado warning issuance.
Records of actual tornado occurrences from 2003 to 2012 were also obtained from EC (EC,
2013f). They are the verified tornado occurrences. The number of severe weather events in each
year was obtained from the reports of Prairie summer severe weather event climatology and
verification results from 2003-2012. Summer severe weather event climatology and verification
reports are published annually by PASPC examining summer severe weather events for the three
Prairie Provinces.
Several caveats were identified in analyzing the warning records. The following criteria were
used to recognize individual warning issuances and false warning records associated with them.

There are cases that tornado warnings are issued for one or several warning areas. In this
case, a warning for one area is considered as one warning and they are counted separately
in determining the number of warning records. If there is a warning for an area that has
experienced a tornado touchdown within the warning period, it is counted as a true
warning. The absence of tornado reports in other areas that have been warned is
considered as false warnings even if the nearby area experienced a tornado.

There are some cases that a warning is issued either simultaneously as the tornado formed
or minutes after initial tornado formation and touchdown but prior to tornado dissipation.
These warnings are also counted as true warnings considering the possibility for future
touchdowns in the area.
161

Sometimes, it is necessary to add new warning areas considering the movement of the
same storm to downstream regions. They are also counted as true warnings.

According to the „15 kilometres and 15 minutes‟ rule, it is assumed that each warning is
associated with one severe weather event.

It is assumed that the issuance of one tornado warning is related to one tornado
occurrence (or not) and there are no multiple tornadoes under one tornado warning. This
complies with the general nature of warnings for isolated tornado events in Canada.
Individual tornado occurrence reports (EC, 2013f) were compared with the tornado warning
records (EC, 2013e) to determine the joint occurrence of a tornado and a warning issuance.
Absence of such an intersection is counted as a false warning record. Furthermore, the number of
tornado occurrences without actual warnings were counted as missed events. A summary of the
data records is presented in Table 7-3. The logical relations of the data sets can also be shown
using a Venn diagram (Figure 7-3).
Table 7-3: Tornado Warning and Occurrence Records in the Canadian Prairies
Year
Total
No. of True Warnings (p)
126
No. of Missed events (q)
191
No. of False Warnings (r)
872
Status Quo (s) (See Section 7.2.4)
1594
No. of Tornado Warnings (p+r)
998
No. of severe weather events (p+q+r+s)
2783
162
Figure 7-6: Venn Diagram of Tornado Warning and Occurrence Records
These datasets were used to analyze and derive probabilities associated with tornado warnings.
7.7.1 True Warning, False warning, Detection Probabilities given a Severe Weather Bulletin
The following equations assume that a severe weather bulletin (not a warning) is in effect. Here,
p,q,r and s are defined in Tables 7-1 and 7-3 and
T = the event that a tornado occurs
T = the event that no tornado occurs
W = the event that a warning is issued, and
W= the event that no warning is issued.
The probability of true warning or the probability of a tornado given a warning can be written as
P(T/W) = p/(p+r).
(7.20)
The probability of false warning, or the probability that a tornado did not occur given that a
warning was issued is
163
P(T/W) = r/(p+r).
(7.21)
The probability of a tornado without a prior warning (missed event) can be written as
P(T/W) = q/(q+s).
(7.22)
The probability of detection can be written as
P(W/T) = p/(p+q).
(7.23)
The probability of false detection can be expressed as
P(W/T) = r/(r+s).
(7.24)
The above conditional probabilities for the Prairie data in 2003-2012 are shown in Table 7-4.
Table 7-4: Probabilities related to Tornado Occurrence and Warning Records (2003-2012)
Probability
Percentage (%)
Probability of True Warning P(T/W)
12.6%
Probability of False Warning P(T/W)
87.4%
Probability of Missed Event P(T/W)
10.7%
Probability of Detection P(W/T)
39.8%
Probability of False Detection P(W/T)
35.4%
The probability of false warning for the ten-year period is 87.4%. This figure shows that, nearly
nine out of ten tornado warnings are recorded as false warnings. Furthermore, Canadian Prairie
region has a probability of detection of around 40%. In other words, only four out of every ten
tornado events are covered by warnings.
These probabilities are not indicators of a good
warning system, although it is not too surprising when compared with the well-established US
tornado warning system. Despite technological advancements, strong spotter networks for
164
ground level observation and verification, and all other efforts to reduce it, the false warning
probability in the US is 75% (Barnes et al., 2007; Brotzge et al., 2011).
7.7.2 Bayes’ Theorem based Inferences
Bayes‟ theorem (Lindley, 1965; Olson, 1965) that connects conditional probabilities to their
inverses can also be used to estimate the probability of false warnings and the probability of
missed events for disaster forecasts. Wickramaratne et al. (2011) have illustrated the use of
Bayes‟ Theorem in estimating false warning probabilities for tsunami given an earthquake. Here,
we use the Bayes‟ theorem to provide some insight into the probabilities related to tornado
forecasting and warning. Given a severe weather event, warning issuance can be related to either
the occurrence or non-occurrence of a tornado event.
Using the Bayes‟ theorem, the probability of false warning is
P(T/W) = [P(W/T) P(T)] / P(W)
(7.25)
which can also be written as
P(T/W) = [P(W/T) P(T)] / [P(W/T)P(T)+P(W/T)P(T)].
(7.26)
The reciprocal of (7.26) is
1/P(T/W) =1+ [P(W/T) P(T)] / [P(W/T)P(T)].
(7.27)
Analysis of (7.27), gives some insights in the terms that a forecaster should pay attention to
reduce the false warning probability. In this equation, P(T) and P(T) values are fixed terms for a
set of severe weather events in a considered time period. If the term P(W/T) can be increased, it
eventually leads to decrease the P(T/W). Furthermore, efforts to decrease the P(W/T) which is
165
known as the probability of false detection can result in reducing the value of P(T/W). This
infers that the higher probability of detection and the lower probability of false detection
decrease the probability of false warnings. This is an important factor to be considered when
using the probability of false warnings as a performance measure.
7.7.3 True Warning and Missed Event Probabilities based on Tornado Intensity
Analysis of the tornado data in the Canadian Prairies for a ten-year period shows that the
P(T/W)=10.7% is lower than P(T/W)=12.6%. The intention in this section is to obtain the
probability values related to the inequality derived in Section 7-6 considering the intensity of the
tornado. Canadian Prairie data was further analyzed to obtain true warning P(T/W) and missed
event P(T/W′) probabilities based on the intensity level as shown in Table 7-5. Unavailability of
data to calculate a probability value is shown by a hyphen (-). These probability values indicate
that the P(T/W) decreases with the tornado intensity level. The chance of missing a warning for
a high intensity tornado is smaller than the chance of missing a warning for a low intensity
tornado. In all cases, P(T/W) < P(T/W) for all the intensity levels.
Table 7-5: True Warning and Missed Event Probabilities based on Intensity (2003-2012)
P(T /W)
P(T /W)
P(TF0 /W) =8.57%
P(TF1 /W) =1.62%
P(TF2 /W) =0.45%
P(TF3 /W) =0.06%
P(TF4 /W) = P(TF5 /W) = -
P(TF0 /W) =9.22%
P(TF1 /W) =22.22%
P(TF2 /W) = 0.80%
P(TF3 /W) = 0.30%
P(TF4 /W) = P(TF5 /W) = 0.10%
166
In this analysis, high probabilities of true warnings are associated with low intensity tornadoes.
This may be due to the fact that the majority of the tornadoes in the Prairies are F0 or F1.
7.8 Conclusion
In this chapter, we analyzed tornado warnings paying special attention on false tornado warnings
and missed events. Two decision trees were developed to represent the household response to a
warning and a forecaster‟s decision to issue a warning. The data analysis results show that the
Canadian Prairie tornado records have a higher probability of false warning and a lower
probability of detection. In this analysis, given a severe weather event, tornadoes reported by EC
were used as actual tornado occurrence records. This does not necessarily include all the
tornadoes that may have occurred over the Prairies during the considered period. In Canada,
tornado verification process is solely based on reliable reports from the ground and radar data is
not used for post-event verification purposes (PASPC, 2010). Due to these reasons, some
tornadoes can go undetected, unreported or unverified. Despite these shortcomings associated
with the database used for calculating these probabilities, this analysis provides an insight into
the key issues such as the probability of false warning, probability of detection and probability of
missed events which are the indicators of the warning performance.
167
Chapter Eight: Suggestions for an Improved Tornado Mitigation System
Based on the overall research work discussed in the previous chapters, this chapter makes
recommendations to assist stakeholders in enhancing the existing system to detect, warn and
communicate tornado warnings to the public and in improving evacuation responses. These
recommendations are offered as guidelines for consideration and possible adoption by
stakeholders who are involved at different stages of the tornado detection, warning,
communication and evacuation process.
8.1 Storm Prediction Centre (SPC)
8.1.1 Improved Technology
Improved technology can assist decision-makers in correctly recognizing the situations that
require them to issue a warning. For example, the improved coverage of Doppler radars enhances
the tornado warning performance, increasing the probability of detection and mean warning lead
time, while reducing the probability of false warnings. In Canada, the Doppler radar density is
much lower than that of the US. The sufficiency and efficiency of the technology to detect
tornadoes should be evaluated. The ways to identify isolated tornado events that can easily go
undetected from radar observations also need to be examined.
8.1.2 Sufficient Forecasters
It has been identified that the lack of a sufficient number of forecasters is a critical issue that
impedes the timely detection of tornadoes. Even if the automated prediction of tornadoes using
various models and algorithms is capable of the initial recognition of environmental conditions
168
supporting tornado development, the tornado prediction capability of automated processes is not
strong enough for warning decision-making. It is important to have a sufficient number of human
forecasters who analyze the conditions to recognize these localized and short-lived events in
advance.
8.1.3 Spotters
The sole reliance on radar-based detection is not always desirable, due to the chaotic nature of
supercell development. Spotters can provide strong evidence about what is happening on the
ground to supplement radar detection in warning decision-making. It was revealed through the
research that spotter groups are not adequate in numbers and organization, particularly in the
City of Calgary; therefore, insufficient information is provided to forecasters for making
decisions on tornado warnings. The implementation of an organized and effective group of
spotters is important to improve the reliability and efficiency of the warning decision-making
process.
8.1.4 Communication between the Storm Prediction Centre and Local Emergency
Management Agencies
The network analysis (Figure 4-6) revealed that there is no direct communication between the
SPC and local emergency management agencies in communicating a tornado warning to the
public. Direct communication with the Calgary Emergency Management Agency will ensure that
they are also aware of the warning and prepared to respond immediately. Close collaboration and
direct communication possibly through push notifications with active alerts are recommended.
169
8.1.5 Weatheradio Canada
Communicating tornado warnings to the public via the Weatheradio Canada network is currently
a low reliability link of the warning communication system. It was revealed that the Weatheradio
application had the lowest rating average (2.2 out of 5) for a source of tornado warning for
Calgary households. It is necessary to raise the awareness of Weatheradio Canada and its ability
to provide warning information not only on tornadoes, but also on other severe weather
conditions, such as thunderstorms, hail and winter storms.
8.1.6 Alternative Dissemination Methods
Official warnings should be disseminated through various media sources, and advantage should
be taken of recent communication trends, such as the use of Internet-enabled mobile phones and
social media, in issuing warnings. The stated preference survey revealed that dissemination
methods such as the Internet, text messages, smart phone apps and social media alerts have
become favourable options with a rating of around 3 out of 5.
8.1.7 Information Content of Warnings
The effectiveness of warnings can be improved by increasing the appropriateness of their
information content. For example, the comparison of Canadian and US systems showed that the
addition of a map showing the warning area provides more information to the public about the
possible touchdown area.
170
8.1.8 False Warning Reduction
Forecasters need to ensure that people trust tornado warnings, or urgent evacuation will not
occur. While paying attention to the warning issuance, it is also imperative that the warning
system finds ways to improve the public response. Repeated issuance of false warnings can
hinder perceptions of trustworthiness, and people may not heed the warnings. Issuing warnings
without a real need can also induce negative consequences. For example, US research has shown
that higher probabilities of false warnings significantly increase fatalities and injuries from future
tornadoes. Although it is unavoidable to have some false warnings, the overall effectiveness of a
tornado warning system can be enhanced by improving the mechanisms used to detect tornadoes
and enhancing the forecasters‟ decision-making process in issuing warnings.
According to the Bayes‟ theorem discussed in Section 7.7.2, forecasters who become successful
in warning issuance for most of the tornadoes that actually occur and not missing warnings for
any other tornadoes (i.e. missed events) have the chance of reducing the probability of false
warnings. The probability of false warnings for the ten-year period of 2003-2012 in the Canadian
Prairies was 87.4%. This figure shows that nearly nine out of ten tornado warnings were
recorded as false warnings. Furthermore, the Canadian Prairie region has a probability of
detection of around 40%. In other words, only four out of every ten tornado events are covered
by warnings. These probabilities do not indicate a good warning system. There is a definite need
to improve these warning performance indicators.
171
8.1.9 Missed Event Reduction
It was shown that when the missed event probability is higher than the true warning probability,
a warning has to be issued even if the public do not respond. However, if the true warning
probability is higher, forecasters seek at least some response probability (>0) to justify a
warning. This indicates the importance of minimizing the missed event probability.
8.1.10 Increased Lead Time
Comparison of the time from the warning issuance to evacuation completion with the present
warning lead time of around 10 minutes indicates why it is imperative to improve the warning
lead time. If the lead time is increased by 5 minutes, there is more than 75% percent confidence
that a household can complete evacuation prior to a tornado touchdown. It is recommended to
take actions to improve the warning lead time using early detection, warning and communication
methods.
8.2 Calgary Emergency Management Agency
8.2.1 Community Awareness and Preparedness
Even if emergency management is capable of responding immediately in the post-disaster stage,
the preparedness at the pre-disaster stage of a tornado is also critically important. According to
the survey results, the majority of respondents know how to take safe cover during a tornado
(61.8%). However, at the community level in the City of Calgary, the majority of the
respondents (61%) believe that their residential community‟s preparedness for a tornado disaster
is not adequate.
172
It is recommended that the Calgary Emergency Management Agency take actions to improve
awareness and preparedness at the community level through training and other applied learning
activities, possibly through community centres and community newsletters. Special educational
efforts are required to increase the awareness of tornado warnings and of safe and timely
evacuation actions among the population under 30 years of age, as discussed in the probit
analysis in Section 6.11.
The tornado awareness and preparedness of public places, such as schools and hospitals, also
need to be addressed with the support of the administration in each organization, so that there are
plans and procedures in place to evacuate these vulnerable groups.
8.2.2 Alberta Emergency Alert System
The network analysis showed that Calgary Emergency Management Agency needs to take
immediate actions to get information from local observers, verify the threat and activate the
Alberta Emergency Alert (AEA) system, if it is not already active. This can be done by
interacting with the weather interest groups or volunteer spotter groups within and around the
city to obtain real-time severe weather information via 911.
8.2.3 Communication Sources
It is important to make sure that people receive warning information from reliable sources at the
right time. As shown by the stated preference survey, some people may seek confirmation from
different sources to reinforce the information prior to making an evacuation decision. It is
recommended that the Calgary Emergency Management Agency utilize various communication
173
media, including traditional media (local radio and television stations) the Internet, social media
and smartphone applications that can reach a diverse population with different preferences, so
that the public receive warnings through at least one reliable source.
8.2.4 Community Evacuation Actions
At the community level, taking evacuation actions in a timely manner is a greater challenge. For
example, evacuation of people from large outdoor activity areas, such as the Calgary Stampede
Grounds and McMahon Stadium, within a very short time can create many problems; and, the
resulting panic may cause injuries with or without any tornado strike. It is important to have
well-designed severe weather plans, especially a well-publicized plan of shelter to reduce the
tornado- and panic-induced causalities.
8.2.5 Improved Communication between the Storm Prediction Centre and the Calgary
Emergency Management Agency
As highlighted in the network analysis in Chapter 4, it is recommended that the communication
between the SPC and the Calgary Emergency Management Agency be significantly improved, in
order to warn the public and launch preparedness activities immediately.
8.3 Alberta Emergency Management Agency
The AEA system is a very important tool in making the public aware of the potential for
tornadoes. As the early detection and warning capability of the SPC is not always desirable, this
alert system can be a big asset. However, the lack of sources of reliable information to activate
the AEA system at the local level is a critical issue. Information from the public generally comes
174
at the last moment; and, the emergency call (911) centres are generally overloaded with more
traditional emergency calls. These factors can delay the immediate activation of the AEA
system. It is, therefore, necessary to take actions to train and create a group of authorized users
who are specialized in the accurate detection of the potential for a tornado.
8.4 Spotter Network
Ground verification of severe weather and tornado potential, such as the appearance of funnel
clouds, improves forecaster confidence and adds more detail and credibility to warning
messages. The role of qualified and experienced spotters is particularly important during the
night at locations far from radar sites and on marginal tornado days. The spotter network
CANWARN (Canadian Weather Amateur Radio Network) is active in some areas, such as
Edmonton, Red Deer and Winnipeg; however, this network is not very active in the Calgary area.
Even if there are volunteers who provide reports of funnel clouds occasionally, their contribution
to the detection of tornadoes is minimal or not always possible. It is recommended to have an
organized group to act during a severe weather event.
8.5 Schools and Children’s Activity Centres
8.5.1 Centres’ Preparedness
Schools and children‟s activity centres are expected to have the fastest, most accurate and
reliable means of receiving critical weather information and have action plans to safeguard
children against tornadoes. Presently, Calgary schools do not practise tornado evacuation
procedures, although they have preparedness procedures for general emergencies. There should
be well-practised plans and procedures that direct all occupants of the upper floors into safe
175
areas, possibly to the basement level. Evacuation drills can be conducted in the spring, so that
school children know how to be safe from tornadoes that can occur in the summer season.
8.5.2 Parents’ Evacuation Actions
As indicated by rating averages of around 3.5 out of 5.0, a significantly high percentage of
respondents intend to drive to pick up their children from schools or other activities before taking
a safe evacuation action. It is, therefore, important that parents are clearly informed of the
schools and activity centres‟ preparedness plans, so they stay calm, seek safe shelter for
themselves, and avoid taking unsafe actions by driving to collect their children.
8.6 Calgary’s Road Operations Centre (ROC)
8.6.1 Variable Message Signs
The stated preference survey showed that variable message signs (VMS) provide a good
potential source of warning to drivers, as indicated by a rating average of around 3 out of 5. It is
necessary to have clear communication and coordination between the SPC or the Calgary
Emergency Management Agency and Calgary‟s Road Operations Centre (ROC) to present realtime warnings to motorists using VMS.
8.6.2 Peak Period Traffic Preparedness
The evacuation response to a tornado warning situation during peak period traffic needs a great
deal of cooperation, so that everyone can stop their vehicles and find the nearest shelter.
According to survey responses, the average time a driver would take to evacuate is around 3
minutes. However, during a peak traffic period, the individual responses and times are dependent
176
on the behaviour of other drivers. The drivers need to be educated to respond to such short-notice
evacuations, possibly through the Alberta driver licensing program. Furthermore, it is
recommended that traffic management technologies that can be applied in response to a tornado
warning be investigated.
8.7 Media
Even if households get warning information from a phone call from a trusted person, evacuation
is not immediate. The research findings show that the direct reception of formal information-rich
warnings to the public is more capable of triggering early evacuation actions than unofficial
warnings that come through third parties. This implies the importance of the maximum
utilization of official and traditional media (local television and radio) in convincing residents to
take timely evacuation actions. When there is a warning issued by Environment Canada or a
critical alert for a tornado issued through the Alberta Emergency Alert (AEA) system, it is
broadcasted through local radio and television networks by interrupting regular programming.
In addition to this warning support, local media needs to play a vital role in educating the public
by facilitating discussions about tornado preparedness and response in the spring and summer
seasons.
8.8 Emergency Services
Emergency Service officers (EMS, Fire and Police) need to be ready to respond during the postdisaster stage. As responsible authorities for public safety, they also can give sighting
177
information to the Calgary Emergency Management Agency when they observe a tornado or
tornado warning signs.
8.9 The Public
8.9.1 Watches and Warnings
The public should be familiar with the Storm Prediction Centre‟s (SPC) watches and warnings,
and the information and critical alerts of the AEA system. The stated preference survey revealed
that around one third of the population is not aware of the difference between a watch and a
warning. Knowing the difference between a tornado watch and a tornado warning is very
important for appropriate reactions. A tornado watch is issued when the conditions are
favourable for the development of a tornado in the coming hours; whereas a tornado warning is
issued when it is likely that a tornado may develop soon in the area, when a tornado is occurring
in a nearby area and may soon move into the area, or when a tornado is already occurring in the
area.
Critical alerts from the AEA system are similar to tornado warnings, which are issued when there
is an imminent life-threatening danger. AEA information alerts provide less critical information
to the public to help them to prepare for an emergency. The public needs to be aware under what
conditions each alert issued, as well as the possibility of a rapid change from one stage to the
other.
178
8.9.2 Environmental Cues
It is likely that the systematic methods of providing warnings to the public are not always
feasible. People can identify incoming tornadoes through environmental cues, such as a dark or
greenish sky, large hail, thunder and lightning, funnel clouds and rumbling sounds. As indicated
in the survey responses, appearance of visible environmental cues has a high rating average for
giving warnings to both the household and driving populations. Improving surveillance to
recognize these environmental cues through awareness and education programs that show visual
information about the tornado development can play a major role in recognizing the imminent
threat of a tornado.
According to drivers‟ responses in the survey, these cues can dramatically increase the perceived
risk level, causing them to take immediate action. It is important to inform the emergency call
(911)centre about what is happening on the ground, if the public can safely do it during a
threatening weather condition. This information helps in issuing an immediate official tornado
warning.
8.9.3 Heeding Warnings
Even when accurate warnings are issued and properly communicated to the public, there is a
portion of people who will choose not to heed the warnings. This may happen when people
frequently experience tornado warnings that do not result in tornadoes, when they are not aware
of the damage that can be caused by tornadoes, or due to the illogical belief that the tornado will
not hit their house. However, given that the timing, location and the intensity of tornadoes cannot
179
be precisely predicted, the public needs to understand the danger communicated through a
warning and take the correct actions immediately.
8.9.4 One Reliable Source
It is always important to have several sources to receive timely and reinforced warnings.
However, there can be situations when these sources may not be active or accessible by a person,
depending on the time of the day or night and the location (indoor or outdoor). The survey results
showed that more than 50% of the respondents make relatively unsafe decisions in waiting for
official warnings or seeking more information. People should not wait until they get all the
information about an approaching tornado. It is sufficient to have a warning from one reliable
source to initiate evacuation.
8.9.5 Personal and Family Preparedness
Personal preparedness can greatly assist people in minimizing the unnecessary wait time before
taking evacuation actions. It is always good for families to have a plan of action that has been
communicated among all family members, in order to be able to act quickly and efficiently
without waiting until it is too late.
8.9.6 Safest Actions
Under a tornado warning, going to the basement or lowest level of the shelter is usually the safest
action for households. In the absence of a basement, lying down in a bathtub or going to a safer
area in the near-neighbourhood can be relatively safer than driving away from the home to avoid
the threat.
180
8.9.7 Road Evacuation
Road evacuation can be more critical, and it requires safety planning procedures at the individual
level. In the absence of emergency officials to guide an urgent evacuation, panic may lead to
more time being consumed before making a safe evacuation action. People need to have
knowledge about possible safe evacuation places along their routes.
According to the survey, an alarming response was found in the driving scenario, which
indicated that driving away from the direction of the tornado would be the most likely evacuation
action, with 58% of respondents intending to take this action. However, the safety level of
driving away from a nearby tornado is quite uncertain, given that the tornado may not be visible
or the direction and speed may be unpredictable. The safest decision in a tornado emergency
would be to stop driving and seek shelter in the lowest level of the nearest building. However,
there are no hard and fast rules about what sources are more reliable and what evacuation actions
are safer. Taking the safest and earliest possible action, depending on the situation, is the wisest
decision.
8.9.8 False Warnings / Missed Events
The efficiency of the warning system should not be judged by only focusing on false warnings or
missed event records. A false warning situation does not always prove that people have been
warned unnecessarily, and a missed event does not show a total failure of the warning system.
With the uncertainties associated with the chaotic nature of tornado development and
complications faced by human forecasters, the public needs to have some patience regarding
false warnings and missed events.
181
8.10 Crucial Factors for each Stakeholder
The crucial factors that need to be considered by each stakeholder – the Storm Prediction Centre,
the Calgary Emergency Management Centre, the Alberta Emergency Management Centre,
Schools and Activity Centres, Calgary‟s Road Operations Centre, the media, the police and the
public –are summarized in Table 8-1. Implementation of the recommendations requires a strong
commitment and actions from the each stakeholder organization and the public. It is also
essential that organizations consider interactions with other stakeholders and the public in
translating these recommendations into actions.
It is worth mentioning that Alberta has experienced two major tornadoes in the recent past,
namely the 1987 Edmonton tornado (F4) that killed 27 people and the 2000 Pine Lake tornado
(F3) that killed 12 people. When considering these incidents and other summer severe storms in
the Prairie region, tornadoes are possible in Calgary; however, the risk should not be over
sensationalized. Understanding of the dangers, long-term planning and practice can improve the
chances of correct responses to a real tornado threat.
182
Table 8-1: Recommendations for Stakeholders
Stakeholder

Storm Prediction
Centre (SPC)




Calgary Emergency
Management
Agency (CEMA)



Alberta
Emergency Mgmt
Agency (AEMA)
Schools
Calgary‟s Road
Operations Centre
Media
Recommendations
Check the sufficiency and efficiency of the technological and human capacity to detect tornadoes and take remedial
measures (Subsections 8.1.1 and 8.1.2)
Implementation of an organized group of spotters to get true ground information (Subsection 8.1.3)
Develop interactions with the local emergency managers (Subsection 8.1.4)
Promote the use of Weatheradio as the primary warning source (Subsection 8.1.5)
Conduct annual information sessions and drills to improve the awareness and preparedness at the individual, institution
and community levels (Subsections 8.2.1and 8.2.4)
Develop interactions with the SPC to receive push notifications with active alerts about tornadoes
Encourage spotters and the public to give tornado information via 911 in order to activate the Alberta Emergency
Alert(AEA)system (Subsections 8.2.2 and 8.2.5)
Utilize various communication media, including traditional local media (radio and TV), the Internet, social media and
smartphone applications, in order to reach a diverse population with different preferences (Subsection 8.2.3)

Strengthen the AEA system to verify tornado information at the local level (Section 8.3)






Practice evacuation drills in the spring season (Subsection 8.5.1)
Improve the awareness of parents regarding school evacuation measures (Subsection 8.5.2)
Initiate the use of variable message signs for severe weather warnings including tornadoes (Subsection 8.6.1)
Educate drivers on how to respond to a tornado emergency (Subsection 8.6.2)
Study traffic management technologies for responses to a tornado warning (Subsection 8.6.2)
Educate the public by facilitating discussions about tornado preparedness and response in the spring and summer
seasons(Section 8.7)
Be ready to respond once a tornado touchdown is reported (Section 8.8)
Improve the awareness of weather alerts, watches, warnings and evacuation actions (Subsections 8.9.1, 8.9.3, 8.9.6 and
8.9.7)
Observe to recognize the environmental cues of tornadoes (Subsection 8.9.2)
Develop a family preparedness plan in responding to a tornado (Subsection 8.9.5)
Emergency Services


The Public


183
Chapter Nine: Conclusion
This chapter summarizes the thesis, highlighting the conclusions and synthesizing the
contributions and deliverables to the body of preexisting knowledge in tornado disaster
mitigation. The chapter further identifies limitations of the work performed and outlines possible
future research directions.
9.1 Research Summary and Conclusions
The research summary and conclusions are chapter specific and they can be summarized as
follows.
9.1.1 Tornado Trend in the Canadian Prairies
Assessing the climatology of tornado events is important in various ways to help plan impact
mitigation. The historical tornado database for the Canadian Prairies (1921-2011) was analyzed
using regression methods, and novel models were developed to represent the trend over time.
They were further refined to capture the fluctuating behaviour of the number of tornadoes
reported in each year.
There appears to be a wave form for the time trend with a period of around 65 years. This wave
pattern shows the presence of a decreasing trend of the number of tornadoes observed in last two
decades. This downward trend may precede an upward trend in the number of tornadoes in
coming years. However, there is no way of confirming this in the absence of data. The
184
conclusion is based on the trend pattern of the available data, and the analysis of the underlying
factors for the wavy form of the trend is beyond the scope of this thesis.
9.1.2 Application of Network Modelling for Tornado Detection, Warning and Communication
(TDWC) System
A detailed analysis of the process from the point of tornado detection to the point of receiving
the warning was conducted, thus evaluating the roles and responsibilities of different
collaborating stakeholders. An activity network was developed, synthesizing the connectivity of
these collaborators in communicating a tornado warning to the public. The network represents
what each stakeholder should do when there is a need to identify a tornado and communicate a
warning to residents.
The application of network modelling was a unique way of analyzing the overall system, taking
the time duration of each activity into account. The overall time consumption distribution
obtained through the Monte-Carlo simulation based network is an indicator of the viability of the
existing system to safeguard the public against tornadoes. According to the simulation results,
the overall time spans from 14 minutes to 28 minutes. On average, it takes 20.5 minutes to
detect, warn and communicate a tornado warning to the public. Based on this analysis,
suggestions to improve the quality and the timeliness of the tornado detection, warning and
communication (TDWC) system was provided.
The total time duration from warning issuance to evacuation completion was obtained by
combining the outputs of the network analysis and the evacuation survey. This time was
185
compared with the Storm Prediction Centre‟s warning lead time of around 10 minutes, and
inferences were made urging the necessity of improving the warning lead time.
It is important to note that the TDWC network has been developed linking the possible activity
sequence and their time durations obtained from various sources. This activity sequence and time
durations can have variations based on different tornado warning situations and the availability
of information.
9.1.3 Comparison of the US and Canadian TDWC Systems
The status of the present TDWC system in Canada can be better understood when compared with
the well-established US system. A qualitative study was conducted to review and compare these
two systems. It focussed on key issues, namely prediction/detection capabilities, warning
provision, emergency preparedness, institutional partnerships, warning area, warning
dissemination methods and the importance of spotter networks.
This comparison supports the recognition of the factors that need to be considered in improving
the present TDWC system in Canada. However, it does not indicate that simply imitating the US
system can solve the problem areas in the Canadian system. There are several factors, such as
tornado risk potential, disaster cost, financial and technical capacity limitations, to be considered
in making changes. Any suggestions to optimize the present system should be the result of a
proper cost-benefit analysis. A quantitative comparison using Monte-Carlo simulation was not
performed because collecting US data on each activity was beyond the scope of this study.
186
9.1.4 Evacuation Behaviour Analysis – Stated Preference Survey
A stated preference survey was conducted to analyze the evacuation behaviour of Calgary
households and drivers during a tornado emergency. This study developed probit models to
examine the factors influencing evacuation behaviour. The behavioural responses that emerged
from the survey provide important factors to be considered in mitigating the impact of tornadoes
at the individual level, as well as at the community level. The responses were analyzed in detail,
and recommendations were provided to improve the overall mitigation system.
9.1.5 Analysis of False Warnings and Missed Events
Warning decision-making is quite challenging for forecasters, particularly due to the negative
consequences associated with wrong decisions, such as false warnings and missed events. The
underlying factors for false warnings and missed events were analyzed in detail. False warning
and missed event probabilities in the Canadian Prairies were also estimated. A model of
household decision-making for evacuation was developed based on decision theory. A logical
basis for the warning decision-making process was also developed and analyzed in detail, and
suggestions were provided to improve the warning performance. A fundamental inequality of
decision-making was identified.
9.2 Research Contributions
Given that the tornado frequency of Canada is second only to the US, mitigation of the impacts
of tornadoes is becoming a major consideration for weather services, local emergency
management agencies and the public. This thesis is the first study centred on the mitigation of
the impact of tornadoes in the Canadian Prairies. The research contributes to a deeper
187
understanding of the TDWC system by providing an overall analysis that spans across different
areas under the general umbrella of tornado disaster mitigation. Furthermore, the research has
provided ideas to improve the preparedness at the individual, institutional level and community
levels. The key contributions can be summarized as follows:

Development of novel models to analyze the long-term tornado time trend in the
Canadian Prairies and the recognition of a wavy form for the tornado trend;

Development of a simulation-based activity network for a Calgary-household TDWC
system and analysis of the probability distribution of the overall time duration to
disseminate a warning;

Analysis of the total time duration from warning issuance to evacuation completion by
combining the outputs of the network analysis and the household and highway evacuation
survey;

Comparative analysis of the Canadian and US TDWC systems and suggestions for
improvements to the Canadian system;

A detailed analysis of the evacuation behaviour of Calgary households and drivers during
a tornado emergency, focussing on the factors influencing evacuation decisions;

Development of a fundamental inequality for the warning decision-making by
forecasters;

Development of a decision tree model for household evacuation; and,

Detailed analysis of the false warnings and missed events in the Canadian Prairies and
suggestions for improvement to the warning performance.
188
This study provides a number of important findings for the improvement of the tornado
detection, warning communication and evacuation process in the Canadian Prairies. Based on
these findings, a set of guidelines for the stakeholders involved in the TDWC process has been
developed. The research will also serve as a base for future studies on the mitigation of the
impacts of other disasters.
9.3 Research Deliverables
As deliverables, this research has culminated in the publication of two peer-reviewed journal
papers that contribute to the scientific body of knowledge. Furthermore, this research has been
presented at two international conferences on disaster management and in two panel sessions of
the Committee on Disaster Risk Management of the World Federation of Engineering
Organizations (WFEO), capturing a wide international audience of disaster professionals.
At the local level, a number of information sessions conducted at the Calgary Emergency
Management Agency and several other city-based presentations have contributed to improving
the awareness of local emergency managers and other interested parties.
The research provides guidelines for several stakeholders who contribute in different ways to the
detection of tornadoes and communication of warnings to the public. A set of recommendations
focussing on the public‟s role in mitigating the impact of tornadoes has also be developed.
189
The limitations of this study can be summarized as follows:

There may have been documentation uncertainties of the historical tornado database
obtained from the Storm Prediction Centre. Furthermore, in the absence of clear
verification processes, there is a possibility that non-tornadic events may have been
recorded as tornadoes. Due to these limitations, there may be minor variations in the
authentic tornado records. However, these variations likely do not have much influence
on the overall trend pattern.

There were some limitations in obtaining the sequence of activities and their time
durations for the development of the Calgary-based activity network. As the City of
Calgary has not experienced major tornado touchdowns in its recorded history, the
reliability of several links is subject to some level of uncertainty. However, they represent
what the collaborating partners should ideally do when there is a need to communicate a
warning to residents.

The stochastic nature of the time duration of each activity has been represented as a
triangular distribution, due to its relative simplicity in obtaining the minimum, most
likely and maximum time estimates. This is an approximate distribution that can be used
in the absence of detailed data sets. It was revealed through simulation that only a minor
variation in the total duration could be observed, even if a more general normal
distribution was used. This shows that the triangular distribution can be used to interpret
activity durations.

The results of the stated preference survey conducted in the City of Calgary represent the
intended behaviour of the public during a tornado emergency. In this absence of real data,
this survey method is an adequate way to obtain behavioural responses from the public.
190
Although these results may not precisely depict the actual behaviour, their inferences can
effectively be utilized in planning efforts to mitigate the impact of future tornadoes.

There have been several caveats identified in analyzing the warning records. A specific
criterion has been used in this research to identify individual tornado occurrences and
false warning records.
9.4 Summary and Recommendations for Future Research
This research is a good example of how a scientific analysis of one particular type of disaster can
identify weaknesses and suggest improvements to the overall mitigation system. These findings
are important for local emergency management agencies in their emergency planning efforts.
Canada‟s National Disaster Mitigation Strategy (PSC, 2010, p.3) also highlights the need to
“apply and promote scientific and engineering best practices in order to build a knowledge base
for sustainable, cost-effective mitigation decisions that contribute to community resiliency”. In
this regard, disaster research is highly influential in mitigating the impacts of future catastrophic
events. For example, given the issues associated with massive devastation of the 2013 Alberta
flood, which has become one of the costliest disasters in Canadian history, more attention should
be devoted to conducting research on long-term planning for flood disaster mitigation. At the
same time, while we make improvements based on recent major disasters, such as the 2013
Alberta flood, preparations for future events that are not repeats of the recent past should not be
neglected.
191
References
Abdelgawad, H., Abdulhai, B. (2009). Emergency evacuation planning as a network design
problem: a critical review, Transportation Letters: the International Journal of Transportation
Research, No.1 , pp. 41-58.
AEA
(2011).
Alberta
Emergency
Alert,
Government
of
Alberta,
Meteorological
Society,
http://www.emergencyalert.alberta.ca. Accessed 29 July 2011.
AMS
(2013).
Meteorology
Glossary,
American
http://glossary.ametsoc.org/wiki/Main_Page, Accessed 20 September 2013.
Barnes, L.R., Gruntfest, E.C., Hayden, M.H., Schultz, D.M., Benight, C. (2007). False
Alarms and Close Calls: A Conceptual Model of Warning Accuracy, Weather and
Forecasting, Vol. 22, pp.1140-1147.
Bartlett, J. E., Kotrlik, J. W., & Higgins, C. C. (2001). Organizational research: Determining
appropriate sample size in survey research. Information Technology, Learning, and
Performance Journal, Vol. 19, No.1, pp. 43-50.
Ben-Akiva ME, Lerman SR (1985). Discrete Choice Analysis: Theory and Application to
Travel Demand. The Massachusetts Institute of Technology, USA.
192
Brockwell, P.J., Davis, R.A. (2002). Introduction to Time Series and Forecasting, Second
Ed., Springer, New York, USA.
Brotzge, J. and Erickson, S. (2009). NWS Tornado Warnings with Zero or Negative Lead
Times, Weather and Forecasting, Vol. 24, pp. 140-154.
Brotzge, J.,Erickson, S.(2010). Tornadoes without NWS Warning. Weather and Forecasting,
Vol. 25, No 1, pp. 159–172.
Brotzge, J.,Erickson, S., Brooks, H. (2011). A 5-yr Climatology of Tornado False Alarms,
Weather and Forecasting, Vol. 26, Issue 4, pp.534-544.
CANWARN (2011). Central Alberta Amateur Radio Club. http://www.caarc.ca/canwarn,
Accessed 4 August 2011.
CANSIM
(2011).
Canadian
Socio-Economic
Information
Management
System,
http://www5.statcan.gc.ca/cansim/home-accueil?lang=eng, Accessed October 2011.
Cao, Z. and Cai, H. (2008). Tornado Frequency and its Large-Scale Environments Over
Ontario, Canada, The Open Atmospheric Science Journal, Vol. 2, pp. 256-260.
Cao, Z. and Cai, H. (2011). Detection of Tornado Frequency Trend Over Ontario, Canada,
The Open Atmospheric Science Journal, Vol. 5, pp. 27-31.
193
Chau, K.W. (1995). The Validity of the Triangular Distribution Assumption in Monte Carlo
Simulation of Construction Costs, Construction Management and Economics, Vol.13, No.1,
pp. 15-21.
Collins, M.L. and Kapucu, N. (2008). Early warning systems and disaster preparedness and
response in local government, Disaster Prevention and Management, Vol. 17, No. 5, pp. 587600.
Comstock, R.D. and Archer, P. (2004). Planning + Practise = Preparedness: A case study in
injury prevention, Work: A Journal of Prevention, Assessment and Rehabilitation, Vol. 23,
pp. 199-204.
Cropper, M.L., and Sahin, S. (2009). Valuing Mortality and Morbidity in the Context of
Disaster Risks, Policy Research Working Paper Series 4832, The World Bank.
Dash,N., and Gladwin, H.(2007). Evacuation Decision Making and Behavioural Responses:
Individual and Household, Natural Hazards Review, Vol. 8, pp. 69-77.
Donner, W.R., Rodriguez, H., Diaz, W. (2012). Tornado Warnings in Three Southern States:
A Qualitative Analysis of Public Response Patterns, Journal of Homeland Security and
Emergency Management, Vol. 9, No.2, pp. _, ISSN (Online) 1547-7355.
194
Dore, M.H.I. (2003). Forecasting the Conditional Probabilities of Natural Disasters in
Canada as a Guide for Disaster Preparedness, Natural Hazards, Vol. 28, pp. 249-269.
Dotto, L., Duchesne, L., Etkin, D. et al. (2010). Canadians at risk: Our exposure to natural
hazards, Canadian Assessment of Natural Hazards Project, paper series – number 48,
Institute for Catastrophic Loss Reduction, Ontario, Canada.
Drury, J., Cocking, C., Reicher, S., Burton, A., Schofield, D., Hardwick, A., Graham, D., and
Langston, P. (2009). Cooperation versus Competition in a Mass Emergency Evacuation: A
New Laboratory Simulation and a New Theoretical Model, Behaviour Research Methods
Vol. 41, No. 3, pp. 957-970.
Durage, S.W., Wirasinghe, S.C., and Ruwanpura, J.Y. (2013a). Comparison of the Canadian
and US Tornado Detection and Warning Systems, Natural Hazards, Vol. 66, Issue 1, pp. 117137.
Durage, S., Wirasinghe, S.C., Kattan, L. and Ruwanpura, J.Y. (2013b). Evacuation
Behaviour of Households and Drivers during a Tornado, Natural Hazards, Springer,
November 2013.
EC (2012). Public Alerting Criteria, Environment Canada, http://www.ec.gc.ca/meteoweather/default.asp?lang=En&n=D9553AB5-1#tornado, Accessed 13 June 2012.
195
EC
(2013a).
Spring
and
Summer
Weather
Hazards,
Environment
http://ec.gc.ca/meteo-weather/default.asp?lang=En&n=6C5D4990-1#tornadoes,
Canada,
Accessed
28 December2013.
EC (2013b). Enhanced Fujita Scale (EF-Scale), Environment Canada, http://ec.gc.ca/meteoweather/default.asp?lang=En&n=41E875DA-1, Accessed 19 August 2013.
EC (2013c). Media's Source for Weather Data and Information Canada, Environment
Canada, http://www.ec.gc.ca/meteo-weather/default.asp?lang=En&n=231F174E-1, Accessed
30 December 2013.
EC (2013d). What is Weatheradio?, http://www.ec.gc.ca/meteo-weather/, Environment
Canada, Accessed 19 June 2013.
EC (2013e). Tornado Warnings in the Canadian Prairies 2003-2012, Obtained from Ontario
Climate Centre, 13 March 2013.
EC (2013f). Prairie Tornado Chronology 2003-2012, Obtained from Prairie Severe Weather
event Climatology and Verification Reports (2003-2012).
Prediction Centre, Canada.
196
Prairie and Arctic Storm
Edwards, R., and Lemon, L.R. (2002). Proactive or reactive? The severe storm threat to
large event venues. Preprints, 21st Conf. Severe Local Storms, San Antonio, Amer. Meteor.
Soc., pp. 232-235.
Edwards, R., Ladue,J.G., Ferree,J.T. Scharfenberg, K., Maier, C. and Coulbourne, W.L.
(2013). Tornado Intensity Estimation - Past, Present, and Future, Bulletin of American
Meteorological Society, Vol. 94, pp. 641–653.
Emergency Management Act (2007). S.C.2007, c.15, Justice Laws Website, Government of
Canada, http://laws.justice.gc.ca Accessed 4 August 2011.
Etkin, D. (1995). Beyond the Year 2000, More Tornadoes in Western Canada? Implications
from the Historical Record, Natural Hazards, Vol. 12, pp. 19-27.
Etkin, D., Brun, S.E., Shabbar, A., Joe, P. (2001). Tornado Climatology Of Canada
Revisited: Tornado Activity during Different Phases of ENSO, International Journal of
Climatology, Vo.21, No. 8, pp. 915-938.
Etkin, D.A., Brun, S.E., Dogra,P., (2002). A Tornado Scenario for Barrie, Ontario, Paper
Series – No. 20, Institute for Catastrophic Loss Reduction, Ontario, Canada.
197
Farley, J.E. (2007).Call-to-Action Statements in Tornado Warnings: Do They Reflect Recent
Developments in Tornado-Safety Research? , International Journal of mass Emergencies and
Disasters, Vol. 25, No. 1, pp.1-36.
Fernando, H.J., Braun, A., Galappatti, R., Ruwanpura, J.Y., Wirasinghe, S.C. (2008).
Tsunamis: manifestation and aftermath. In: M. Gad-el-Hak, ed. Large-scale disasters:
prediction, control and mitigation. Cambridge University Press, pp. 258–292.
Fujita, T.T. (1971). Proposed Characterization of Tornadoes and Hurricanes by Area and
Intensity, SMRP Research Paper, No. 91, University of Chicago, USA.
Fujita, T.T. (1973). Tornadoes around the World, Weatherwise, Vol. 26, No.2, pp.56-83.
GA (2011). Traffic Safety Act 115(2)(i), Distracted Driving Legislation, Government of
Alberta,http://www.transportation.alberta.ca/Content/docType3679/Production/FactSheet.pdf
, Accessed 19 June 2013.
Golden, J.H. and Adams, C.R. (2000). The Tornado Problem: Forecast, Warning, and
Response, Natural Hazards Review, Vol. 1, No.2, pp. 107-118.
Goliger, A.M. and Milford, R.V. (1998). Reprinted from Journal of Wind Engineering and
Industrial Aerodynamics, Vol. 74-76, A Review of Worldwide Occurrence of Tornadoes,
pp.112, Copyright (2013), with permission from Elsevier.
198
GP(2013).
During
a
Tornado,
Get
Prepared,
Government
of
Canada
http://www.getprepared.gc.ca/cnt/hzd/trnds-drng-eng.aspx, Accessed 23 September 2013.
Grazulis, T.P. (2001). The Tornado: nature‟s ultimate windstorm, University of Oklahoma
Press, Norman, USA, 2001.
Grosvenor, G.M., Fahey J.M., Allen, W.L. and Carroll. A. (1998). Tornado Distribution in
North America, Natural Hazards of North America.
National Geographic Society,
http://www.utoronto.ca/imap/ Accessed 15 August 2011.
Hage, K. (2003). On Destructive Canadian Prairie Windstorms and Severe Winters, Natural
Hazards, Vol. 29, pp. 207-228.
Hammer, B.O., Schmidlin, T.W. (2001). Vehicle-occupant deaths caused by tornadoes in the
United States, 1900-1998. Environmental Hazards, Vol. 2, No. 3, pp. 105-118.
Henstra, D. and McBean, G.A. (2005). Canadian Disaster Management Policy: Moving
Toward a Paradigm Shift? Canadian Public Policy, Vol. 31, No. 3, pp. 303-318.
Holden, J., Wright, A. (2004). UK Tornado Climatology and the Development of Simple
Prediction Tools, Quarterly Journal of the Royal Meteorological Society, Vol. 130, pp. 10091021.
199
Hosmer, D.W. Jr., Lemeshow,S., Sturdivant, R.X. (2013). Applied Logistic Regression,
John Wiley & Sons, New Jersey, USA.
Hwacha, V. (2005). Canada‟s Experience in Developing a National Disaster Mitigation
Strategy: A Deliberative Dialogue Approach, Mitigation and Adaptation Strategies for
Global Change, Vol. 10, pp. 507-523.
IPCC (2007). Climate Change 2007: Synthesis Report, Contribution of Working Groups I, II
and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change
[Core Writing Team, Pachauri, R.K and Reisinger, A.(eds.)]. IPCC, Geneva, Switzerland,
104 pp.
ITS (2007). Evacuation Management Operations (EMO) Modelling Assessment:
Transportation Modelling Inventory, Intelligent Transportation Systems, US Department of
Transportation, USA.
Joe, P., Dance, S., Lakshmanan, V. et al. (2012). ,Automated Processing of Doppler Radar
Data for Severe Weather Warnings, http://www.intechopen.com/books/doppler-radarobservations-weather-radar-wind-profiler-ionospheric-radar-and-other-advancedapplications/automated-processing-of-doppler-radar-data-for-severe-weather-forecasting,
Accessed 18 June 2012.
200
Jordaan, I. (2005). Decisions Under Uncertainty: Probabilistic Analysis for Engineering
Decisions, Cambridge University Press, Cambridge, UK.
King, P., (1997). On the Absence of Population Bias in the Tornado Climatology of
Southwestern Ontario, Weather Forecasting, Vol. 12, pp. 939–946.
Kockelman K, Kweon Y. (2002). Driver injury severity: an application of ordered probit
models. Accid. Anal. Prev. Vol. 34, No.3, pp. 313–321.
Kwak, C., Matthews, A.C.(2002). Multinomial Logistic Regression, Nursing Research Vol.
51, No. 6, pp. 404-410.
League, C.E., Diaz, W., Philips, B., Bass, E.J. Kloesel, K., Gruntfest, E., Gessner, A. (2010).
Emergency manager decision-making in tornado warning communication, Meteorological
Applications, Vol. 17, No. 2, pp. 163-152.
Lee, D.E. & Shi, J.J. (2004).Statistical analyses for simulating schedule networks”,
Proceedings of the 2004 Winter Simulation Conference, WSC, pp.i-XI.
Lindley, D. V. (1965). Introduction to Probability and Statistics from a Bayesian Viewpoint.
Part 1: Probability. Part 2: Inference. Cambridge University Press. London.
201
Louviere, J.J., Hensher, D.A., Swait, J.D. (2000). Stated Choice Methods Analysis and
Applications, Cambridge University Press, U.K.
Lowe, A.B., McKay, G.A. (1962). The Tornadoes of Western Canada. Meteorological
Branch, Internal Report. Department of Transport, Queens Printer, Ottawa.
Mahapatra, K., Kant, S. (2005). Tropical deforestation: a multinomial logistic model and
some country-specific policy prescriptions, Forest Policy and Economics, Vol. 7, pp. 1– 24.
Marshall, T.P. (1993). Lessons Learned from Analyzing Tornado Damage, The Tornado: Its
Structure, Dynamics, Prediction, and Hazards, Geophysical Monograph 79, American
Geophysical Union, pp. 495 - 499.
Mathwave (2013). EasyFit : Distribution Fitting Made Easy, MathWave Technologies,
http://www.mathwave.com/products/easyfit.html, Accessed 23 October 2013.
Matyas, C., Srinivasan, S., Cahyanto, I., Thapa, B., Pennington-Gray, L., and Villegas, J.
( 2011). Risk perception and evacuation decisions of Florida tourists under hurricane threats:
a stated preference analysis. Natural Hazards, Vol. 59, pp. 871-890.
McBean, G.A. (2005). Risk Mitigation Strategies for Tornadoes in the Context of Climate
Change and Development, Mitigation and Adaptation Strategies for Global Change, Vol. 10,
pp.357-366.
202
McCarthy, D.H. (2001). The Role of Ground-Truth Reports in the Warning Decision-Making
Process during the 3 May 1999 Oklahoma Tornado Outbreak, weather and Forecasting, Vol.
17, pp. 647-649.
McCarthy, P. (Ed) (2011). Prairie and Northern Region Severe Weather Database (18262010). Prairie and Arctic Storm Prediction Centre, Environment Canada, CD-ROM.
McGovern,A., Rosendahl, D.H. and Brown, R.A. (2014). Toward Understanding Tornado
Formation through Spatiotemporal Data Mining (pp. 29-47). In Cervone, G., Lin, J.,Waters,
N. (Eds.), Data Mining for Geoinformatics- Methods and Applications, Springer New York,
USA.
Mileti, D., and O‟Brien, P. (1992). Warning during disaster: Normalizing communicated risk,
Social. Problems, Vo. 39, No. 1, pp. 40–55.
Mileti, D., and Sorensen, J. (1990). „„Communication of emergency public warnings.‟‟
ORNL-6609,
Oak
Ridge
National
Laboratory,
http://emc.ed.ornl.gov/publications/PDF/CommunicationFinal.pdf , Accessed 15 June 2012.
Monfredo, W. (2008). Blown away in Greensburg, USA: prediction and analysis of an EF-5
tornado, Weather, Vol. 63, No. 5, pp. 116-120.
203
Mooney, C.Z. (1997). Monte Carlo Simulation, Series: Quantitative Applications in the
Social Sciences, No, 116, SAGE Publications, Inc., USA.
Moussa, M. (2007). Development of Decision Support System (DSS) Simulation Tool for
Risk Analysis, MSc Thesis, Department of Civil Engineering, University of Calgary, Canada.
MSC (2003). Summer severe weather bulletins in Alberta - Watches and warnings,
Meteorological Service of Canada, Canada.
Murphy, B., Falkiner, L., McBean, G., Dolan, H., Kovacs, P. (2005).Enhancing Local
Level Emergency Management: The Influence of Disaster Experience and the Role of
Households and Neighbourhoods, Institute for Catastrophic loss Reduction, ICLR Research,
Paper Series – No. 43, Canada.
Murray, R.C. and McDonald, J.M. (1993). Design for Contaminants of Hazardous Materials,
The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophysical Monograph 79,
American Geophysical Union, pp. 379 - 387.
Newark, M.J. (1984). Canadian Tornadoes, 1950-1979. Atmosphere Ocean, Vol. 22, No. 3,
pp. 343-253.
Newark, M.J. and McCulloch, D. (1992). Using Tornado Climatology to Help Plan a
Doppler Radar Network, Natural Hazards, Vol. 5, pp. 211-219.
204
Nirupama, N. (2012). Risk and Vulnerability Assessment: A Comprehensive Approach,
International Journal of Disaster Resilience in the Built Environment, Vol. 3, No. 2, pp. 103114.
Nirupama, N. and Maula, A. (2013). Engaging Public for Building Resilient Communities to
Reduce Disaster Impact, Natural Hazards: Natural Hazards: Special Issue on Sociological
Aspects of Natural Disasters, Vol. 66, pp.51-59.
NOAA (1995). Tornadoes... Nature‟s Most Violent Storms, A Preparedness guide including
safety information for schools, National Oceanic and Atmospheric Administration,
http://www.nws.noaa.gov/om/severeweather/resources/ttl6-10.pdf,
Accessed
2 August
2013.
NOAA (2011a). Preliminary Assessment of Climate Factors Contributing to the Extreme
2011
Tornadoes,
National
Oceanic
and
Atmospheric
Administration,
http://www.esrl.noaa.gov/psd/csi/events/2011/tornadoes/climatechange.html, Accessed
2
August 2013.
NOAA (2011b). Tornado Forecasting and Warnings, National Oceanic and Atmospheric
Administration,http://www.noaa.gov/factsheets/new%20version/Tornadoes_web_version_fin
al.pdf , Accessed 19 November 2013.
205
NOAA(2012a). Tornado Detection and Warnings, National Oceanic and Atmospheric
Administration,http://celebrating200years.noaa.gov/breakthroughs/tornadowarnings/#doppler
,Accessed 23 September 2013.
NOAA(2012b). Mobile weather warnings on the way!, National Oceanic and Atmospheric
Administration,http://www.noaa.gov/features/03_protecting/wireless_emergency_alerts.html
, Accessed 6 June 2013.
NOAA (2013). National Weather Service Glossary, National Oceanic and Atmospheric
Administration, http://w1.weather.gov/glossary/, Accessed 20 September 2013.
NSSL
(2009).
Tornado
Basics,
National
Severe
Storms
laboratory,
http://www.nssl.noaa.gov/primer/tornado/tor_basics.html, Accessed 13 April 2011.
NSSL (2013a). Research Tools: Dual-Polarized Radar, National Severe Storms Laboratory,
http://www.nssl.noaa.gov/tools/radar/dualpol/, Accessed 23 September 2013.
NSSL
(2013b).
Tornado
Detection,
National
Severe
Storms
Laboratory
http://www.nssl.noaa.gov/education/svrwx101/tornadoes/detection/, Accessed 23 September
2013.
206
NSSL
(2013c).
Warn-on-Forecast,
National
Severe
Storms
Laboratory,
http://www.nssl.noaa.gov/news/factsheets/WoF_09jul2013.pdf , Accessed 23 September
2013.
NWS (2010a). National Weather Service Reference Guide, National Weather Service, USA,
http://www.nws.noaa.gov/om/guide/ Accessed 9 September 2011.
NWS
(2010b).
National
Warnings,
National
Weather
Service,
USA,
http://www.nws.noaa.gov/view/nationalwarnings.php, Accessed 14 July 2010.
Olson, R.H. (1965). On the use of Bayes‟ theorem in estimating false alarm rates, Monthly
Weather Review, vol. 93, no. 9, pp.557-558.
Oxendine, C., Sonwalkar, M., Waters, N. (2012). A Multi-Objective, Multi-Criteria
Approach to Improve Situational Awareness in Emergency Evacuation Routing Using
Mobile Phone Data, Transactions in GIS, Vol. 16, No. 3, pp. 375–396.
Oxford
(2013).
Definition
of
tornado
in
English,
Oxford
Dictionaries,
http://oxforddictionaries.com/definition/english/tornado, Accessed 16 August 2013.
Papastavrou, J.D. and Lehto, M.R. (1996). Improving the effectiveness of warnings by
increasing the appropriateness of their information content: some hypotheses about human
compliance, Safety Science, Vol. 21, pp. 175-189.
207
Paruk, B.J. and Blackwell, S.R. (1994). A Severe Thunderstorm Climatology for Alberta,
National Weather Digest, Vol. 119, No.1, pp. 27–33.
PASPC (2010). 2010 Prairie Summer Severe Weather Event Climatology and Verification
Results Report, Prairie and Arctic Storm Prediction Centre, Canada.
PASPC (2011). 2011 Prairie Summer Severe Weather Event Climatology and Verification
Results Report, Prairie and Arctic Storm Prediction Centre, Canada.
Paul, A.H. (1982). The Thunderstorm Hazard on the Canadian Prairies, Geoforum, Vol . 13,
No . 4, pp . 275-288.
Personal Communication (2011). Discussion with Greg Carbin, Warning Coordination
Meteorologist, National Storm Prediction Center, Oklahoma, USA, 23 August 2011.
Personal Communication (2012). E-mail Discussion with Shannon Bestland, Operational
Meteorologist (MT-03), Prairie and Arctic Storm Prediction Center , Winnipeg, Canada,
December 2012.
Pfarr, C., Schmid, A., Schneider, U. (2011). Estimating Ordered Categorical Variables Using
Panel Data: A Generalized Ordered Probit Model with an Autofit Procedure, Journal of
Economics and Econometrics Vol. 54, No.1, pp. 7-23.
208
PSC
(2010).
Canada's
National
Disaster
Mitigation
Strategy.
http://www.publicsafety.gc.ca/prg/em/ndms/index-eng.aspx Accessed 15 August 2011.
Raiffa, H. (1968). Decision Analysis - Introductory Lectures on Choices under Uncertainty.
Addison-Wesley, Reading, MA, USA.
Ravindra, M.K. (1993). State-of-the-Art and Current Research Activities in extreme Winds
Relating to Design and evaluation of Nuclear Power Plants, In: The Tornado: Its Structure,
Dynamics, Prediction, and Hazards, Geophysical Monograph 79, American Geophysical
Union, pp. 389 - 397.
Robinson M. and Khattak A. (2011). Route change decision-making by Hurricane Evacuees
Facing Congestion, Transportation Research Record: Journal of the Transportation Research
Board, Vol. 2196, pp. 168–175.
Ruwanpura, J., Wickramaratne, S., Braun, A., and Wirasinghe, S.C. (2009). Planning and
modelling for Mitigation of Tsunami Impacts, Civil Engineering and Environmental
Systems, Vol.26, No 2, pp. 195-209.
Schiermeier, Q. (2011). Climate and weather: Extreme measures, Nature, Vol.477, pp.148149.
209
Schiermeier, Q. (2011). Extreme Measures- Can violent hurricanes, floods and droughts be
pinned on climate change? Scientists are beginning to say yes, Nature, Vol. 477, pp.148-149.
Schiermeier, Q. (2012). Disaster toll tallied, Nature, Vol. 481, pp.124-125.
Schmidlin, T.W., Hammer, B.O., Ono, Y.,
King, P.S. (2009). Tornado shelter-seeking
behaviour and tornado shelter options among mobile home residents in the United States,
Natural Hazards, Vol. 48, pp. 191–201.
Schmidlin, T.W., King Vitae, P.S., Hammer, B.O., Ono, Y. (1998). Behaviour of Vehicles
during Tornado Winds, Journal of Safety Research, Vol.29, No.3, pp.181–186.
Schultz, D.M., Gruntfest,E.C., Hayden,M.H., Benight,C.C., Drobot,S., Barnes, L.R. (2010).
Decision Making by Austin, Texas, Residents in Hypothetical Tornado Scenarios, Weather,
Climate and Society, Vol.2, pp. 249-254.
Schumacher, R.S., Lindsey, D.T., Schumacher, A.B. et al (2010). Multidisciplinary Analysis
of an Unusual tornado: Meteorology, Climatology, and the Communication and
Interpretation of Warnings, Weather and Forecasting, Vol. 25, No, 5, October.
Shepherd,M.,
Niyogi, D.,
and Mote, T.L. (2009). A Seasonal-Scale Climatological
Analysis, Correlating Spring Tornadic Activity with Antecedent Fall–Winter Drought in the
Southeastern United States, Environmental Research Letters, Vol. 4, 7pp.
210
Sherman-Morris, K.(2010). Tornado warning dissemination and response at a university
campus, Natural Hazards, Vol. 52, pp. 623–638.
Sills, D. M. L. (2009). On the MSC Forecasters Forums and the Future Role of the Human
Forecaster. Bull. American Meteorological Society, Vol. 90, No. 5,pp. 619-627.
Simmons, K.M. and Sutter, D. (2006). Improvements in Tornado Warnings and Tornado
Causalities, International Journal of Mass Emergencies and Disasters, Vol. 24, No. 3, pp.
351-369.
Simmons, K.M. and Sutter, D. (2009). False Alarms, Tornado Warnings, and Tornado
Casualties, Weather, Climate, and Society, Vol.1, pp. 38-53.
SKYWARN (2011). National SKYWARN Homepage.
http://skywarn.org Accessed 15
September 2011.
Slovic, P. (2000). The Perception of Risk, Earthscan Publications LTD, USA.
Stensrud, D. J., Wicker, L. J. and Kelleher, K. E. et al. (2009). Convective Scale Warn-onForecast System, Bulletin of the American Meteorological Society, Vol.90, pp.1487-149.
Tay, R., Barua, U., Kattan, L. (2009). Factors contributing to hit-and-run in fatal crashes,
Accident Analysis and Prevention, Vol. 41, pp. 227-233.
211
Taylor,N.,
Sills,D.,
Hanesiak,J., Milbrandt,J., Smith,C.,McCarthy,P., Strong,G. (2012).
Understanding Severe Thunderstorms and Alberta Boundary Layers Experiment,
http://ebookbrowse.com/unstable-science-final-pdf-d314471272, Accessed 15 June 2012.
Timbal, B., Kounkou, R.,
Mills, G.A. (2010). Changes in the Risk Of Cool-Season
Tornadoes Over Southern Australia due to Model Projections of Anthropogenic Warming,
Journal of Climate, Notes and Correspondence. Vol. 23, pp.2440-2449.
TTU (2006). A Recommendation for an Enhanced Fujita Scale, Wind Science and
Engineering Center, Texas Tech University, USA.
Verbout, S. M., Brooks, H. E., Leslie, L. M. and Schultz, D. M. ( 2006). Evolution of the US
tornado database: 1954-2003. Weather Forecasting, Vol. 21, pp. 86-93.
Wickramaratne, S. (2010).
Design and Analysis of Tsunami Warning and Evacuation
Systems, Ph.D. Thesis, Department of Civil Engineering, University of Calgary, Calgary,
Canada.
Wickramaratne, S., Ruwanpura, J.Y., Wirasinghe, S.C. (2011). Decision analysis for a
tsunami
detection
system
–
case
study:
Sri
EnvironmentalSystems, Vol. 28, No. 4, pp. 353–373.
212
Lanka,
Civil
Engineering
and
Williams, R. (2006). Generalized ordered logit/partial proportional odds models for ordinal
dependent variables, The Stata Journal, Vol. 6, No.1, pp. 58–82.
Ye Z, Chaudhari J, Booth J, Posadas B. (2010). Evaluation of the Use of Rural
Transportation Infrastructure in Evacuation Operations. Journal of Transportation Safety and
Security. Vol. 2, pp. 88–101.
Yuan, M. , Dickens-Micozzi, M. and Magsig, M.A. (2002). Analysis of Tornado Damage
Tracks from the 3 May Tornado Outbreak Using Multispectral Satellite Imagery, Weather
and Forecasting, Vol. 17, pp. 382-398.
213
APPENDIX A: Dataset used for Regression Models
Year
Population (In
Millions)
P
Observed
number of
tornadoes
T0(t)
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1.9561
1.977
1.99
2.013
2.04
2.0674
2.125
2.184
2.244
2.3
2.3535
2.369
2.384
2.395
2.405
2.4155
2.413
2.415
2.418
2.418
2.4219
2.348
2.346
2.371
2.368
2.363
2.4
2.438
2.474
2.514
25
16
35
13
12
11
32
23
13
11
24
32
18
27
43
27
27
9
2
11
12
12
12
22
11
16
27
15
20
13
Year
Population
(In Millions)
P
Observed
number of
tornadoes
T0(t)
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
2.5477
2.614
2.682
2.753
2.808
2.8538
2.906
2.972
3.046
3.112
3.1789
3.235
3.285
3.331
3.365
3.3817
3.41
3.455
3.496
3.519
3.5423
3.6165
3.6446
3.6813
3.7511
3.8327
3.9303
4.0156
4.094
4.193
7
13
23
12
22
24
18
16
14
23
11
25
23
41
22
19
10
22
35
23
26
14
35
20
69
34
36
56
48
28
214
Year
Population (In
Millions)
P
Observed
number of
tornadoes T0(t)
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
4.3024
4.4016
4.4546
4.4803
4.5119
4.5532
4.572
4.587
4.6216
4.6609
4.7046
4.7494
4.7918
4.8334
4.8779
4.9283
4.9839
5.0539
5.1097
5.1591
5.2097
5.2818
5.3437
5.4105
5.4941
5.5974
5.7072
5.8152
5.922
6.003
6.087
27
62
60
60
26
69
63
73
69
67
79
37
43
48
28
51
37
36
68
51
26
28
33
35
41
26
31
39
16
19
18
APPENDIX B: Simulation of Regression Models
%Load tornado data and define variables
load('Tornado Data.mat');
Y = (log(untitled(:,2)));
d=-20:0.1:20;
k=-20:0.1:20;
c=0.03;
r_squared_values=zeros(length(k),length(d));
%Find R square values for different d,k combinations
for i=1:1:length(k)
for j=1:1:length(d)
X= log(untitled(:,1))+(d(1,j).*((1+c.*untitled(:,1)).^k(1,i)));
[b,bint,r,rint,stats] = regress(Y,[ones(91,1),X]);
r_squared_values(i,j)=stats(1,1);
end
end
%Find the maximum R squared and the corresponding d,k values
max_r_squared= max(max(r_squared_values));
[index_k_at_r_max index_d_at_r_max]=(find(r_squared_values==max_r_squared));
d_at_r_max= -20+(0.1)*(index_d_at_r_max-1);
k_at_r_max= -20+(0.1)*(index_k_at_r_max-1);
%Find the K1(), K2() values
X1= log(untitled(:,1))+(d_at_r_max.*((1+c.*untitled(:,1)).^k_at_r_max));
[final_b,final_bint,final_r,final_rint,final_stats] =
regress(Y,[ones(91,1),X1]);
K1= final_b(1,1);
K2= final_b(2,1);
%Plot the countour
contour(d,k,r_squared_values)
xlabel('d Value');
ylabel('k Value');
title('Contour plot');
215
APPENDIX C: Stated Preference Survey
Tornado Evacuation Behaviour Analysis
Dear Respondent,
I am a PhD student in the Department of Civil Engineering at the University of Calgary,
conducting research on tornado mitigation. My research is focussed on the development of a
more efficient tornado warning and communication system for Alberta to mitigate the impact on
the public and property.
Tornadoes are violent storms with rotating columns of high velocity wind capable of creating
significant damage, injuries and even fatalities. Alberta is located on the fringe of tornado alley
in North America. Although the frequency and the impact of tornadoes have not been high in
Alberta compared to the rest of tornado alley, Alberta has experienced two major events in the
recent past in 1987 and 2000.
I am undertaking a survey about the evacuation behaviour of the public during a tornado event. I
would like to collect information on how Calgarians get tornado warnings and respond to
them.This survey will only take around 10 minutes to complete. The survey has been approved
by the Conjoint Faculties Research Ethics Board (CFREB) at the University of Calgary.
Thank you in advance for your co-operation.
Yours sincerely,
Samanthi Durage
Before filling out the survey questionnaire, please read the consent form carefully.
216
Name of Researcher, Faculty, Department
Samanthi Durage , Schulich School of Engineering, Department of Civil Engineering, University
of Calgary
Supervisors
Prof. S.C. Wirasinghe, Prof. Janaka Ruwanpura and Dr. Lina Kattan, Department of Civil
Engineering, University of Calgary
Title of Project
Tornado Warning and Evacuation Behaviour Analysis - Stated Preference Survey
Purpose of the Study
The main purpose of this study is to investigate the tornado warning and communication system
in Alberta, Canada, and propose improvements to plans and systems to further mitigate the
impacts of tornadoes.
What will I be asked to do?
The questionnaire seeks to gather information on how Calgary residents will receive tornado
warnings and what evacuation actions they will take.
What type of personal information will be collected?
No personal identifying information will be collected, and all participants shall remain
anonymous. Should you agree to participate, you will be asked to provide generalized
information such as your gender, age group, dwelling type, household size, presence of school
age children, household income range(optional), and level of education (optional). All
information you give us will be treated in a confidential manner.
Are there risks or benefits if I participate?
217
Your support will be helpful in the effort to develop a better tornado warning and evacuation
process in Calgary. In case you feel distressed when imagining a tornado during the filling of the
questionnaire, please end the activity immediately. In the event that you do feel distressed during
these interactions or later, we suggest that you consider contacting your family physician or, in
the absence of one, visiting a drop in clinic. The online version of this survey is being
administered by Surveymonkey(c), an American software company. As such, your responses are
subject to U.S. laws, including the USA Patriot Act. The risks associated with participation are
minimal, however, and similar to those associated with many e-mail programs, such as
Hotmail(c) and social utility spaces, such as Facebook(c) and MySpace(c).
What happens to the information I provide?
Participation is completely voluntary, anonymous and confidential. No one except the researcher
and her academic supervisors will be allowed to see or hear any of the answers to the
questionnaire. There are no names on the questionnaire. Only group information will be
summarized for any presentation or publication of results. Potential venues for the dissemination
of results will be journal articles, PhD thesis, conference presentations, seminars and reports to
the Calgary Emergency Management Agency. The online survey results are kept in a passwordprotected computer only accessible by the researcher and her supervisors. The anonymous data
will be stored for two years, at which time, it will be permanently destroyed.
Written Consent
By submitting the completed or partially-completed survey, you are indicating your consent as a
participant in this research study. You are free to withdraw from this survey at any time by
simply not submitting the completed survey.
218
Questions/Concerns
This consent form is only part of the process of informed consent. If you have any further
questions or want clarification regarding this research and/or your participation, please contact
Ms.Samanthi Durage at (403) 398-6497; email [email protected] or Prof. S.C. Wirasinghe
at (403) 220-7180; email [email protected].
If you have any concerns about the way you've been treated as a participant, please contact the
Senior Ethics Resource Officer, Research Services Office, University of Calgary at (403) 2203782; email [email protected].
Do you agree to the consent information listed on this form?
 Yes, I agree to the above consent form.
 No, I don't agree to the above consent form.
Tornado Knowledge
1. Most tornadoes are born in the turbulence of a thunderstorm.
Yes
Don‟t Know
No
2. Summer is the peak tornado season.
Yes
Don‟t Know
No
3. A tornado can occur at any time of the day or night.
Yes
Don‟t Know
No
4. The lowest level of a shelter is generally the safest location during a tornado.
Yes
Don‟t Know
No
219
5. An intense tornado can uplift vehicles away from roads.
Yes
Don‟t Know
No
6. Environment Canada is the official source of tornado warnings in Canada.
Yes
Don‟t Know
No
Please select your level of agreement or disagreement with each statement.
7. Precise tornado forecasting is difficult due to uncertainties in estimating weather conditions
and technology limitations.
Strongly Disagree
Disagree
Neutral
Agree
Strongly Agree
8. Tornadoes always give warnings through various environmental cues when they are about to
occur.
Strongly Disagree
Disagree
Neutral
Agree
Strongly Agree
9. I know the difference between a „Tornado Watch‟ and „Tornado Warning‟.
Strongly Disagree
Disagree
Neutral
Agree
Strongly Agree
10. It is important to wait until an official warning is issued to take evacuation actions.
Strongly Disagree
Disagree
Neutral
Agree
Strongly Agree
11. I will seek more information about an official warning before evacuating.
Strongly Disagree
Disagree
Neutral
Agree
Strongly Agree
Agree
Strongly Agree
12. I know how to take cover during a tornado.
Strongly Disagree
Disagree
Neutral
13. Under a tornado warning, I am not able to concentrate on the instructions given.
Strongly Disagree
Disagree
Neutral
Agree
Strongly Agree
14. Overall, my residential community‟s preparedness for a tornado disaster is adequate.
Strongly Disagree
Disagree
Neutral
220
Agree
Strongly Agree
Disaster Experience
15. Do you have previous experience of any emergency or disaster of any kind which required
immediate reaction to evacuate?
Yes
No
16. If yes, what were they? Select all applicable disasters or emergencies.
Fire
Imminent Flooding
Tornado
Wind Storm
Avalanche
Other (Please Specify)......................................................................................
Please read this information before answering the following questions.
Tornadoes
Tornadoes are nature‟s most violent atmospheric storms that can have devastating impacts on
lives and property.

Tornado warnings have very low lead times.

Forecasters cannot precisely predict a tornado touchdown time, location, path and size.

Tornado warnings have to be given for a large area although the impacts are localized.

Tornado warnings are given by Environment Canada only when a tornado has been detected
by a reliable source.
Watch a 30 second tornado video.
https://www.youtube.com/watch?v=b1WWXpvfi3Q
221
Household Warning and Evacuation
17. Please rate each likely source of warning to your household in case of a tornado, on a scale
from 1 to 5.
Least Likely-1
Most Likely -5
Local radio
1
2
3
4
5
Emergency weather radio
1
2
3
4
5
Social media
1
2
3
4
5
Police public warnings
1
2
3
4
5
From neighbours
1
2
3
4
5
Television
1
2
3
4
5
Smart phones
1
2
3
4
5
Weather websites
1
2
3
4
5
Visible Environmental cues
1
2
3
4
5
1
2
3
4
5
(e.g. funnel clouds)
Call from a trusted person
Other (please specify) - if applicable.........................................................................................
18. Imagine that you just received a tornado warning. Before you evacuate, how much
importance will you place on the following actions? Rate on a scale from 1 to 5. (Do not fill
rows if you will definitely not take that action.)
222
Action
1- Very Low
5- Very High
Verify information with a different source
1
2
3
4
5
Go outside to confirm
1
2
3
4
5
Inform others in the house
1
2
3
4
5
Telephone family members/ close friends
1
2
3
4
5
Inform neighbours
1
2
3
4
5
Plan for taking photos or videos
1
2
3
4
5
Tweet, or post on Facebook to share the 1
2
3
4
5
2
3
4
5
2
3
4
5
message with others
If the children are out of the house, leave 1
to pick them from the local school or other
place
Other............................................
1
19. What are the possible evacuation actions that you will take at home under tornado warning?
Rank the following actions from 1 to 6. (Rank 1- Most Likely & Rank 6-Least Likely)
 Go to a safer area in the basement
 Lie down in a bath tub
 Stay in the house but take no evacuation action
 Go to a safer building within the neighbourhood
 Drive away to avoid the threat
 Other (please specify in question 20 - if applicable)
223
20. Please specify the other action indicated above (if applicable).
.............................................................................................................................................
21. How long will you wait before completing the most likely evacuation action?
0-1min
1-5min
5-10min
10-15min
More than 15min
Warning and Evacuation while Driving
22. Now assume that while you are driving, a tornado warning is to be announced. During your
trip, from what sources are you likely to receive the warning? Rate each source of warning
on a scale of 1 to 5.
Least Likely-1
Most Likely -5
Local radio
1
2
3
4
5
Social media
1
2
3
4
5
other road users
1
2
3
4
5
a call from home or friends
1
2
3
4
5
Mobile text alerts
1
2
3
4
5
Variable Highway Message Signs
1
2
3
4
5
Visible environmental cues
1
2
3
4
5
1
2
3
4
5
(Funnel cloud, flying debris etc. ...)
Other..........................................
23. Imagine that while you are driving you received a tornado warning through one source or
another. Before you evacuate to a safe place, how much importance will you place on the
224
following actions? Rate on a scale from 1 to 5. (Do not fill rows if you definitely not take that
action.)
Action
1- Very Low
5- Very High
Verify information from different sources
1
2
3
4
5
Inform others on the road
1
2
3
4
5
Call family members
1
2
3
4
5
Plan for photos or videos
1
2
3
4
5
Tweet or Facebook to share the message 1
2
3
4
5
2
3
4
5
2
3
4
5
with others
Drive to pick children from school or 1
other activity
Other……………………………….
1
24. What are the possible actions to seek shelter under tornado threats? Rank the following
actions from 1 to 7. (Rank 1- Most Likely & Rank 7-Least Likely)
 Drive away from the direction of the tornado
 Drive to the nearest building (house or commercial facility)
 Do nothing and continue driving to my destination
 Stop on the roadside and stay in the vehicle
 Stop on the roadside and hide under the car
 Seek road-side shelters such as bridges, highway overpasses and stop underneath
 Other (please specify in question 25 - if applicable)
225
25. Please specify the other action indicated above (if applicable).
......................................................................................................................................
26. How long will you wait before completing the most likely evacuation action?
0-1min
1-5min
5-10min
10-15min
More than 15min
40-49
50-59
Above 60
4
5
Above 5
Socio-Economic Factors
27. Gender:
Male
Female
28. Age (years):
Less than 20
20-29
30-39
29. Dwelling type:
Detached single family
Multifamily one or two storey
Multifamily three or more storeys
Mobile home
Basement suite
Other ..........................................
30. Household size:
1
2
3
31. How many bathtubs, with no adjacent glass enclosures or large mirrors, are in the house?
0
1
2
3
226
4
More than 4
32. Select the floors with bathtubs with no adjacent glass enclosures or large mirrors. Check (√)
all applicable answers.
Basement
Ground
Upper
33. Presence of school age children:
Yes
No
34. Presence of members with reduced mobility (optional):
Yes
No
35. Household income per annum (optional):
Less than 30,000
30,000-59,999
60,000-89,999
90,000-119,999
More than 120,000
36. What is the highest level of education that you have completed (optional):
High school or less
Some college
Post-secondary
Other........
College graduate
227
APPENDIX D: Copyright Permissions
Copyright Permission - Figure 2-2
228
Copyright Permission - Figure 2-3
229
Copyright Permission - Figure 7-1
230