University of Southern California Dissertations and Theses

Defining the geospatial needs of a ubiquitous disaster relief organization

(USC Thesis Other)

Defining the geospatial needs of a ubiquitous disaster relief organization

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

DEFINING THE GEOSPATIAL NEEDS OF A UBIQUITOUS DISASTER RELIEF
ORGANIZATION

by

Evan Lue

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
GEOGRAPHY

December 2014

Copyright 2014 Evan Lue

ii
DEDICATION
I dedicate this document to my grandfather Eugene Yeh, who passed away during its
development. He is remembered as an advocate for education. I am proud to carry on his
legacy.
iii
ACKNOWLEDGEMENTS
Thanks to the staff at the University of Southern California for their support: Robert
Alvarez, Leilani Banks, Kate Kelsey, Melissa Salido, and Sandra Vasquez. Also to those
at the American Red Cross: Corey Eide, Alyssa Hilario, Rick Hinrichs, Cliff Hu, Michael
Kleiner, Tomoyo Kuriyama, Kendra Pospychalla, Nancy Rodriguez, Alex Rose, Greg
Tune, Christopher Underwood, and Scott Underwood.

Thanks also to Dr. Travis Longcore and my advisor, Dr. John Wilson, who gave
me a chance to work on this degree as USC’s last geographer, and particularly to John
who demonstrated great patience and magnanimity through a long process. Thanks to Dr.
Andrew Curtis to whom I’m deeply indebted, because without him I would not have had
the opportunity to let the Red Cross play a formative role in both my personal and
professional developments. And thanks to my qualifying exam and dissertation
committee members: Rod McKenzie, Meredith Franklin, Manual Pastor, Lisa
Schweitzer, and Bob Vos.

Most importantly, I am grateful for the unwavering support of my family: Judy
Lue, Dale Lue, Alan Parker Lue, Ryan Lue, Irene Lue, Hsiang-Yun Yeh, Chin Su Lue,
Shu Chih Lue, Tai Lue, Anchi Lue, John Elson, Laura Lu, Vivian Chou, and Jessica Lee.
I could not have completed this body of work without their support.

iv
TABLE OF CONTENTS
Dedication ii
Acknowledgements iii
List of Tables vii
List of Figures x
List of Abbreviations xii
Abstract xiv
Chapter One: GIS and the American Red Cross 1
Chapter Two: Geospatial Technology Implementation at Local Red Cross Chapters 6
2.1 Introduction 6
2.1.1 Disaster GIS 7
2.1.2 GIS Diffusion 11
2.2 Methods 15
2.3 Results 18
2.3.1 Type 1: Building a Common Operating Picture (COP) 19
2.3.2 Type 2: Technology for Disaster Assessment (DA) 21
2.3.3 Type 3: Unplanned Exploration in Mapping 23
2.3.4 Type 4: The Traditional GIS Department 23
2.3.5 Trends in GIS Adoption 24
2.4 Conclusions 30
Chapter Three: Mapping Fires and Red Cross Aid with Demographic Indicators of
Vulnerability 33
3.1 Introduction 33
3.1.1 Disaster Risk and Vulnerability 35
v
3.2 Methods 40
3.2.1 National Fire Information and Reporting System (NFIRS) 41
3.2.2 Client Assistance System (CAS) 44
3.2.3 Demographic Variables 46
3.2.4 Spatial Joins and Expected Distributions 47
3.3 Results 48
3.3.1 High and Low Counts 51
3.3.2 The Intersection of Variables 53
3.4 Discussion and Conclusions 56
Chapter Four: Conducting Disaster Damage Assessments with Spatial Video, Experts,
and Citizens 59
4.1 Introduction 59
4.2 Damage Assessment, Crowdsourcing, and the GeoWeb 63
4.3 Methods 67
4.3.1 Photo Collection 68
4.3.2 Online Survey 69
4.3.3 Photo Selection 71
4.3.4 Comparing Respondent Groups and Inter-Rater Reliability (IRR) 74
4.4 Results 74
4.4.1 Respondent Summary 74
4.4.2 Damage Scores 76
4.4.3 Image Quality 86
4.4.4 Repeated Images 90
vi
4.4.5 Confidence 91
4.4.6 Respondents’ Comments 93
4.5 Discussion and Conclusions 94
Chapter Five: Synthesis and Conclusions 98
References 105
Appendix A: Chapter Two Interview Guiding Document 120
Appendix B: Chapter Four Survey (Abridged) 131

vii
LIST OF TABLES

Table 1: Eleven factors for GIS diffusion with 34 sub-factors
between them, taken from Onsrud and Pinto (1993).
Sub-factors in bold represent variables considered
highly important to the decision to adopt GIS. Factors
with one or more sub-factors considered highly
important also appear in bold. 14
Table 2: Eleven processes for GIS diffusion in order, taken from
Onsrud and Pinto (1993). 15
Table 3: Interest in map layers by Red Cross GIS users,
averaged and then rank ordered from most to least
interest (average score in parentheses; 5=most and
1=least). 25
Table 4: Important factors for GIS success by Red Cross GIS
users, averaged and rank ordered from most to least
important (average score in parentheses; 5=most and
1=least). 26
Table 5: Eleven factors of social vulnerability resulting from
Cutter et al. (2003). The dominant variable in each
factor is also indicated. 39
Table 6: Demographic variables used for analysis with fire and
aid events. Every census block group was assigned
either a Higher (H) or Lower (L) vulnerability
designation for each variable with the cutoff between
the two classes being the median variable value. 47
Table 7: A sample of representative variables with sums of
population, households, response events, and aid for
each H or L class. 52

viii

Table 8: Example of comparing actual fire response events to
what may be expected when block groups are classed
on the intersection of percentage minority and financial
vulnerability (i.e. low median household income). 55
Table 9: Example of comparing actual Red Cross aid events to
what may be expected when block groups are classed
on the intersection of percentage minority and financial
vulnerability (i.e. low median household income). 56
Table 10: Damage assessment classes as defined on American
Red Cross street sheets. These class definitions were
reproduced in the damage survey. 70
Table 11: Number of pictures used in the online survey belonging
to different damage scores as determined by the author. 72
Table 12: Number of respondents organized into groups and
subgroups. 76
Table 13: Percentages of scores for each picture as assigned by
Group A (n = 108). Cases of “no response” are ignored
for these calculations. Bold and shaded values designate
the mode while italicized scores designate the median.
An asterisk (*) indicates a picture where the median and
mode were not the same. 78
Table 14: Percentages of scores for each picture as assigned by
Group E-All (left; n = 83) and Group I (right; n = 23).
Cases of “no response” are ignored for these
calculations. Bold and shaded values designate the
mode while italics designate the median. An asterisk (*)
indicates a picture where the median and mode were not
the same. 81
ix
Table 15: The p-values from MWW tests between Group E-All
and Group I’s damage scores. Only Picture 19, in bold,
had a p-value within a 95% confidence limit. 82
Table 16: Krippendorff's α values for IRR and the damage scores
assigned by different respondent groups to multiple
subject sets. Green values are reliable, black values are
acceptable for tentative conclusions, and red values
should not be accepted as indicators of agreement. 85
Table 17: The p-values from MWW tests between Group E-All
and Group I’s image quality scores resulted in no cases
where the null hypothesis could be rejected. 87
Table 18: Krippendorff's α values for IRR and the image quality
scores assigned by different respondent groups to
multiple subject sets. All values are red, indicating that
they should not be accepted as indicators of agreement. 88
Table 19: Percentages of pictures with different combinations of
damage scores (rows) and image quality scores
(columns) out of 3,456 total picture responses. One
rater’s response to one picture accounts for 0.03% in
the table. Cells are colored on a gradient from red to
green corresponding with smallest to largest values
(0.09% to 10.73%). 89

x
LIST OF FIGURES

Figure 1: The American Red Cross divides its service areas into
regions and chapters. The Red Cross GIS Project is
primarily concerned with the Los Angeles Region,
encompassing Los Angeles County (larger map), Inyo
and Mono Counties, and part of Kern County (smaller
map). 2
Figure 2: A simplified disaster cycle depicting four stages to
disaster where an application for GIS is described in
each stage (from Cova 1999 and Godschalk 1991). 8
Figure 3: Distribution of focus (between planning and response)
for GIS activities at surveyed Red Cross chapters. 29
Figure 4: The PAR model describes disasters as a function of
hazards and vulnerability (Blaikie et al. 1994). 36
Figure 5: The model for hazards of place shows the factors that
contribute to place vulnerability (Cutter 1996). 37
Figure 6: Fire response locations mapped on top of median
household income by census tract. The map is zoomed
in to focus on the more populous areas of the county,
with 93% of fire responses visible in the view extent. 50
Figure 7: Red Cross aid locations mapped on top of median
household income by census tract. The map is zoomed
in to focus on the more populous areas of the county,
with 92% of aid events visible in the view extent. 51
Figure 8: A map depicting block groups by the combination of
classes (high and low) for percentage minority and
median household income. The orange HL code
represents block groups with a high percentage of
minorities and a low percentage of households with
low-income (i.e. households with higher income). 54
xi
Figure 9: Map of the neighborhood used for the ordered picture
set. 73
Figure 10: Mode of damage scores for the ordered (a) and random
picture sets (b) as scored by Groups A, E-All, and I. 79
Figure 11: Picture 11 (a) and Group E-All's assessment of it (b),
split between the “Affected” and “Minor” damage
scores. 83
Figure 12: Picture 21 (a) and Group E-O's assessment of it (b),
evenly split between the “Affected”, “Minor”, and
“Major” damage scores. 84
Figure 13: Contingency tables for each pair of repeated pictures as
rated by Group A. The rows in the tables represent
scores the first time the picture appeared while the
columns represents scores the second time. 91
Figure 14: Confidence related responses for Group I and Group E-
All, with general confidence on the left (a) and
assessments of image quality and likelihood of on-the-
ground reproducibility on the right (b). 93
Figure 15: Screenshot of the Red Cross GIS Project website
showing preparedness education events layered over a
choropleth map of disaster events (e.g. house fires). 100
xii
LIST OF ABBREVIATIONS
AGI Ambient Geographic Information
CAS Client Assistance System
COP Common Operating Picture
DA Disaster Assessment (Red Cross activity)
DFW Dallas Forth-Worth
DST Disaster Services Technology
EDR Emergency Disaster Response
EMDC Emergency Mapping and Data Center
GIS Geographic Information Systems
GPS Global Positioning System
IP Information & Planning (Red Cross group)
IRR Inter-rater Reliability
JSON JavaScript Object Notation (file format)
KML Keyhole Markup Language (file format)
LACoFD Los Angeles County Fire Department
LAFD Los Angeles Fire Department (City of Los Angeles)
MN Manager (Red Cross position)
MWW Mann-Whitney-Wilcoxon test
NFIRS National Fire Incident Reporting System
NHQ National Headquarters
PAR Pressure and Release
PCA Principal Components Analysis
PPGIS Public Participation GIS
xiii
RFP Request For Proposal
RSS Rich Site Summary (file format)
SA Service Associate (Red Cross position)
SDI Spatial Data Infrastructure
SES Socioeconomic Status
SoVI Social Vulnerability Index
SV Supervisor (Red Cross position)
UAV Unmanned Aerial Vehicle
USC University of Southern California
VDAT Virtual Disaster Assessment Team
VDV Virtual Disaster Viewer
VGI Volunteered Geographic Information

xiv
ABSTRACT
The field of study lying at the intersection of geospatial technology and emergency
management has grown significantly in recent years and continues to rapidly evolve.
Developments are largely the result of advancing technology, making display and
analysis of geographic information easier and faster. Collaborative capabilities of
technology have also improved, allowing a skilled volunteer to contribute to a disaster
response operation taking place thousands of miles away. These developments in “crisis
mapping” enable individuals to participate in providing aid and relief through
information creation, dissemination, and analysis. While this is happening, an opportunity
exists to ensure that long-established organizations keep pace with technology and can
adopted to properly accommodate new tools for rapid response.
The American Red Cross is a non-profit organization that plays a significant role
in humanitarian aid and disaster relief in the US. The organization provides many
services, including sheltering, feeding, bulk distribution, health services, and information
dissemination to victims of disaster. However, many decisions made by local chapters of
the organization are not informed by spatial awareness and intelligence. The vast
majority of American Red Cross chapters have no GIS capabilities in their operations.
Despite the status of the Los Angeles Region as one of the country’s most active
Red Cross Regions and a long-standing interest by its emergency disaster response
department, geospatial capabilities were not part of its operations until recently. A Red
Cross GIS Project began in 2009 when the Los Angeles Region acquired GIS data, tools,
and volunteers and developed a framework for the use of geospatial data. This
dissertation uses elements of the project development process to provide insight into how
geographic information can be utilized within the Red Cross.
xv
The dissertation details how the Los Angeles Region and a few other Red Cross
chapters across the country use geospatial technologies. The act of technology adoption is
discussed, including how those tools can be used in day-to-day operations. The
distinction between GIS for disaster planning and disaster response is made and barriers
to implementation are highlighted in the hope that Red Cross chapters interested in GIS
can learn from the trials of others.
To highlight GIS for planning, a case study of GIS to enhance disaster
understanding is also provided. Red Cross disaster response data is analyzed spatially in
Los Angeles County. The analysis is performed with fire response data and demographic
data to provide context regarding the communities affected and those who seek aid.
Exploring GIS as a tool for disaster response, a model for crowdsourced damage
assessment is presented. The value of such a model is that some aspects of disaster
response at any scale, whether local or national, can be rapidly responded to by people
remotely. Chapters that are not regularly engaged in geospatial technology use can
benefit from a network of technically trained volunteers to perform certain tasks, such as
damage assessment following a disaster.
This dissertation documents some of the key ways that GIS adoption has been and
can be accomplished. It relates how GIS can be used in this and similar organizations in
an age where humanitarians can contribute to a cause from behind their computers. Much
opportunity still exists to improve understanding of how aid can be best provided through
these technologies. As the American Red Cross partakes in technological advancement,
lessons are learned that can help inform the development of the digital humanitarian.

1
CHAPTER ONE: GIS AND THE AMERICAN RED CROSS
The American Red Cross has played a significant role in humanitarian aid in the US since
its inception in 1881, and was designated through a congressional charter in 1905 as a
national provider of sheltering, feeding, bulk distribution, health services, and
information to victims of disaster. This non-profit organization provides disaster relief
following hazards at varying geographic scales, ranging from local events such as single-
family house fires to regional and multi-state catastrophes such as hurricanes and
earthquakes. While the Red Cross gets much of its media attention from its work during
the large catastrophes, the majority of its operations are the day-to-day responses to the
local events. The organization accomplishes the task of responding to this multitude and
variety of disasters with a large volunteer base that far exceeds its paid staff.
The organizational structure of the Red Cross can be simplified as consisting of
tiers at national, regional, and chapter (i.e. local) scales. National policies and
management are handled by the organization’s single National Headquarters (NHQ),
which coordinates responses for large disasters using both local and national resources.
The mid-level scale is the Red Cross Region, composed of a number of local chapters,
similar to the cities that make up a county. The Red Cross is able to fulfill its mission
through its network of more than 600 local chapters spread across the nation. The fine-
scale local chapters enable the organization to provide relief following the day-to-day
events such as single-family fires. Each of the 600+ chapters has a jurisdictional reach
that determines the office that independently responds to an event based on its geographic
location, but these local chapters can also receive direction and assistance from NHQ
during large events.
2
Of all Red Cross Regions in the country, the Red Cross Los Angeles Region
(referred to hereafter as the “Los Angeles Region” and not to be confused with the Los
Angeles metropolitan area as a region) is the most populous and covers the most area,
serving over 10 million people in approximately 19,000 square miles. Most of these
people live in Los Angeles County, but the Los Angeles Region extends beyond that
county’s boundaries, including one-third of Kern County as well as Inyo and Mono
Counties in their entirety (Figure 1). The Los Angeles Region is also among the most
active of the nation’s Red Cross Regions, with single-family fires occurring at an average
in excess of one fire per day.

Figure 1 The American Red Cross divides its service areas into regions and chapters.
The Red Cross GIS Project is primarily concerned with the Los Angeles Region,
encompassing Los Angeles County (larger map), Inyo and Mono Counties, and part of
Kern County (smaller map).
3

Despite the status of the Los Angeles Region as one of the country’s most active
Red Cross Regions and a long-standing interest by its emergency disaster response
department (EDR), geospatial capabilities were not part of its operations until recently. A
Red Cross GIS Project began in 2009 when the Los Angeles Region acquired GIS data,
tools, and volunteers and developed a framework for the use of geospatial data. This
effort sprung from this dissertation as a collaboration whereby GIS technology
introduction and use for disaster response could be explored. This dissertation uses
elements of the project development process to provide insight into how geographic
information can be utilized within the Red Cross.
The vast majority of American Red Cross chapters still have no GIS capabilities
in their operations (through this dissertation research, an estimate of less than 15% of Red
Cross Regions). Because of the Red Cross GIS Project, the Los Angeles Region is
considered a leader in the implementation and use of GIS at the local chapter level. While
different Red Cross chapters have different needs, similarities in chapter structures and
functions allow for the lessons learned at this chapter to be transferred to others. In
general, these lessons may prove useful for other similar organizations, whether they are
similar in focus (i.e. disaster relief and humanitarian aid) or workforce management (i.e.
a non-profit that utilizes volunteers and interns for essential support). Further, the use of
GIS in Los Angeles is uncommon, but not wholly unique. Other chapters that use GIS
can help inform and support the validity of these lessons.
The dissertation examines GIS at the Red Cross at two scales: the local and
national levels. Specifically, the next dissertation chapter examines GIS implementation
4
at the local level within chapters across the country, providing fine-scale exploration on a
national stage. It details how the Red Cross uses those technologies as well as how those
tools affect day-to-day operations. The distinction between GIS for disaster planning and
disaster response is made. Barriers to implementation are also highlighted in the hope that
Red Cross chapters interested in GIS can learn from the trials of others.
Following this, a use case of GIS to enhance disaster understanding is provided in
Chapter 3. Red Cross disaster response data is analyzed spatially in Los Angeles County.
The analysis is performed with fire response data and demographic data to provide
context regarding the communities affected and the communities that seek aid. This use
case can be replicated at different chapters as an introduction to the utility of geospatial
technology beyond the desire to simply put points on a map. This analysis also represents
GIS use for planning, where historical data is used to better understand the communities
that the Red Cross works in.
The fourth chapter returns to a national scale and discusses a model for virtual
disaster response work, with an emphasis on response over planning. The value of such a
model is that some aspects of disaster response at any scale, whether local or national,
can be rapidly responded to by people remotely. Chapters that are not regularly engaged
in geospatial technology use can benefit from a network of technically trained volunteers
to perform certain tasks, such as damage assessment following a disaster.
The Red Cross GIS Project effectively introduced a GIS department to the Los
Angeles Region along with the concept of thinking about their data spatially. This
produced a broad range of ideas and work that included the conceptualization of GIS
needs for planning and operations as well as the implementation of a framework designed
5
to create, distribute, and display geospatial information. The data products and human
network that sprung from this effort have further helped create a larger role for GIS at
other Red Cross chapters and have aided in the development of national GIS standards,
workflows, and best practices. This dissertation documents some of the key ways that this
has been accomplished, relates how GIS can be used in the organization, and where the
applications may head as technology adoption grows.

6
CHAPTER TWO: GEOSPATIAL TECHNOLOGY IMPLEMENTATION AT
LOCAL RED CROSS CHAPTERS

Publication Information
This chapter was submitted to URISA Journal for publication in August 2014.

2.1 Introduction
Geospatial technologies such as geographic information systems (GIS), vehicle tracking,
and location-based mobile applications have the potential to add great value to disaster
response operations. While these technologies have been available for years, the majority
of local chapters of the American Red Cross do not use any such solutions for responding
to disaster events. Such technology could also be used for day-to-day Red Cross
activities, since much of their planning work can be improved by location information.
For example, the pre-positioning of shelter supplies in shipping containers can be aided
by spatial intelligence, helping to answer questions regarding areas of greatest disaster
risk or areas with the densest populations. This paper explores the adoption and use of
geospatial technology by some Red Cross chapters and identifies barriers to
implementation as well as opportunities for engagement. An important caveat on the
findings presented here is that new developments are always like to occur with frequent
technology advancement and adoption, but this snapshot can remain useful as providing
several lessons in GIS development for any organization.
The process of creating a GIS department and identifying the geospatial needs of
a local Red Cross chapter has not proven to be as straightforward as hiring a GIS analyst
and procuring GIS software. In cases of success, a geospatial culture had been introduced
7
to a non-profit organization which was previously only using maps to identify driving
directions with Google Maps and hard-copy Thomas Guides. As an example, the
Emergency Disaster Response Department (EDR) of the Los Angeles Region readily
understood that GIS could be important to emergency response but was largely unaware
of how to use it in relief operations. Despite the ubiquity and important role of the
organization, technical resources were lacking. However, the identification of the need
for GIS by EDR reflects a spatial turn in information management that was partly fueled
by the popularization of Google Maps and Google Earth.
Through structured interviews of geospatial leaders at GIS-enabled Red Cross
chapters, lessons have been learned that can be used by other chapters interested in
similar technology solutions. While these interviews were tailored toward an audience of
emergency managers, some of the findings are universally applicable, independent of an
organization’s professional field. In particular, the needs of local chapters of the
American Red Cross for planning and response are determined. Accomplishments have
also been observed and challenges have been identified, creating insight into the
endeavor of GIS adoption.

2.1.1 Disaster GIS
GIS and geospatial tools have been used for emergency management and disaster
response in many ways (Cova 1999, Gunes and Kovel 2000, Cutter 2003). This is best
understood in the context of a community’s needs surrounding disasters, and those needs
are often expressed through the emergency response cycle (Cutter 2003) which is
commonly approximated as consisting of four steps: mitigation to reduce risk of
disasters, preparedness to reduce vulnerability, response during the event, and recovery
8
afterward (Godschalk 1991, Cova 1999, Tierney et al. 2001). GIS has relevance for many
activities associated with each of these phases (Figure 2). For example, before a disaster
strikes, GIS can play a part in mitigation through analytical modeling, assessment, and
management. The threats of a natural hazard such as a fire or hurricane or a man-made
hazard such as toxic materials storage can be modeled with GIS, allowing policies or land
use decisions to be made to reduce the hazard (Zerger 2002). The same reasoning can be
applied to assessments of vulnerability and risk.

Figure 2 A simplified disaster cycle depicting four stages to disaster where an
application for GIS is described in each stage (from Cova 1999 and Godschalk 1991).

There exists some overlap between the mitigation and preparedness phases as
well the mitigation and recovery phases in terms of how some disaster research
9
applications of GIS can be used. For example, risk mapping can also be utilized for
preparedness purposes. However, some GIS applications are clearly preparedness
oriented, such as the distribution of disaster relief equipment. Another example is the pre-
determination of the best routes for emergency vehicle deployment. Other aspects of
spatially informed planning include evacuation models for community residents (Pidd et
al. 1996, de Silva and Eglese 2000, Shahabi and Wilson 2014). Real-time or near real
time data can also provide early warnings to prepare for disasters, such as hurricanes and
tornadoes (Cutter 2003).
There is a clear role for mapping capabilities during the response phase of a
disaster. One of the more obvious needs by emergency managers in terms of GIS is a
map for situational awareness (Johnson et al. 2011). The availability of technologies such
as mobile GIS and digital video have expanded the ability of geospatial technologies to
bring real-time or near real-time data to improve situational awareness (Montoya 2003).
Disasters create demands on society beyond its ability to supply (Tierney et al. 2001),
with a clear example being food, water, and shelter. GIS can be used to aid in triaging the
distribution of supplies with resource deployment maps. Search and rescue efforts are
aided by the analysis of an area in GIS. Evacuation routing for communities that need to
move away from danger is also made possible with geospatial tools (e.g. Cova 1999,
Shahabi and Wilson 2014).
Geographic information technology continues to stay relevant to the emergency
response cycle once the response phase has progressed to the recovery phase. This can
include damage assessment done with GIS to determine the who or what requires
assistance in recovery (Kiltz and Smith 2011). Damage from disasters can vary spatially
10
and so it follows that recovery should too. Another GIS application that follows is the
analysis of recovery and rebuilding efforts over time.
An example of a disaster where geospatial tools were implemented in response is
the World Trade Center attacks of September 11, 2001. The response to these terrorist
attacks demonstrated the utility of geospatial tools in emergency management for both
rescue and relief as well as the distribution of resources and information dissemination
(Cutter 2003). From this disaster an Emergency Mapping and Data Center (EMDC) was
established and provided decision support for several months after the event (Kevany
2003).
GIS also played a role in Hurricane Katrina in 2005. During the disaster, the
Louisiana State University GIS Clearinghouse Cooperative was created to centralize
geospatial data germane to the hurricane and related relief efforts (Mills et al. 2008). This
clearinghouse allowed for data dissemination that could be used in all phases of the
emergency response cycle. In particular regard to recovery, geospatial analyses have been
extensively employed to aid affected communities such as through the development of
recovery maps and recovery report cards (Mills 2008). In another example of its role in
the recovery and mitigation stages following the disaster, GIS was used in the analysis of
health disparity that arose as a consequence of the hurricane’s destruction (Curtis 2008).
Geospatial tools also helped aid recovery efforts with the use of spatial video to
understand the root causes of post-traumatic stress disorder (Curtis et al. 2007).
There are many documented uses for GIS in the field of disaster response, but all
the above mentioned uses require some knowledge of the technology. Bringing GIS to
the Red Cross is difficult not because there is no role for it there, but because the
11
technology has not permeated very far into the organization. The acquisition of the tools
and the geospatial culture in general is the first challenge that faces an organization
interested in GIS.

2.1.2 GIS Diffusion
Acquisition of a technology does not guarantee successful adoption, as it is just an
intermediate step to full integration (Onsrud and Pinto 1993). The successful introduction
of innovative technologies into a company or organization is referred to as diffusion, and
so the introduction of GIS technology is referred to as GIS diffusion. The classical model
of diffusion of innovations is composed of four elements: (1) the development of an
innovative idea or technology; (2) the spread of that idea via communication channels;
(3) a period of time for the diffusion to occur; and (4) the social system through which
the introduced innovation propagates (Rogers 1976). The adoption of innovation is
typically generalized by a sigmoidal curve with time on the X-axis and number of
adopters on the Y-axis, signifying few early adopters followed by a steep rise in adoption,
followed finally by a few late-comers. The field of diffusion research tends to be more
appropriate for private market analysts given the applicability of the topic to technology
sales, making the data from this sector publicly inaccessible. However, GIS diffusion
research did occur in the 1990’s, providing insight into how organizations and agencies
are able to implement this technology.
The reason that technology diffuses may be understood by any of three theories:
technological determinism, economic determinism, and social interactionism (Campbell
1996). Technological determinism assumes that diffusion occurs naturally given
technological advancements. Economic determinism is similar, but focuses more on
12
computerization as the big step to economic development. This view links technology to
economic progress. While these two views make sense, these models are easily criticized
as over-simplified. In social interactionism, technology is assumed to be socially
constructed and diffusion happens because of social interaction. This process is not based
on technology or economic growth and accounts for varied experiences (Campbell 1996).
Another common reason for adoption is simply the need to keep up or get left behind at
both the organizational and individual levels (Moore 1993).
Given that much of this research is over a decade old, organizations adopting GIS
now could be considered late-comers, but the ever-changing nature of geospatial tools
makes this field of research relevant today. There is a high degree of re-invention in GIS
due to the wide range of problems that the technology can be used to solve. In particular
regard to multi-jurisdictional GIS, new tasks and needs are known to arise (Greenwald
2000). Additionally, GIS has traditionally not been user-friendly, slowing its adoption
rates (Rogers 1993) and explaining the absence of these technologies in many
organizations.
While different companies will adopt technology differently due to variations in
organizational structures, systems, resources, and users (Budić and Godschalk 1994,
Campbell 1996), there are commonly recurring themes in diffusion. GIS diffusion can be
explained with a content-model approach, where focus is put on the variables that
influence the success of adoption. For example, these variables might be the visibility of
benefits, the complexity in learning to use the innovation, “trialability” of the innovation,
its compatibility with existing systems, or social norms. A contrasting method of
understanding diffusion is the process model, which describes adoption as a function of
13
processes. These include the processes of initiation, where an organization first learns of
the innovation and decides to adopt it, and implementation, where an organization takes
steps toward the adoption (Onsrud and Pinto 1991, 1993).
Many variables have been identified under the content-model that help to explain
the success of GIS diffusion. GIS vision in itself can be considered to be the driving force
behind adoption. Senior management may be behind that vision and has been identified
to be important (Chan and Williamson 1999) if not completely necessary (Onsrud and
Pinto 1991, Sieber 2000, Croswell 2009). Successful projects require senior management
support for acquisition, but all groups must be on board for full implementation success
(Moore 1993). User training, involvement in system design and implementation, previous
technological experience, and support at the administrative level all increase
implementation success, while organizational conflict and instability reduce the
likelihood of success (Nedović-Budić and Godschalk 1996).
Onsrud and Pinto (1993) used questionnaires to examine GIS diffusion in local
government agencies all over the world, though the majority of surveys came from the
U.S. With the content-model in mind, they identified 34 variables associated with
diffusion and constructed questions around those factors. After responses were collected,
a factor analysis was performed to group the 34 variables into 11 factors (Table 1). The
responses revealed that the five most important factors for GIS adoption in that survey
were access to learning, ease of use, cost, utility, and benefit extended to users, with the
greatest of these being utility.

14
Table 1 Eleven factors for GIS diffusion with 34 sub-factors between them, taken from
Onsrud and Pinto (1993). Sub-factors in bold represent variables considered highly
important to the decision to adopt GIS. Factors with one or more sub-factors considered
highly important also appear in bold.
Factor Sub-Factor
Access to learning Ability to talk with previous adopters
Accessibility to GIS vendors
Accessibility to informal educational opportunities
Availability of learning materials
Availability of formal education opportunities
Visibility of uses and benefits
Ease of use Availability of existing data
Ease of transferring data
Availability of skilled GIS people
Compatibility of GIS with existing computer system
Effects of use Adverse ramifications to advocates if GIS fails
Adverse ramifications to organization if GIS fails
Benefits to advocates if GIS is a success
Cost Ease of making pilot study
Cost of hardware and software
Cost of data entry
Cost to retrain staff
Utility Data accuracy
Ease of generating results
Advantages of GIS over existing processes
GIS consistency with corporate goals
Ability to expand types of uses
Benefits to extended users Ability to serve several units of the organization
Ability to serve those outside organization
Adapting GIS to meet your specific needs
Importance of GIS champion
Communications channels Existence of formal communication channels
Existence of informal communications channels
Compatibility and past
success
Extent of existing computer usage skills
History of successful computer systems in organization
History of past failures History of unsuccessful computer systems in organization
Fallback options Ability to revert in case of GIS failure
Availability of informal methods to cover costs
Proximity to other users Close proximity to other GIS users

15
Important processes for GIS acquisition were also identified in the same survey
and respondents indicated whether or not these steps were taken in their organization
(Table 2). These processes were common for both successful and unsuccessful adopters.
One of the least common processes undertaken was the acquisition of a GIS consultant,
but for many who did, this also happened to be the first step. The ordering of processes
was typically the same regardless of whether or not that first step was taken.

Table 2 Eleven processes for GIS diffusion in order, taken from Onsrud and Pinto
(1993).
Process Percentage of Respondents Who Undertook the
Process
Seek and acquire a GIS consultant 55
Prepare informal proposal for GIS introduction 78
Identify GIS user needs 93
Seek staff support for GIS 87
Match GIS to tasks and problems 85
Identify GIS location within organization 83
Prepare formal proposal for GIS introduction 76
Undertake request for proposal (RFP) 80
Conduct a pilot project 76
Enter a contract for purchase 96
Acquire GIS technology 100

2.2 Methods
A call for GIS users was distributed to each of the 600+ Red Cross chapters in the
country during the summer of 2013 regarding their current and desired uses of geospatial
tools through the Disaster Services Technology (DST) newsletter published and
16
distributed to all chapters by the National Headquarters (NHQ) of the American Red
Cross. The GIS users and eventual interviewees were identified in this way, either
through direct responses or referrals from another GIS user. In addition, the geospatial
technology manager at the Red Cross Headquarters had already begun compiling a list of
GIS users who had approached him for access to software.
The questionnaire used to guide interviews was split into four sections: (1) basic
information; (2) existing GIS efforts; (3) GIS needs; and (4) GIS adoption hurdles. All
interviews were completed over the phone and ranged between 24 to 77 minutes in
length. Questions were ordered to accommodate for the diversity of project maturity,
since chapters have different types and degrees of GIS adoption. For example, a chapter
with a fully mature GIS team will likely provide map services to other departments
within their office, whereas other teams may be completely dependent on external
partners to provide maps.
Prompts were utilized to indicate which questions should be answered during an
interview depending on the relevance of those questions to the chapter’s GIS situation.
However, many questions were designed to be common for all respondents so that a
comparison can be made across the different types of GIS efforts. For example, questions
regarding what the chapter’s mapping needs are were asked in a way that was relevant to
both mature and new GIS teams.
A part of the “basic information” section inquires about the person being surveyed
and his or her role within the organization. This information was used to determine the
capacity of chapters to devote time to mapping endeavor and to ensure that the
respondent was a good representative of his or her chapter. Everyone interviewed was
17
either a Red Cross employee or authorized to speak about the organization by an
employee. The remainder of the sections consisted of a combination of closed-ended
response formats, utilizing nominal-polytomous formats and Likert scaling.
The second section of the questionnaire focused on the current state of geospatial
tool use including both electronic and paper-based tools. Some chapters already do some
form of computer-based mapping, including the creation of KML files for viewing in
Google Earth. Some chapters may collaborate with agencies and universities to do
mapping, and some may have even developed map applications and use GIS software
(e.g. ArcGIS, Depiction). The answers to this section of the questionnaire provide insight
into common paths and efforts in enabling the Red Cross with GIS. The subject of
datasets is again important to understanding the needs of the particular chapter. If the
chapter has done some form of mapping, a focus is put on what datasets have been
mapped, and which may have garnered enough interest to initiate mapping in the first
place.
The following section of the questionnaire then inquired about the mapping goals
of the chapter. These questions examined the extent to which the chapter has
contemplated its potential use and integration of geospatial tools in their planning and
operations. Geospatial datasets were discussed here as well in the context of a computer-
based map viewer with pre-loaded map layers. The purpose of these questions was to
determine what the respondent expects to see when they pull up a map and how they
expect to use those datasets to assist them in their activities. They also ask about the
importance of mapping common Red Cross datasets, some of which are tracked
18
throughout the country by NHQ databases. There was an additional focus here on
whether those activities are more related to planning or response.
The final section of the questionnaire was meant to identify the hurdles that exist
between goal and the current state. For most chapters, the hurdle could likely be that
mapping was never considered, while other chapters could describe some mapping they
have done (as identified in the previous section). These chapters can provide further
information on the obstacles that face an organization like a local Red Cross chapter.
Many of these obstacles may be easy to predict, but this questionnaire helped pinpoint the
obstacles that are most difficult to overcome in bringing a spatial tool to their
organization.

2.3 Results
A total of 13 interviews were performed over the phone, though a total of 18 chapters
were represented by the interviewees, as five chapters were represented as receiving aid
from one of the 13. In addition, a representative from International Services at NHQ was
interviewed, representing GIS in an international context. While their interests may differ
from domestic relief operations in some ways, the lessons learned from their GIS
adoption process are equally valuable.
There were four recurring impetuses for GIS at Red Cross chapters. Everybody
interviewed fell within one of these origins, listed below in order of least to most
common. While most chapters could list several reasons for exploring geospatial
technology, the single driving force behind each chapter’s decision was clear, allowing
for classification by impetus. These impetuses can be characterized as: (1) building a
common operating picture (COP); (2) using GIS for damage assessment (DA); (3) an
19
unstructured exploration of GIS; and (4) a deliberate attempt at building a GIS
department.

2.3.1 Type 1: Building a Common Operating Picture (COP)
The idea of a single and comprehensive COP is pervasive in the emergency management
community. Disaster situations create the need for a variety of information which comes
from multiple sources (e.g. traffic information from transportation departments and fire
perimeters from fire departments). A COP unifies this information, enabling decision-
makers to see many variables at once. The virtue of a COP becomes clear when data
sources should clearly interact, such as road closures informing the opening or closing of
evacuation centers and shelters.
Geographic information is a major component of any COP, as a map provides
visualization for variable interaction. To create a COP, an organization needs to be able
to work with geospatial data. Thus the development of a COP creates a necessity for
geospatial infrastructure, and becomes a pathway to adopting GIS technology.
A prime example of a COP created for the Red Cross is the San Diego chapter’s
SitCell, used to provide a comprehensive look at situations. As any good COP does,
SitCell goes beyond the map. Other elements of the tool include the ability to centralize
calendars for scheduling and hosting a single hub for document sharing. The result is a
platform that removes less efficient workflows of e-mailing others to find and submit
forms, such as those used by damage assessment teams. Coordination of assets can be
performed in real-time, ensuring that vehicles and supplies are always tracked and
accounted for. The map component of the tool allows for a quickly interpretable view of
the multiple pieces of a response operation.
20
Chicago, wanting to create a COP but not wanting to re-invent any processes or
tools already developed by others, has adopted SitCell technologies and procedures. The
intent was the same as that of San Diego: to put all the moving pieces of the operation
into a single place where they can be assessed together. An added benefit to adopting San
Diego’s tools was ensuring interoperability of platforms. While the two chapters are
geographically distant and are unlikely to ever need to share data for a response
operation, this interoperability becomes extremely important for the likely scenarios of
one chapter providing remote auxiliary support for the other, as well as sharing lessons
learned and best practices in the parallel evolution of their technologies.
A common thread between these ambitious projects of creating a full COP was
that they were both driven by Red Cross employees rather than volunteers. This does not
imply that volunteers did not play a role in these projects; in the case of San Diego, the
project lead is a volunteer. However, San Diego was still able to invest employee hours
and their operating budget in the project. This fact occurs in contrast to other projects to
be described below, which were driven by volunteers and had fewer resources (either
time or funding) devoted to them.
Because the use of GIS data is less a focus in COP development than a
component, neither of these efforts was informed by traditional GIS practices. Rather,
both were heavily focused in less traditional web GIS, where data formats were often
GeoRSS feeds, GeoJSON, or KML as opposed to the shapefiles and geodatabases that
make up major data sources for many GIS professionals. Specifically, neither effort was
initially led by a technician trained in desktop GIS software. However, the traditional
21
desktop-based GIS methodologies and workflows were later added to the COP software
solution during research for its development.
The construction of a COP may not have been the driving force for GIS use at
many Red Cross chapters, but it was very often a part of the overall vision for technology
and information management. The remaining three impetuses do not preclude the use of
GIS in a COP later on, and simply illustrate different origins for GIS adoption. In these
cases, GIS can be integral in the eventual development of a COP.

2.3.2 Type 2: Technology for Disaster Assessment (DA)
Volunteers of the American Red Cross can choose from a variety of activities to be
involved in. For example, those with experience in financial reporting can offer their
services to that function in a response operation. While spatial information can be
relevant to virtually all activities, volunteers from just one activity, DA, have emerged in
multiple chapters and states as pioneers in GIS use. This activity is responsible for these
individuals visiting disaster-affected communities and determining the extent and degree
of damage sustained by residences. Information produced by these volunteers is used by
the Red Cross to determine how to provide relief. DA volunteers have made much use of
maps for planning assessment routes and for visualizing the results of their work. In
particular, the chapters that are active in mapping for damage assessment are in states
where tornadoes are a recurring threat.
The three respondents representing this front are from the Central and Western
Oklahoma Region, the Mid-Illinois Chapter, and the Greater Indianapolis Region.
However, none of these efforts could be described as being contained within the borders
of any of these functional geographies. Just as disasters know no geopolitical boundaries,
22
none of these efforts considers the boundaries of their home offices before creating maps
to help with a response to a tornado or storm. At the same time, none of these efforts is
nationally comprehensive in coverage; a tornado in Missouri is likely to be covered by
one of these three efforts while a wildfire in California is not. The geographies for
inclusion are not well-defined, but are generally aligned with “Tornado Alley”, the area
of the U.S. where tornadoes are most frequent, encompassing the mid-west, the south,
and parts of the east coast.
Likely because of the geographically widespread nature of tornado threats, much
coordination occurs between these three efforts, particularly on the social media website
Facebook. Among them, a team has developed with the intent of being able to provide
intelligence (much of it geographic) following a disaster. The Virtual Disaster
Assessment Team (VDAT) is completely volunteer-driven and organized so that those
experienced in DA and technologies in GIS can remotely assist in a disaster if requested
to do so. VDAT currently exists simply as a relatively informal list of such trained
volunteers.
Just as was the case with the chapters interested in full COPs, the volunteers from
these chapters were not traditionally trained in desktop-based GIS. The key difference in
implementation between the COP chapters and the DA chapters is that all these projects
are initiated and run by volunteers, not employees. In every case, the impetus came from
volunteers who attempted to provide good information and visualizations with geospatial
tools that were widely available and easy to learn.

23
2.3.3 Type 3: Unplanned Exploration in Mapping
Three chapters were identified as beginning an exploration of GIS simply through
volunteer or employee interest in the technology. In all cases, the GIS users had no
background in GIS and were not looking at it for a specific purpose such as DA. There
was also no larger goal, such as the development of a COP or a GIS department.
These individual chapters in Oregon, North Carolina, and Ohio had all heard
about GIS at some point and were simply exploring ways to put their data on a map. The
future of these initiatives is unclear, as the efforts largely reside with a single person
developing a skill, similar to developing database design skills. In this sense, these may
not yet be large enough efforts to be considered GIS projects, but the individual efforts
put into these was significant to warrant an interview.

2.3.4 Type 4: The Traditional GIS Department
The single most common type of GIS project was effectively the development of a stand-
alone GIS department. A total of 10 chapters fell into this category, consisting of one
each in Colorado, Florida, New Hampshire, Pennsylvania, Texas, and Virginia, and four
in California. Additionally, the two GIS teams at NHQ (domestic and International
Services) would be classified as beginning with this impetus.
A slight majority of these initiatives were started by Red Cross employees while
the others had begun with a volunteer with a background in GIS. The idea of providing
geospatial services to the chapter was intended to capitalize on the horizontal nature of
geographic information. While several of these projects had some beginnings with
providing situational awareness in a response operation, the difference between these
designs and those by the COP developers was: (1) the complete focus on having those
24
involved handle just geographic information; and (2) a smaller focus on providing a
single tool to achieve that goal. These projects also tended to have a greater interest in
how the data would be used for planning than for response.

2.3.5 Trends in GIS Adoption
Three potential map layers were tied for the most desired: hazard/risk area, Red Cross
shelters, and logistical supplies (Table 3). This is unsurprising given that the geographic
location of hazards is important for both disaster planning and response, and there is no
more effective way to communicate that location information than with a map. Shelters
and supplies are also likely to be top picks because they are the two most important assets
with geographical importance in a disaster: the location of a shelter must be far from a
danger zone while being close in proximity to those being served, and supplies have to be
located nearby to stock shelters and other emergency centers. Human capital such as
partner organizations and volunteers can be easily transported, so their “home” location is
not as important.
The three least desired map layers were volunteers, outreach events, and donors.
These are also likely the three least important on this list in terms of a disaster response.
As mentioned, the volunteers are movable, and outreach events and donors represent
different stages of the disaster cycle. In fact, outreach events and donors have the least to
do with the disaster response departments of a Red Cross chapter, and were only desired
by those who had been building out a traditional GIS department. These traditional teams
focused on GIS data management for the entire organization are much more likely to
work with other departments on issues such as outreach and fundraising than the COP or
DA teams would.
25

Table 3 Interest in map layers by Red Cross GIS users, averaged and then rank ordered
from most to least interest (average score in parentheses; 5=most and 1=least).
Map Layers Rank (Average Score)
Hazard/Risk Areas 1 (4.54)
ARC Shelters 1 (4.54)
Red Cross Logistical Supplies 1 (4.54)
ARC Disaster Responses 4 (4.46)
Census/Demographics 5 (4.31)
Response Boundaries 6 (4.23)
Red Cross Partner Locations 6 (4.23)
Political Boundaries 8 (4.15)
Critical Infrastructure (e.g. public utility facilities) 9 (4.00)
ARC Volunteers 10 (3.69)
ARC Outreach Events (e.g. preparedness education) 11 (3.54)
ARC Donors 12 (2.85)

Interest in layers is influenced by varying levels of program maturity. The value
of non-response layers such as outreach and donors may be clear to disaster managers,
but is a much lower priority than the basics of disaster response (e.g. sheltering, incident
locations) when starting a new program. Demographic layers may seem like a lower
priority to those outside the field, but understanding communities is important even
before people are deployed to respond. The disaster responder is trained to be prepared
for a variety of situations, but having knowledge of languages spoken, for example, can
influence who best to deploy.
26
The most important factors to GIS success were software availability and the
presence of a GIS champion (Table 4). This result is also unsurprising, as GIS requires
the software and user at a minimum. However, it was clear to all the interviewees that a
GIS champion was recognized as more than a user. Several individuals commented that
GIS had been too complex a technology to properly adopt at their chapters without an
individual devoted to the process who could act as an ambassador for spatial data. The
emphasis on software was due to the fact that many chapters already had a willing GIS
champion, but did not have access to the software needed to create and analyze spatial
data. For most chapters, this was a process that could span many months, particularly
when the efforts were volunteer-driven and no funding was available for these projects.

Table 4 Important factors for GIS success by Red Cross GIS users, averaged and rank
ordered from most to least important (average score in parentheses; 5=most and 1=least).
Importance to Project Rank (Average Score)
Software availability 1 (4.62)
Presence of a map project lead (GIS champion) 1 (4.62)
Hardware availability 3 (4.38)
Access to data internally from various sources 4 (4.31)
General interest of senior management 5 (4.15)
Tech savvy volunteers 5 (4.15)
Access to data from partners 7 (3.62)
Well-defined workflows 7 (3.62)
General interest of chapter 9 (3.54)
Standards for data security/privacy 10 (3.46)

27
Hardware availability was listed as the third most important factor, even though
the interviewees had a clear understanding that this factor related to their needs despite
their current equipment. While desktop computers can be found nearly everywhere, the
interviewees generally had an impression that the GIS software had to be run on state-of-
the-art machines. Because the majority of office computers in these disaster response
departments are designed for word processing and spreadsheets, they are likely to have
slow processors, little memory, and relatively little hard disk space. In addition, there was
a general impression that specific computers would be needed to be dedicated to the
activity of GIS. Even though people technically have hardware, the availability of fast
and powerful hardware was perceived as being among the most important factors
influencing the likelihood of success.
The four factors of least interest all received average scores of less than 4 on a
scale from 1 to 5, 1 being least important and 5 being most important. One of these four,
general interest of the chapter in GIS activities, is likely of little concern because the
chapter’s interest is unrelated to the sustainability of a project (as opposed to interest of
senior management). The other three factors relate to program maturity: access to data
from partners, well-defined workflows, and standards for data security and privacy. For
projects that start with the immediate goal of mapping a chapter’s own data, these goals
are all more advanced. Data access from partners is a secondary concern after producing
Red Cross GIS data. Well-defined workflows assume recurring data processes and a team
to share the workflow with. Privacy and security, while important at all stages of
development, are most important once the data is packaged to be widely distributed.
28
The distribution of efforts focused more heavily on planning than response
(Figure 3). In Red Cross nomenclature, GIS efforts are currently oriented toward “blue
sky” planning as opposed to “gray sky” response. This result may seem counterintuitive
given the GIS industry’s emphasis on tools and mobile applications for rapid response to
an incident. The likely explanation for this distribution is that GIS technicians and
analysts are able to prepare the infrastructure for GIS, but the implementation of new
technology during a disaster requires greater effort to achieve. Well defined tasks and
workflows are essential to efficient response, and the introduction of a new technology to
the situation must be thoughtfully incorporated with cooperation from existing
responders. The many people involved in response represent a variety of activities within
the relief operation, including sheltering, mass care, client services, feeding, public
relations, etc. GIS must be introduced intelligently by coordinating across these groups so
as not to disrupt current processes or chains of command.
The DA-driven teams are unsurprisingly focused on response. Their activities
center around events such as tornadoes, and the maps they produce were always intended
for immediate field use. The COP teams are similar, but as COPs require careful design
to effectively produce useful knowledge, those teams find value in both planning and
response. The COP projects described themselves as about being equal between the two,
though the impetus is probably more response related. Some of the traditional GIS teams
emphasized response, but they were largely more planning focused.
In general, teams are very small, with no more than six active volunteers in any
group. At least two chapters had single-person efforts (though they have likely grown
29
since the time of the interviews). The DA teams have no Red Cross employees,
consisting only of volunteers (no more than four on any team).

Figure 3 Distribution of focus (between planning and response) for GIS activities at
surveyed Red Cross chapters.

A little more than half of projects were employee-driven, including most of the
traditional GIS teams (approximately two-thirds) and both of the COP efforts. This
makes sense, as these larger, more ambitious, and more focused efforts require a broad
vision and thrive best when coupled with funding. The DA efforts, on the other hand, are
born from a need in field response.
The primary impediments to map creation were lack of time and resources, consistent
with small teams and limited volunteer work hours.

0%
46%
23% 23%
8%
0%
10%
20%
30%
40%
50%
Planning
Focused
Mostly Planning Evenly Split Mostly Response Response
Focused
30
2.4 Conclusions
The contingent of GIS-enabled chapters is small but growing. These chapters
have found value in geospatial technology in several areas, including visualizing the data
that they collect everyday (but infrequently analyze) and understanding the distribution
and reach of their resources. Many of these have no paid personnel dedicated to GIS and
began their efforts with limited access to proprietary GIS software, but a growing number
of chapters have been able to find some kind of funding, whether for software, hardware,
or employee salaries.
Chapters of the Red Cross that wish to integrate GIS capabilities into their
operations can learn from the lessons of these early adopters. The identified need for a
GIS champion to guide development is largely a result of the complexity of GIS
software; it is a technical skill that is not as quick to learn as spreadsheets or word
processing. However, geospatial software programs have become easier to use in recent
years and have enabled quicker entry into the field. The need for a GIS champion may
diminish slightly over time as a result of the advances in software usability, but likely
only for the less robust GIS solutions. This shift may also represent a widening gap
between full GIS capabilities (i.e. the ability to perform complex geoprocessing and data
management) and geospatial technology as a whole (which includes GIS, but represents
simple geographically-oriented tools as well).
Software and training are also keys to success, but may be difficult to obtain for a
new program. Perhaps the best first step for a new chapter is to obtain the GIS champion
of another chapter as a GIS consultant. With tools and people in place, a basic program
can be quickly set up. Full GIS adoption occurs when the program is sustainable and
integrated with the organization’s other functional units. This integration is a process
31
which can take a long time, but given the similar structures of multiple Red Cross
chapters, early adopters’ experiences can be invaluable.
Due to the digital nature of most geospatial technologies, there is also a ripe
opportunity for virtual GIS and remote assistance. The VDAT team is an excellent
example of leveraging an organization’s resources across a large geography. Rather than
individual adoption, a chapter office can find success in becoming a member of an
existing program like VDAT. Similarly, Chicago emulated San Diego’s SitCell in such a
way that the two programs are fully interoperable with a reduced start-up time for
Chicago.
Physically present teams have proven successful with just one person, but the
scope of work that the effort is capable of obviously increases with more volunteers and
employees. In each case of a successful single-person effort, that person was either an
employee who was paid to devote time to GIS development or a retired volunteer who
had more time than a volunteer with a day job. The teams with the most capabilities
consisted of at least four people.
The Red Cross as an organization has much internal data that are worthy of being
spatially enabled to provide geographic context, including shelter locations, partner
locations, logistical stores, historic disaster cases, and past outreach events. Chapters
interested in creating a traditional GIS department have worthwhile map layers to
develop which have clear geographic meaning and interaction. For example, mapping
sites of historic disaster incidents can help determine the communities that have suffered
the most damage, and can thereby be used to plan outreach events for disaster prevention.
The potential to create a GIS department without an impetus such as the DA field needs
32
still exists, and existing GIS-enabled chapters can provide many use cases for the
technology as it applies to specific Red Cross activities.
Since the beginning of this study, the disaster services branch of the American
Red Cross has undergone a re-engineering of its critical functions, including putting a
greater emphasis on the idea of “Situational Awareness”, which gives emergency
responders access to all the information needed to appropriately respond to an event
using effective platforms. Maps and geographic data are a significant component of
Situational Awareness since they are able to visually see a wealth of data at once, identify
geographic patterns, and observe spatially-based interactions between datasets. While this
emphasis on Situational Awareness does not necessitate development of GIS programs
per se, descriptions of the Situational Awareness activity do include mentions of mapping
and GIS. As chapters develop these information and intelligence departments, the lessons
learned by early GIS adopters can aid in creating systems for comprehensive disaster
intelligence with geographic analysis capabilities.

33
CHAPTER THREE: MAPPING FIRES AND RED CROSS AID WITH
DEMOGRAPHIC INDICATORS OF VULNERABILITY

Publication Information
This chapter was submitted to Disasters for publication in August 2014.

3.1 Introduction
A key component of the mission of the American Red Cross is to help people prepare for
disasters before they happen. Understanding the social vulnerability of the different
communities they serve helps the Red Cross accomplish its mission in at least two ways:
(1) by allowing the organization to strategically distribute logistical assets for pre-disaster
placement based on expected need; and (2) by determining the communities that could
most benefit from outreach and education. This chapter focuses on validating
demographic variables commonly identified as social vulnerability indicators with the
goal of providing support for decision-making and providing a richer understanding of
those served by the Red Cross. A distinction is also made between the first responses by
fire departments and the aid that may be provided by the Red Cross following those
incidents.
Spatial patterns of response events become evident after geographically enabling
records on both fire department runs and Red Cross aid. If all households were at equal
risk of fires, the expected distribution of fires would be correlated only with the presence
of households across a landscape. Through analysis of fire records, however, studies have
shown that fire risk will vary depending on a number of factors such as the presence of
smokers, age of housing, installation of smoke alarms, and related demographic
34
characteristics such as wealth (Runyan et al. 1992, Warda et al. 1999, Folz et al. 2011,
Gaither et al. 2011). Additionally, catastrophic events will disproportionately affect
people depending on their vulnerability, which many studies have also related to some of
these same characteristics (Morrow 1999, King and MacGregor 2000, Cutter et al. 2003,
Rygel et al. 2006, Flanagan et al. 2011).
To better understand the communities assisted by the Red Cross in Los Angeles
County, fires and Red Cross aid were mapped over a 5-year time period and compared
with geographic data on demographic variables commonly associated with vulnerability.
The first goal of the mapping was to determine whether or not fires occur to households
evenly across Los Angeles County. For example, if Los Angeles was split into census
block groups and these groups were assigned to classes based on one or more social
vulnerability indicators, would fire response events occur in these classes evenly, or
would they trend in an explainable direction? For which demographic variables will there
be bias? When Red Cross aid events are examined in the same manner, are trends also
observed? Will the biases be generally greater for aid cases than for fire responses?
This study does not take into account likely contributors to fire risk that are not
purely demographic in nature, such as the age of housing or crime rate. The purpose of
this study is not to determine what causes fires, but to determine the kinds of
demographic profiles of those who are affected by fires. The results are intended to assist
organizations such as the Red Cross in understanding the communities they serve and
help them better tailor their approach to achieving their goals. The wider-reaching
findings will include a broader understanding of those who seek relief; not strictly
35
populations that have been identified as vulnerable, but those who have found themselves
in situations of distress.

3.1.1 Disaster Risk and Vulnerability
Disaster situations result in stresses to both organizations and individuals that hinder
expected conditions of life, creating a greater demand for those conditions than society
may be capable of providing (Tierney et al. 2001). Some organizations and individuals
suffer greater stress than others or may have more difficulty restoring those expected
conditions of life, leading to varying levels of vulnerability for different parties. While
there are many definitions of vulnerability that have made their way into the academic
literature (Cutter 1996, Weichselgartner 2001), it can be broadly defined as the potential
for loss; the greater a person’s potential for loss, the greater the vulnerability. Just as
there are many definitions that have nuanced distinctions, there are various models for
understanding disasters where the concept of vulnerability plays an important role.
The Pressure and Release (PAR) disaster model heavily features vulnerability
(Figure 4). This particular model describes a disaster (or risk) formed from two sources
of pressure: (1) the processes that create vulnerability; and (2) the natural hazard or
process that threatens a population (Blaikie et al. 1994). Elements of the PAR model have
been used by the International Federation of the Red Cross and Red Crescent Societies
for vulnerability and capacity assessments (Pelling 2007). The processes that create the
“progression of vulnerability” are the interaction of vulnerabilities’ root causes
(consisting of limited access to resources and political and economic ideologies),
dynamic pressures from society (consisting of the lack of social infrastructure and macro-
36
forces such as rapid population changes), and unsafe conditions (consisting of both
physical and social elements of safety).

Figure 4 The PAR model describes disasters as a function of hazards and vulnerability
(Blaikie et al. 1994).

The hazards that interact with this progression of vulnerability consist of physical
hazards such as earthquakes, fires, and floods. The magnitude of the disaster can be
reduced via a release of pressure on the vulnerability side. A formulaic representation of
this concept is R = H × V, where R is risk, H is hazard, and V is vulnerability.
Shortcomings of the model are the equal weighting of hazards and vulnerability in the
production of risk (Adger 2006), the failure to consider the vulnerability of biophysical
subsystems, and the lack of consideration of the causal sequence of hazards and feedback
mechanisms (Turner II et al. 2003).
The hazards of place model proposed by Cutter (1996) extends the risk-hazard
approach taken by the PAR model, providing more explanatory model parameters (Figure
5). This integrated approach incorporates both risk-hazard and political economy
37
approaches, considering external socioeconomic and biophysical factors (Füssel 2007). In
this model, the progression toward place vulnerability begins with the risk of hazards and
the mitigation measures associated with that place.

Figure 5 The model for hazards of place shows the factors that contribute to place
vulnerability (Cutter 1996).

The risk of hazards is comprised of both the type of disaster and the probable
frequency and magnitude of that disaster while mitigation accounts for steps taken to
ameliorate losses. These factors interact and form the hazard potential of the place, with
risk increasing hazard potential and mitigation reducing that potential. Hazard potential is
filtered given the geographic context and the social fabric of the place it affects,
representing biophysical and social aspects of vulnerability, respectively. These two
aspects of vulnerability are responsible for place vulnerability, which feeds back to alter
both risk and mitigation strategies. This model is commonly cited in the disaster research
literature (e.g. Brenkert and Malone 2005, Greiving et al. 2006, Collins and Bolin 2007,
Holub and Fuchs 2009) and has been used as a framework for the development of other
models (e.g. Borden et al. 2007, Zhou et al. 2009).
38
When viewing the individual components that contribute to vulnerability, the
most significant factor in social vulnerability can be broadly described as socioeconomic
status, which can include percent of households below a federally designated poverty
level, percent population with less than a high school education, per capita income, and
median house value as indicators (Cutter and Finch 2008). While there is no widespread
consensus on the exact indicators that should be used to determine social vulnerability,
there is general agreement about the major factors.
In the most commonly cited study on a social vulnerability index, Cutter et al.
(2003) describe six key groups of vulnerability factors consisting of: (1) lack of access to
resources and knowledge; (2) limited access to political power and representation; (3)
reduced social capital and social networks; (4) building stock and age; (5) frailty and
physical limitations of individuals; and (6) the type and density of infrastructure and
lifelines. Given that many of these parameters are typically measured by seeking data
directly from people, social vulnerability mapping involves the use of demographic data,
including the use of previously collected census data (e.g. Cutter et al. 2003, Chakraborty
et al. 2005) or the use of data collected from interviews and surveys (e.g. Adger and
Kelly 1999, Collins 2005).
In the work by Cutter et al. (2003), the Social Vulnerability Index (SoVI) was
produced at the county level for the entire U.S. The model started with 250 variables as
inputs, obtained from a variety of sources. These variables were narrowed to 85 after a
test for multicollinearity was performed. The variables were furthered narrowed to 42
after a set of computations and normalizations were completed to format the data into
appropriate percentages, per capita estimates, and density functions. From there, a
39
principal components analysis (PCA) was performed to create a total of 11 factors, as
shown in Table 5. These factors have been consistently used by other studies dealing with
vulnerability to disaster (e.g. Morrow 1999, King and MacGregor 2000, Rygel et al.
2006, Flanagan et al. 2011).

Table 5 Eleven factors of social vulnerability resulting from Cutter et al. (2003). The
dominant variable in each factor is also indicated.
Name Dominant Variable
Personal wealth Per capita income
Age Median age
Density of the built
environment
No. commercial establishments per
square mile
Single-sector economic
dependence
% employed in extractive
industries
Housing stock and
tenancy
% housing units that are mobile
homes
Race-African American % African American
Ethnicity-Hispanic % Hispanic
Ethnicity-Native
American
% Native American
Race-Asian % Asian
Occupation % employed in service
Infrastructure
dependence
% employed in transportation,
communication, and public utilities

All of the aforementioned studies have demonstrated that people with less wealth
suffer disproportionately from disasters. When broadly dividing disaster events into
segments of preparedness, response, and recovery, wealth can play a huge role in
resilience at all stages. Socioeconomic status (SES) has been shown to negatively
40
correlate with preparedness behavior (Turner et al 1986). This lack of preparation appears
to be independent of perceived risk, as lower SES populations have been observed to
have heightened risk perception (Fothergill and Peek 2004). This trend may be related to
the cost of preparation, but it may also be related to perceptions of how much control
people feel they have over their own lives (Vaughan 1995).
The second stage, response, can be related to wealth as well. In terms of the
likelihood of a fire event, the most likely causes for residential fires are heating
equipment or smoking materials such as cigarettes, and tobacco smoking has been shown
to be more common for low-income populations (Runyan et al. 1992, Jennings 2013).
Older buildings and poorly maintained buildings are at greater risk as well. And in the
last stage of this simplified disaster cycle, the ability of people to recover and restore their
homes and lost possessions after a fire is related directly to their wealth and/or insurance.
For all these reasons, this study looks closely at indicators of wealth as they relate
to fire response and aid. Other variables examined here relate to age and race/ethnicity
due to the identification of these categories as important to social vulnerability. Also, a
focus can easily be placed on them when planning outreach in communities. Age
variables relate to Red Cross training courses for families and babysitters while
race/ethnicity may relate to language as a matter of special attention, ensuring that the
message of disaster preparedness and disaster relief reaches as many people as possible.

3.2 Methods
Fires are tracked by responding agencies to enable reporting on the services they provide.
In the United States, individual fire departments have the opportunity to report their
responses to a national database to contribute to the larger picture of fire response in the
41
country. This system, called the National Fire Information and Reporting System
(NFIRS), provides a rich and valuable data source for understanding the country’s fires
and how local fire departments respond to them. Detailed NFIRS data is made available
to the public by request.
The American Red Cross also keeps a national database where local chapters can
report their activities. The Client Assistance System (CAS) tracks incidents where the
organization has provided assistance to someone affected by any kind of disaster. In other
words, CAS is not limited to fires. This database is not public and is used internally to
better understand Red Cross response and operations.
Both datasets track large amounts of information measured in terms of both detail
and volume. A thorough understanding of each is needed to maximize utility and improve
interpretation. In particular, expert knowledge is needed to filter the databases so that
they can be properly aligned and compared for spatial analysis. While both systems
possess opportunities for tracking some demographic variables, these fields in the
databases are usually blank. This is likely due to the volume of records produced by both
databases and the priority given to essential elements such as time, number of people
affected, organizations involved, etc. when populating these databases.

3.2.1 National Fire Information and Reporting System (NFIRS)
The National Fire Data Center initiated NFIRS in the 1970s. The system is currently at
Version 5.0 and has evolved over the years to incorporate a greater wealth of information.
The U.S Fire Administration’s NFIRS website
(http://www.usfa.fema.gov/fireservice/nfirs/) currently reports that 23,000 fire
42
departments from every state and Washington, D.C. enter their incidents into the
database, accounting for 75% of all reported fires that occur annually.
NFIRS uses a three-tiered system composed of local fire departments, state fire
agencies, and the federal government. When a fire occurs, a local fire officer will
complete a fire report in a standard format provided by NFIRS, creating a public legal
record that will be reported first to the state and then to the national database.
Additionally, each state may set its own reporting requirements for the local agencies.
For some states, reporting is voluntary. Examples of varied requirements include
differences in which “runs” get counted (e.g. fire responses only) or a dollar loss
threshold in a fire. Local communities may also have varying standards for what gets
included (or not) (Ahrens et al. 2003).
The fire report incorporates information such as the data and time that the incident
occurred, estimated property damage, the occupancy of the structures involved, and any
resulting casualties. Finer levels of detail are also possible, including whether or not
sprinklers and smoke detectors were found to be working. Information on the nature of
the fire, such as its cause and origin, may also be captured (Ahrens et al. 2003). Some
data in NFIRS is made available through the website, but a public request for data can be
made to a fire agency representative for each state.
Selecting an appropriate set of attribute queries for the NFIRS database was an
iterative process that required consulting with fire data specialists who work with NFIRS
regularly. The ultimate goal of the export was to have a tabular output with one record for
each fire in Los Angeles County. An attribute-based geographic filter (i.e. by fire
department codes in Los Angeles County) and a time-based filter (i.e. fires between 2007
43
and 2012) provided a start, but there were many more filters that needed to be applied. A
basic query of fire events would yield multiple records for the same fire because multiple
departments could respond to a single event, and each department would report its
involvement as a record. A refined query removed the fire departments that were
providing “mutual aid” to another department that was the primary responder.
Another goal was to analyze only records for fires that the Red Cross might also have a
role in responding to. Residences, including mobile homes, were included in the query,
but other location types were excluded, such as commercial buildings, parked cars, and
trash cans. The final component of the query filtered for incidents occurring in Los
Angeles County using fire department identification codes.
An initial search of NFIRS for fire events in 2011 was conducted with a less
refined filter. This resulted in records for all fire department responses, including non-fire
responses such as medical emergencies. Duplicate values due to mutual aid entries were
not removed and commercial property responses were included. This query resulted in
18,280 records for dispatches in Los Angeles County. The refined query for this year
reduced the number of responses by 89% to 2,092.
Even though the exclusion of certain “mutual aid” values in the query was
extremely helpful in removing multiple records for a single fire, there were still cases
where this occurred. Regardless of mutual aid, this can happen if a house fire happens to
spread from one structure to an adjacent one. Such “duplicates” were manually removed
from the input dataset for this analysis. While those records are often appropriate for
inclusion in many types of analysis (e.g. when looking at the number of structures that
44
received fire response), they were not appropriate in this case, as the focus here is on
singular fire events captured by fire departments and the Red Cross.

3.2.2 Client Assistance System (CAS)
American Red Cross nomenclature designates any event that the organization responds to
as a “disaster”, including both large catastrophes like wildfires and events with smaller
geographic footprints like house fires. When any incident occurs, the Red Cross may
open a case for each household, and the household will consist of one or more individuals
to be given assistance. As a result, a case is likely to be associated with several
individuals, and multiple cases can be generated from an incident. For example, a single
incident consisting of an apartment building fire will generate multiple cases (i.e.
apartment units) and each case will help multiple individuals. The vast majority of
disasters in the Red Cross Los Angeles Region are single-family house fires.
Information on assisted individuals, the cases they belong to, and the incidents
that brought on the response are stored in CAS. This system is not open to the public and
is used internally for reporting. The use of American Red Cross data in this study is
permitted through a local partnership whereby any data products that arise from the
analysis are carefully reviewed to ensure that no sensitive information is released.
CAS only tracks disasters that the Red Cross has a role in responding to. If a house fire
occurs and the affected party declines Red Cross assistance, the incident will not be
recorded in CAS. An example of assistance that could be provided is a small set of funds
designated to help an individual purchase enough clothing to temporarily mitigate the
loss of their closet in a fire. Red Cross assistance is not based on financial need but rather
on damages incurred due to the incident. Because the amount of assistance is typically a
45
small fraction of total damages (i.e. this assistance is far from comprehensive insurance)
and requires effort to obtain, an assumption is made that this dataset represents
individuals with greater financial need.
Red Cross disaster response cases were queried from CAS for all fire-related
events (including single-family fires, multi-unit residential fires, and wildfires) in Los
Angeles County between 2007 and 2012. Because CAS records any event that can lead to
the Red Cross providing aid (e.g. a building collapse, civil disturbance, explosion, flood,
or landslide), non-fire event types were excluded from the query. With only the
geographic and time-based filters applied, fires made up 87% of Red Cross events in the
county.
Each record in CAS corresponds to an assisted individual, and individuals may
belong to a single case. Multiple cases in turn could belong to a single incident. As a
result, “duplicate data” were removed based on incident number to produce a table where
each incident was represented by one data record. A fire in a multi-family household that
affected several apartment units would thus be counted as one fire and could be better
compared to the NFIRS data, which also only produces one record for a multi-family fire.
Both datasets were geocoded using a variety of address locators to maximize the
number of usable records. While the NFIRS database does include fields for latitude and
longitude, those fields were sparsely populated. The absence of these data could be due to
a large number of factors, including the lack of a GPS unit in the field for acquiring
coordinates and the cumbersome task of entering long strings of numbers weighed
against immediate need. Moreover, data entry is not necessarily completed by someone
with knowledge of geographic data or GIS skills. Given this lack of latitude/longitude
46
coordinates, automatic geocoding of addresses was performed followed by manual
iterations where unmatched addresses could be resolved by making small obvious
changes to addresses. For example, many intersections recorded by fire departments were
written in the form of “Main St X Broadway Ave”, which had to be corrected to “Main St
and Broadway Ave”.

3.2.3 Demographic Variables
Analysis of demographic variables was performed using data from the Esri Business
Analyst software package (2012 version, Esri, Redlands, CA) from a variety of sources,
including the U.S. Census Bureau. The variables chosen reflect the common indicators of
vulnerability related to age, wealth, and race/ethnicity. The census block group was
chosen as the geographical unit for this study because it was the finest scale unit available
with comprehensive data coverage. In Business Analyst, an attribute-based query for
block groups in Los Angeles County was performed, yielding 6,422 block groups.
Three variables were chosen for each of the age, wealth, and race/ethnicity
categories for a total of nine variables. Expected correlation to vulnerability was
determined and summary statistics were calculated for the indicators (Table 6). Every
census block group was assigned either a Higher (H) or Lower (L) vulnerability
designation for each variable with the cutoff between the two classes being the median
variable value. Block groups cannot be generalized as being either H or L, as these
designations are specific for each variable: a block group may be H for median age but L
for median household income. Which block groups were H or L depended on the specific
variable of interest.

47
Table 6 Demographic variables used for analysis with fire and aid events. Every census
block group was assigned either a Higher (H) or Lower (L) vulnerability designation for
each variable with the cutoff between the two classes being the median variable value.
Correlation to
Vulnerability
Category Variable Median
- negative Age Percentage >65 years old 10.7%
Age Median Age 35.6
Wealth Median Home Value $308,788
Wealth Median Household Income $50,626
Wealth Per Capita Income $20,887
+ positive Age Percentage <20 years old 26.5%
Race/ Ethnicity Percentage Black Population 3.3%
Race/ Ethnicity Percentage Hispanic Population 43.7%
Race/ Ethnicity Percentage Minority Population 80.6%

The two designations shown in the first column of Table 2 are based on how a
variable is expected correlate with vulnerability. For example, minority population is
expected to have a positive correlation with vulnerability, so block groups with
percentages greater than 10.7% were classed as H. Conversely, median household income
is expected to be negatively correlated, so block groups with less than $50,626 median
household income were classed as H, even though they are the smaller values in that
dataset.

3.2.4 Spatial Joins and Expected Distributions
Census block groups were identified for every point event in the Red Cross and NFIRS
datasets through a spatial join. Once the block group codes were matched to the fire and
48
aid locations, the classes for each variable were also matched to each record. The result
was an L or H designation for each of the nine variables at every location.
To determine how reality deviates from a ceteris paribus expectation of fire
response/aid (i.e. independent of indicators such as wealth), expected numbers of
response/aid needed to be calculated. A common incorrect assumption of a ceteris
paribus scenario might be that 50% of fires occur in the L percentage minority block
groups, 50% occur in the H percentage minority block groups since the 6,422 CBGs were
divided into two groups based on the median CBG value for that variable. The more
accurate assumption would be based on the number of households in each of these classes
since households suffer fires, not block groups. In reality, 55.7% of households fall in the
L percentage minority block groups, suggesting that in a world without residential fire
bias, 55.7% of fires would occur there. The counts of households by block group were
thus used to calculate expected fire rates.
With “expected” and “actual” distributions for each variable, a statistical
comparison can be made to determine how likely the observed values in these two
datasets differ by chance. A chi-squared test was selected as the statistical test where the
null hypothesis is that the distributions are not different. The p-value for the chi-squared
tests offers a determination for the statistical significance of the result.

3.3 Results
The majority of fire and aid events could be geocoded with automated services by using
the addresses assigned to them in their respective databases, but a significant percentage
of those records had incomplete address attributes or improper formatting. The process of
fixing these addresses was time-consuming, with one of the more common problems
49
being the blanket assignment of “Los Angeles” as the city of record in a county that
includes 88 cities. Manual correction of addresses increased geolocated aid events from
90% to 96% (2,057 records to 2,214) and fire events from 84% to 89% (10,766 records to
11,397).
Most of the NFIRS data that were left unmatched came from the fire departments
of the City of Los Angeles (LAFD) and Los Angeles County (LACoFD). This is
unsurprising given the sheer volume of responses by these two departments; together,
they account for 78% of responses in the study time period. The LAFD was responsible
as the primary responder for 41% of events while the LACoFD was responsible for 37%.
The next most frequent responses were by the Long Beach Fire Department at 8% and
then the Glendale Fire Department at 2%. On average, LAFD handled 910 fires a year as
the primary responder and LACoFD handled 809.
Over the time period queried, a total of 30 fire departments in Los Angeles
County reported to NFIRS. The largest number of departments reporting in a single year
was 27 in 2008 while the fewest reporting was 22 in 2011. There are 88 cities in the
county, all of which either have their own fire department or have a partner agreement
with a nearby department (in most cases LACoFD). The total number of departments that
can report to NFIRS in the county is 55, though only 36 belong to a city or the county;
the remaining 19 are fire departments for private organizations such as aerospace
companies, state recreation areas, and universities. After corrections between the two
datasets, a total of 13,611 of 15,067 records were successfully located to a street address
(Figure 6).

50

Figure 6 Fire response locations mapped on top of median household income by census
tract. The map is zoomed in to focus on the more populous areas of the county, with 93%
of fire responses visible in the view extent.

A visual examination of the response locations overlaid on a thematic map of
median household income shows that there may be a higher incidence of fires in lower
income areas. Such a conclusion is difficult to draw from this view, however, with many
data points obstructing visibility of the underlying income theme. A map of Red Cross
aid events with fewer points makes its spatial pattern more apparent (Figure 7).
51

Figure 7 Red Cross aid locations mapped on top of median household income by census
tract. The map is zoomed in to focus on the more populous areas of the county, with 92%
of aid events visible in the view extent.

3.3.1 High and Low Counts
After summing the number of households, fires, and aid events that fell into the L and H
classes for each variable, the same general trend was observed for every demographic
indicator chosen: the majority of the county’s households, even if slight, were in the
block groups that were classed as having lower vulnerability according to that indicator
(Table 7). At the same time, the majority of both fire responses and Red Cross aid events
occurred in the block groups with higher vulnerability as represented by each indicator.
In the case of fires, for example, the smallest difference between percentages in H and L
52
block groups occurred for the percentage Hispanic variable at 8 percentage points (54%
in H, 46% in L) and the largest was for the percentage black variable at 18 percentage
points (59% in H, 41% in L). These differences are even larger for aid events, in all cases
exceeding the maximum observed for fires. The smallest difference was 35 percentage
points (both in percentage black and percentage older than 65, with 68.7% in H and
32.3% in L), nearly double the fire maximum difference of 18 percentage points. The
maximum difference occurred for the variable per capita income, where 75% of aid
happened in H block groups and 25% happened in L.

Table 7 A sample of representative variables with sums of population, households,
response events, and aid for each H or L class.
Variable Clas
s
Total
Population
Total
Household
s
Fire
Responses
Expected
Fire
Responses
Red Cross
Aid
Expected
Red Cross
Aid
Percent
Minority
H 5,234,902
(53%)
1,447,848
(44%)
6,175
(54%)
5,024
(44%)
1,629
(74%)
1,206
(54%)
L 4,669,439
(47%)
1,819,270
(56%)
5,162
(46%)
6,313
(56%)
585
(26%)
1,008
(46%)
Median
Age
H 5,451,493
(55%)
1,578,770
(48%)
6,424
(57%)
5,478
(48%)
1,615
(73%)
1,255
(57%)
L 4,452,848
(45%)
1,688,348
(52%)
4,913
(43%)
5,859
(52%)
599
(27%)
959
(43%)
Median
Househol
d Income
H 5,232,193
(53%)
1,607,969
(49%)
6,484
(57%)
5,580
(49%)
1,633
(74%)
1,266
(57%)
L 4,672,148
(47%)
1,659,149
(51%)
4,853
(43%)
5,757
(51%)
581
(26%)
9,48
(43%)
Total 9,904,341 3,267,118 11,337 11,337 2,214 2,214

A chi-squared test was performed comparing actual vs. expected fire responses
for all variables described in Table 6 where the null hypothesis represented that there was
53
no difference between the H and L groups. Every comparison resulted in the rejection of
the null hypothesis with p-values < 0.01, meaning that there was a statistically significant
difference between the H and L groups. In other words, using the ratio of households in
the H and L block groups for each variable to determine an expected distribution of the
11,337 fires, the actual division between H and L was found to be so different from the
expected division that it is highly unlikely they could have occurred by chance. These
results suggest that fire events tend to occur in areas with more minorities, where the
median age is younger, and median household income is lower.
A similar chi-squared test was performed comparing the frequency of Red Cross
aid events in H and L block groups for each variable with an expected frequency based
on the percentage of fire responses in those block group classes. For example, 54% of fire
responses occurred in block groups with a higher percentage minority population and
46% occurred in block groups with a lower percentage. These ratios were used to
calculate the expected Red Cross aid event counts for the tests. For all variables, the chi-
squared tests yielded p-values < 0.01, indicating that the actual distribution of aid events
was different from what would be expected.

3.3.2 The Intersection of Variables
Representative variables were chosen from each category to look at how the fire and aid
events are associated with a combination of variables. Median age was chosen for the age
category, median household income was chosen for wealth, and percentage minority was
chosen for race/ethnicity. A scheme was created to categorize census tracts by their
classes for two variables at a time, resulting in four classes: HH, HL, LH, and LL. For
example, the combination of median age and percentage minority has an HH class that
54
represents a higher median age and therefore higher age-based vulnerability coupled with
a larger fraction of minorities and therefore a higher race-based vulnerability, an HL class
that represents higher age vulnerability and lower race vulnerability, and so on (Figure 8).
Using the three aforementioned categories, three combinations were created (the first
combining age and race, another combining age and wealth, and the third combining
wealth and race).

Figure 8 A map depicting block groups by the combination of classes (high and low) for
percentage minority and median household income. The orange HL code represents block
groups with a high percentage of minorities and a low percentage of households with
low-income (i.e. households with higher income).

The chi-squared tests were performed on the actual count of fire responses in
these new combinative classes. For every combination, the distribution of fire counts
among the four classes were shown to be significantly different (p-value < 0.01) than
55
what would be expected (Table 8). Again, the expected counts were calculated by
considering the number of households in each class for each combination of variables.

Table 8 Example of comparing actual fire response events to what may be expected
when block groups are classed on the intersection of percentage minority and financial
vulnerability (i.e. low median household income).
Class Actual Fire
Response Count
Expected Count Explanation of Expected Count
Total fires (11,337) was multiplied by:
HH 5,069 3,849 34%; 1.1M of 3.3M households are in block
groups with a HIGHER percentage minorities
and a HIGHER financial vulnerability
HL 1,355 1,629 14%; 0.5M of 3.3M households are in block
groups with a HIGHER percentage minorities
and a LOWER financial vulnerability
LH 1,106 1,175 10%; 0.3M of 3.3M households are in block
groups with a LOWER percentage minorities and
a HIGHER financial vulnerability
LL 3,807 4,684 41%; 1.3M of 3.3M households are in block
groups with a LOWER percentage minorities and
a LOWER financial vulnerability
TOTAL 11,337 11,337

A final set of chi-squared tests were performed on the three combinations, this
time comparing actual Red Cross aid events to what would be expected if that aid was
provided in the same proportion to the classes as the 11,337 fire events (Table 9). Again,
the values were found to be different in all cases (p-values < 0.01). These results indicate
both a vulnerability-related disproportion in fire occurrence and a related, and therefore
unsurprising, disproportion in the distribution of Red Cross aid.

56
Table 9 Example of comparing actual Red Cross aid events to what may be expected
when block groups are classed on the intersection of percentage minority and financial
vulnerability (i.e. low median household income).
Class Actual Red
Cross Aid Count
Expected Count Explanation of Expected Count
Total aid (2,214) was multiplied by:
HH 1,386 1,018 46%; 5.2K of 11.3K fires are in block groups
with a HIGHER percentage minorities and a
HIGHER financial vulnerability
HL 229 236 11%; 1.2K of 11.3K fires are in block groups
with a HIGHER percentage minorities and a
LOWER financial vulnerability
LH 247 248 11%; 1.3K of 11.3K fires are in block groups
with a LOWER percentage minorities and a
HIGHER financial vulnerability
LL 352 711 32%; 3.6K of 11.3K fires are in block groups
with a LOWER percentage minorities and a
LOWER financial vulnerability
TOTAL 2,214 2,214

3.4 Discussion and Conclusions
The distributions of these datasets when organized by demographic variables indicated
that fire events in Los Angeles County were more likely to have occurred in block groups
that demonstrate elements of greater social vulnerability. Specifically, indicators of
wealth, age, and race were examined and found to be associated with a higher incidence
rate for fires. Moreover, these demographic characteristics are even more strongly related
to populations that receive aid from the American Red Cross following a fire.
The results surrounding the American Red Cross dataset are reaffirming; because
the Red Cross aspires to provide aid based on need, and need is related to the capacity for
recovery. The results surrounding the NFIRS dataset, however, may suggest that socially
vulnerable populations face greater risk of fire occurrence in an urban area. The dataset
will need to be put in proper context via the addition of other data sources in order to
57
perform more robust analyses. While the NFIRS database allows for the recording of
information such as ignition source or effectiveness of smoke alarms, these fields are
sparsely populated. Further study regarding the reasons for this skewed distribution of
fire occurrences may be aided by information regarding building quality and age, as
socially vulnerable populations are likely living in older buildings with higher potential
for fires. Smoking behavior and the presence or absence of fire prevention equipment
(e.g. smoke detectors and fire extinguishers) are also important variables that are likely
related to vulnerability, and such data would add value in understanding fire risk.
More research on the spatial clustering of these demographic variables is needed,
along with information on how population density plays a role in the distribution of fires
and aid. Identification of clusters would help provide geographic context for a more
comprehensive analysis of home fires. Results from such analysis could contribute to the
understanding of place vulnerability within the county.
Adoption of digital information infrastructures such as NFIRS has streamlined
data reporting, and the program’s long history makes for a relatively comprehensive
collection of records. In addition, recent open data initiatives have increased accessibility
and enabled the general public to make visualizations and maps from bulk tabular
exports. A problem still exists, however, in the quality of that information. Much of the
data processing involved in this study was devoted to “clean-up” and data sanitization.
Thus, even in well-established databases such as NFIRS, there can be a need to improve
the format in which data is distributed and the ease of data querying. The sanitization
performed in this study can help guide future researchers interested in NFIRS data, and
could be used to produce streamlined workflows to access commonly requested data.
58
Regardless of the progress left to be made in preparing the quality of these
datasets, the American Red Cross can use such existing information to better clarify the
places and kinds of aid to prepare. Many groups have been found to be affected by fires,
so diverse tactics for planning and preparedness have been previously recommended
(DiGuiseppi et al. 2000). However, these results showing a relationship between
demographic indicators and actual fire and aid events helps justify targeted outreach
programs. General reporting of such information can also be helpful for members of the
public to understand how they can be affected in a disaster and how people similar to
them have been affected in the past.
59
CHAPTER FOUR: CONDUCTING DISASTER DAMAGE ASSESSMENTS
WITH SPATIAL VIDEO, EXPERTS, AND CITIZENS

Publication Information
This chapter has been published in Applied Geography and can be cited as:
Lue E, Wilson J P, and Curtis A 2014 Conducting disaster damage assessments with
Spatial Video, experts, and citizens. Applied Geography 52: 46–54

4.1 Introduction
Damage assessment (DA) is an important part of the disaster recovery process. It
helps agencies and organizations provide assistance and relief to those most affected.
These assessments are common following large disasters which have caused property
damage to homes, such as earthquakes, tornadoes, and hurricanes. Most DA efforts will
score damage to individual homes with an ordinal series of classes such as “Unaffected”,
“Minor”, and “Destroyed”. Using these classes, organizations like the American Red
Cross and agencies like FEMA can create an informed action plan to provide relief to
individual clients and to a community as a whole.
DA is a distinct volunteer activity at the Red Cross, falling under the Information
& Planning (IP) group. As such, guidelines for performing DA are well established in
training programs and manuals. After a disaster, such as when a tornado occurs, a field
crew of Red Cross volunteers trained in DA will arrive on the scene within days to
perform on-the-ground field surveys. These volunteers can be service associates (SA),
supervisors (SV), or managers (MN) depending on their level of experience and training.
60
During a Red Cross DA response, Red Cross personnel are supposed to survey
homes only from their vehicles on the road. They are trained to not get out of their cars to
perform inspections. From the road, they determine details of the residence such as the
street address, dwelling type (e.g. single-family, mobile, or apartment) and number of
floors. They also estimate damage on the following scale: “Affected”, “Minor”, “Major”,
and “Destroyed” (or inaccessible if road conditions do not permit an assessment). All this
information is recorded on a paper data entry form known as a street sheet.
Spatial Video has been proposed as a method for DA with enriched data capture
(Curtis et al. 2007, Curtis and Fagan 2013). This method utilizes one or more GPS-
enabled cameras mounted to a car that drives through the neighborhoods affected by the
disaster. The camera captures video that can then be taken back to a computer
workstation and used to do a virtual DA, reviewing the spatially enabled imagery to
assess damage levels. Because the video is recorded from a car, the video’s content is
presumably the same as what a Red Cross ground surveyor would see.
Data collected via Spatial Video allows people to monitor recovery and better
understand how a neighborhood is making its way through the disaster cycle (Curtis et al.
2007). It provides a lasting digital record of the damage that was captured at a scale that
currently cannot be matched with available aerial imagery. If roads are accessible, the
data can also be collected in a timely manner after an event occurs. With the consent of
local authorities, an agency or organization such as the Red Cross could do a drive
through quickly and collect all the imagery needed for DA. This effort is already similar
to surveys done by the Red Cross; windshield surveys are currently done where
volunteers drive around and perform assessments by looking through their car windows.
61
Adding a video camera to this methodology only adds value and produces digital content.
Without having to stop and observe houses one-at-a-time, neighborhoods can be covered
in hours and then the work of damage assessment can be distributed to technicians in
front of computers. This would reduce the amount of time needed to perform the damage
assessment and reduce the number of field crews that need to be sent out to a disaster site,
saving both time and money. It can enable an assessment of individual homes at a later
time with only one data collection effort.
Of course, the technology does not have to replace existing methods and protocols
entirely, and the availability of these technologies does not imply that field crews will not
be needed. A ground survey and Spatial Video are not mutually exclusive. For example,
Spatial Video can aid in ground surveys and provide a digital record that can be used as a
reference. There is certainly value in having people in the field, building experience and
discovering damage that video or aerial imagery may miss due to obstructions between
the road and the homes or because of other problems that may diminish image quality.
This study explores the potential to use this technology to distribute disaster
assessment work and crowdsource the assessments. The inherent benefit of
crowdsourcing is that it can take a large task that may take a long time and break it into
tasks that are manageable by a large number of individuals working autonomously. An
online survey was conducted using imagery from a Spatial Video data collection to
determine the potential for the technology to support crowdsourced damage assessment.
The survey was distributed to DA managers, supervisors, and service associates at
various chapters of the American Red Cross. The survey was also distributed to
individuals associated with CrisisMappers, a mailing list for people interested in the
62
intersection between crises and geospatial technology. Through that list, all people over
the age of 18 were invited to take the survey, whether they had experience in DA or not.
The primary purpose of the survey is to look at inter-rater reliability (IRR). In
other words, how do damage scores assigned to the same houses compare across different
assessors? This knowledge will be important in building production-level tools to
crowdsource damage assessment, and is particularly important in regard to the
respondents’ varying experience levels with DA.
While assessing a home as “Unaffected” or “Destroyed” is relatively easy given
the superlative nature of those classes, discerning the difference between “Affected” and
“Minor” damaged homes and “Minor” and “Major” damaged homes is more difficult.
One of the key aims of this survey is to determine whether the ratings of a crowdsourced
group of assessors would converge on common damage scores. In other words, will there
be cases where the distribution of score counts is clearly bimodal or multi-modal? If so,
the multi-modality will need to be addressed in either assigning or interpreting a final
damage score.
Additionally, information was collected not only on damage scores but also on an
assessment of the quality of the Spatial Video images presented to the respondent. The
underlying question here concerns how feasible it is to use these images for DA. Do the
more experienced assessors see these images as being sufficient and valuable for an
accurate assessment?
Other questions include the topics of repeatability and whether bias gets
introduced when the assessor has an understanding of the spatial relationships between
homes. On the issue of repeatability, the main question is: Given the same pictures
63
provided more than once, how do people perform in terms of their ability to score the
same picture the same way twice? And when examining pictures where people have a
sense of the surrounding homes and their levels of damage, do respondents record less
variability in terms of damage scores and an assessment of the image quality?

4.2 Damage Assessment, Crowdsourcing, and the GeoWeb
Damage Assessment can be performed in a number of ways. Ground-based
surveys are commonly performed by field teams which assess homes on foot or in
vehicles. DA can also be accomplished using mail or telephone surveys (Brown et al.
2002). A more common method in recent years utilizes remotely sensed imagery to
identify tornado paths (Yuan et al. 2002, Jedlovec et al. 2006, Joyce et al. 2009) and
perform the DA (Barrington et al. 2011, McCarthy et al. 2008).
The American Red Cross performs assessments on the ground by sending
volunteers to drive through the affected neighborhoods. This effort can employ many
people and coordinating it can be both expensive and time-consuming. In addition, only
those few volunteers will be able to offer their perspectives on the assessments. The
resulting data may ultimately provide limited utility to other partner organizations since
there are no universally accepted standards for damage classification, and different
organizations may assess damage using their own guidelines. Capturing digital imagery
or video allows just a few people to provide a versatile dataset to many, giving others the
option to make their own assessments. The inclusion of GPS data for tracking collection
routes also aids in the logistics of organizing large assessments. The GPS tracks provide
information on what roads were covered and what roads are left to be explored, reducing
the possibility of different teams duplicating each other’s efforts (Curtis et al. 2006).
64
Spatial perspectives on damage assessment are useful to inform the placement and
distribution of resources and critical infrastructure. This perspective can also help answer
where risk is highest. One of the issues with remote sensing, especially automated
processes, is that mistakes have been made using this method due to the difficulty of
accurately determining the content of the imagery (Speheger et al. 2002). Using oblique
imagery after the Haiti earthquake, two operators agreed only 76% of the time (Gerke
and Kerle 2011). However, the use of multiple assessors allows for a much more robust
system for crisis response. Crowdsourcing after a disaster creates a huge workforce in a
short amount of time.
Crowdsourcing as a method for data collection after a disaster has gained traction
in recent years (Roche et al. 2013, Gao et al. 2011, Goodchild and Glennon 2010).
Through crowdsourcing, large numbers of people can contribute to a project through a
common workflow that centralizes and processes their user generated content. The four
general benefits of crowdsourcing are speed, cost, quality, and diversity (Alonso 2012).
While the first of the two benefits relate to the need for rapid information collection
following a disaster, the latter two relate to the reliability of the information and the law
of large numbers, where a repeated experiment will tend to converge toward an expected
outcome. Physical resources in disaster-affected areas are limited and deploying
resources to those areas can be expensive and take time. Crowdsourcing allows disaster
responders to distribute tasks and make them more manageable while having secondary
benefits such as spreading awareness of the situation on the ground.
The presence of spatial data, software, and tools has also grown sharply in recent
years as the GeoWeb has developed (Roche et al. 2013, Hall et al. 2010, Crampton 2009,
65
Goodchild 2009, Haklay et al. 2008). While the technical growth in this realm is clear, a
parallel cultural growth has also occurred in the democratization of map-making. These
activities are no longer exclusively the purview of GIS and cartographic professionals,
and this shift has been made possible by Web 2.0 technologies that offer availability,
interactivity, and customizability of spatial content. The ability to customize the content
to fit various needs is instrumental to the democratization of spatial information
production, resulting in map mashups, which blend and combine data from various
sources and formats, to serve any number of needs (Batty et al. 2010, Liu and Palen
2010). As offspring of Web 2.0, the GeoWeb and crowdsourcing promote and support
boundless opportunities that go hand-in-hand.
Perhaps the best known example of crowdsourcing spatial information post-
disaster is the response to the catastrophic 2010 earthquake in Haiti that claimed 230,000
lives (Heinzelman and Waters 2010, Yates and Paquette 2011, Zook et al. 2010). One of
the most visible crisis mapping efforts following the earthquake was the implementation
of an Ushahidi instance for the disaster event. Ushahidi is an open source platform that
collects information from SMS texts, RSS feeds, and social media (Okolloh 2009). It
allowed anyone to access the information, removing many restrictions from information
dissemination that may have been associated with individual organizations’ sharing
policies. It also enabled the collection of information from non-authoritative sources
(particularly boots-on-the-ground), which may prove to be more valuable than official
data streams. This, of course, is a cause of concern, as false or misleading reports may
arise, leading to the need for information verification (Heinzelman and Waters 2010).
66
Quality control is necessary to effectively and properly utilize crowdsourced information
(Alonso 2012).
Some crowdsourced damage assessments have utilized remotely sensed imagery,
such as the Virtual Disaster Viewer (VDV) created by ImageCat following the 2008
earthquake in Wenchuan, China, the GEO-CAN collaboration for the 2010 earthquake in
Haiti, and the collaboration between GEO-CAN and Tomnod, Inc. to build a web-based
viewer for the 2011 earthquake in Christchurch, New Zealand (Barrington et al. 2011).
These DA solutions support the participation of multiple users in the assessment process
using high resolution imagery taken with a bird’s eye view of the affected areas. These
images can be rapidly collected and processed, but do not provide perspective at ground
level.
Spatial Video allows for fine scale ground level imagery focusing on the streets
that have been damaged (Curtis et al. 2009, Mills et al. 2010, Curtis and Mills 2012). For
a tornado, the data collection follows the tornado’s path. One of the key benefits to this
method (similar to aerial imagery but better than most ground surveys) is that a digital
record will be produced that can be referred to later. For example, a tornado may damage
the roof of a home, and the homeowner may want to repair it immediately to move back
in. If there is later a dispute with a relief organization or insurance company over the
level of damage, the Spatial Video provides imagery to back up the homeowner’s claims.
Such photographic evidence is typically not collected when field surveys are conducted.
Another benefit is that the digital data can be assessed at multiple peoples’
computers rather than having to send field crews to a site. This can be accomplished by a
few select experts or through crowdsourcing. Either way, it allows for the quick creation
67
of time-sensitive data. Once collected, the video can also be shared with any relevant
parties. This is also important because assessments can differ between organizations in
terms of how damage is rated or what assessors look for, but the video will provide a
single source of data that can be used by multiple organizations.
If the video were to be collected immediately after the event, another virtue of the
imagery would be that residents of the affected homes could see those homes before they
may be allowed back into their neighborhoods. This could provide relief for those whose
homes were less damaged, or it might better prepare people to deal with the damage that
was inflicted to their homes. The time period between evacuation and return may
sometimes be short, but such timely information can be valuable during recovery,
especially since so many victims have nothing to do but wait.
Crowdsourcing allows organizations and agencies to benefit from the general
public’s increased interest in disaster relief following a catastrophic event. The American
Red Cross is an example of just one organization that receives an influx of calls after a
disaster. People call to offer monetary donations, in-kind donations, and their time as
volunteers. The Los Angeles office of the Red Cross has observed a surge in applications
for new volunteers soon after a large-scale disaster occurs, even if that disaster is on the
other side of the country.

4.3 Methods
This study utilized footage obtained from a Spatial Video data collection effort.
The footage was exported to still photographs and an online survey was built for
assessing a selection of these pictures. The results of the survey were then analyzed to
address the usefulness of crowdsourcing for damage assessment.
68

4.3.1 Photo Collection
The tornado event that led to the imagery collection for this project occurred on
Tuesday, April 3, 2012 in the Dallas Fort-Worth (DFW) area and all tornadoes were rated
EF 2 on the Enhanced Fujita scale. The tornadoes were individually not unusual for the
Dallas area and would have a recurrence interval of approximately one year, but the
simultaneous occurrence of multiple such tornadoes made the situation unique and
attracted national media attention. The data collection occurred four days after the event
so that the imagery captures a mixture of damaged homes and some that were already
being repaired. The majority of homes with roof damage already had tarps covering
them. Access to a few streets was limited due to police blockades, which can be
addressed in more comprehensive data collection efforts by working with local law
enforcement.
Data collection began at 0830 local time on Saturday, April 7 using a rental
vehicle mounted with three Contour GPS cameras, one on each of the rear side windows
and one on the front windshield. Each camera recorded video along with a GPS track
synchronized with the video playback. The rear windows were tinted, but they were
lowered for the collection and did not come between the side cameras and the homes.
Neighborhoods with the highest levels of reported damage were included in the survey.
The collection finished at 1630 local time, making for 8 hours of field collection.
Homes were visually inspected from a distance before the decision was made to
include a neighborhood in the collection. The presence of blue or black tarps on roofs
was a general indicator of damage. Due to time and information constraints, the coverage
69
of these neighborhoods was not comprehensive; the primary goal was to collect video for
homes that would be most useful in cross-method comparisons of damage assessment.

4.3.2 Online Survey
An online survey was produced for two audiences: DA experts at the American
Red Cross (experienced assessors) and the general public (inexperienced assessors). The
untrained eyes largely come from an announcement posted in the mailing list of
CrisisMappers (http://www.crisismappers.net), a group of people interested in crises and
the role that geospatial technology plays in responding to them. The survey was created
using the Qualtrics survey software. The survey presented 32 images of damaged homes,
which the respondent assigned as either an “Unaffected” class or as belonging to one of
four damage classes consistent with the American Red Cross terminology: “Affected”,
“Minor”, “Major”, or “Destroyed” (Table 10).
The results were stratified by the respondents' experience with damage
assessment efforts. Those with previous experience were classified as experts and
included both Red Cross volunteers and others from the CrisisMappers community who
had worked on similar efforts with other organizations. The inexperienced respondents
also came from both the Red Cross and CrisisMappers, including people who had been
involved in aspects of disaster relief other than damage assessment.

70
Table 10 Damage assessment classes as defined on American Red Cross street sheets.
These class definitions were reproduced in the damage survey.
Affected Minor Major Destroyed
Some shingles and/or
siding missing

Debris against or around
dwelling

Structure damage
considered to be
nuisance

Dwelling is livable
without repairs
Minor structural damage

Damage to small
sections of roof

Numerous broken
windows

Large portions of
roofing material and/or
siding missing

Penetration damage
where it is believed no
structural damage has
occurred
Large portions of roof
missing or debris
penetration

One or two walls
missing
Total collapse

Shifted on foundation

Not economically
feasible to repair

Each image was estimated to take 25 seconds to assess, for a total of
approximately 14 minutes. Six minutes are estimated for reading instructions and
answering questions about background, making for a total time of 20 minutes. The
images themselves were split into two groups: ordered images and random images.
Ordered images show homes in a sequence, where it is clear that the houses being
assessed are located next to one another (and thus likely to introduce proximity bias in
assessment). While the damage levels that tornadoes cause can vary greatly between
adjacent houses, there exists the potential for a damage assessor to assign a house a score
that is influenced by the score of the previous nearby house that was assessed. The
random images showed homes in a random sequence. The 16 ordered images were
presented first, followed by the 16 randomized images. Of the 16 random images, six
were repeats from the set of ordered images to test for discrepancies in re-scoring.

71
4.3.3 Photo Selection
The videos were recorded with GPS data at a rate of one latitude/longitude
coordinate pair per second. Other data collected by the GPS include altitude, speed,
heading, and time. A KML file was created directly from the videos and then converted
into an Esri shapefile. This process produced a driving track that could be symbolized to
show the direction of the camera along the track by taking vehicle heading and adding
90° for right-facing cameras and subtracting 90° for left-facing cameras.
The videos required minor post-processing. Field conditions sometimes led to a
slight rotation applied to the camera angle. While all the data was still captured, the
rotation of the imagery made assessment more difficult. Select files were modified with
video editing software that applied a corrective rotation to the video.
The videos were then parsed into individual screen captures at a rate of one frame
per second and then linked to the shapefiles of driving tracks. The pictures were linked
simply by their sequence in the original video; because the GPS device recorded
coordinates at the equivalent rate of screen capturing, one image was produced per point
in the track.
In total, approximately 36,000 seconds of good Spatial Video using the three
cameras (3.3 hours on average per camera), translating into the same number of photos
when taking one freeze frame per second. This number does not include video where the
GPS signal was lost or other minor technical errors occurred. To ensure that only direct
images of homes were used, video from the front-facing camera was excluded. Photos
were also excluded if there was no structure in the view of the picture, which happened
when recording took place on highways and where the data collection vehicle was parked
and motionless along its route. Photos where there was no single structure in the center of
72
the picture were also excluded so there would be no confusion about which structure was
the subject for damage assessment. After these exclusions, approximately 5,000 good
photos remained and were eligible for inclusion in the online survey. Faces of people
standing in front of houses were covered with a block box to protect privacy.
The images chosen for the survey were selected to represent all possible damage
levels with an emphasis on the middle classes where the distinction between two
successive classes is more difficult to determine. An “Unaffected” house and a
“Destroyed” house are less similar than houses with “Minor” and “Major” damage. A
street segment from the data collection with 16 consecutive houses was chosen to
represent the set of ordered pictures with at least one house in each damage class
(according to the author’s assessment of damages). The random set of pictures was
composed of only “Affected”, “Minor”, and “Major” damage levels (again, according to
the author’s assessment). For each of those levels, two images from the ordered set were
repeated (Table 11).

Table 11 Number of pictures used in the online survey belonging to different damage
scores as determined by the author.
Estimated Damage Score Ordered Set (16 total) Random Set (16 total)
(1) Unaffected 4 0
(2) Affected 3 5 (1 repeat, 3 new)
(3) Minor 6 6 (1 repeat, 4 new)
(4) Major 2 5 (1 repeat, 3 new)
(5) Destroyed 1 0

73
When selecting the neighborhood for the 16 ordered pictures, the first step was to
find an image of a “Destroyed” home. Then adjacent homes were examined to determine
the surrounding level of destruction and ensure diversity of damage classes (Figure 9).
The resulting set included four “Unaffected” homes, which were useful in setting
respondent expectations; knowing that there are “Unaffected” homes among the pictures
would help remove bias from an incorrect expectation that all homes would be damaged.

Figure 9 Map of the neighborhood used for the ordered picture set.

Aside from the repeated images in the random picture set, pictures were selected
by first classifying damage for approximately 100 random photos. Once those photos had
been placed into their damage classes, random photos were selected from each to fulfill
the quotas for the random picture set.

74
4.3.4 Comparing Respondent Groups and Inter-Rater Reliability (IRR)
After the survey period closed, respondents’ answers were categorized into
groups based on the experience levels of the respondents. Responses by the experienced
and inexperienced groups were compared for each picture using the Mann-Whitney-
Wilcoxon test (MWW), a non-parametric test where the null hypothesis is that two
populations are the same. This test is appropriate given the non-normal distribution of the
survey results.
Krippendorff’s α is a statistical test to determine IRR within an individual group
and was used on each of the survey respondent groups. It was chosen for its acceptance
of multiple observers (i.e. respondents) and missing data. The value of α represents the
percentage of the data that is coded better than it would be if the data were coded
randomly. Suggested guidelines for interpreting α are that data should be considered
reliable when α ≥ 0.800 and can be considered for drawing tentative conclusions between
0.667 and 0.800. However, interpretations of α values can vary depending on the datasets
used and their implications, and there is more leeway for accepting lower α values in the
social sciences than there is in the physical sciences (Krippendorff 2013).

4.4 Results
4.4.1 Respondent Summary
The survey received a total of 144 responses, including 36 people who attempted
the survey but did not complete it. While many of those 36 responses had usable photo
scoring data, they were excluded from the analysis because there was no way to
determine if some of these respondents represented people who took the survey again and
completed it later. Exclusion of their responses helps to ensure that there was no double-
75
counting of individuals who may have had to restart the survey due to time constraints or
technological problems.
The remaining 108 responses consisted of 23 members of the general public who
were inexperienced in disaster assessment, two who did not reveal any information on
prior DA experience, 13 who did have experience but were not affiliated with the Red
Cross, and 70 Red Cross volunteers and employees (Table 12). Of the Red Cross
respondents, eight did not indicate their position within the Disaster Assessment activity
and the remainder did, with this group comprising of 37 service associates, 13
supervisors, and 12 managers.
For most of the following analyses, respondents were arranged into several non-
exclusive groups, the largest being Group A (all 108 viable responses). Two more groups
– Group I (the 23 inexperienced respondents) and Group E-All (the 83 experienced) –
were compared with one another. Other groups include Group E-M (the 12 managers),
Group E-S (the 13 supervisors), Group E-SA (the 37 service associates), and Group E-O
(the 13 with experience outside the Red Cross). Group U (the two who did not indicate
any experience) and Group E-U (the eight with Red Cross experience but did not provide
DA experience information) were used only in analyses that ignored experience level.

76
Table 12 Number of respondents organized into groups and subgroups.
Group Description Group Symbol No. of Respondents
Red Cross Managers E-M 12
Red Cross Supervisors E-S 13
Red Cross Service Associates E-SA 37
Red Cross, unknown position E-U 8
Other Experience E-O 13
Total Experienced E-All 83
Unknown U 2
Inexperienced I 23
All Respondents A 108

The majority of respondents heard about the survey through various Red Cross
channels, both official (e.g. announcement on a national Disaster Assessment call) and
unofficial (e.g. Facebook groups for disaster assessment enthusiasts). Unsurprisingly, the
70 respondents who had indicated a Red Cross affiliation also indicated that they heard
about the survey through these channels. The remainder learned of the survey largely
through CrisisMappers (21 f, five via word-of-mouth, two via Facebook, five via Twitter,
and five via other channels, with some indicating they had learned about the survey from
multiple sources).

4.4.2 Damage Scores
The response data for both damage class and image quality assessments were
ordinal and were coded for analysis. Damage classes were assigned numeric codes from 1
(Unaffected) to 5 (Destroyed) and the same was done for image quality scores (1 – Poor,
77
5 – Excellent). Modes and medians were calculated as measures of central tendency. In
most cases, mode and median values were the same, although the results presented here
focus on modes since this method is not susceptible to skewing by a few outliers.
For Group A, only three pictures in 32 had different values for medians and
modes (Table 13). In only one picture did the range of values received span all five
damage classes (Picture 29). The range spanned four classes in 16 cases, three classes in
10 cases, and two classes in five cases. Of those five cases, four occurred between the
highest two damage classes and one occurred between the lowest two damage classes.
This makes sense given that pictures of “Unaffected” and “Destroyed” homes are more
likely to have one or two scores due to more visibly clear damage states. However, range
size did not necessarily correlate with the percentage value of the mode score. Of the four
pictures where the mode consisted of over 80% of responses (Pictures 6 – 9 and 32),
three had ranges with three damage classes (Pictures 6 – 8).

78
Table 13 Percentages of scores for each picture as assigned by Group A (n = 108). Cases
of “no response” are ignored for these calculations. Bold and shaded values designate the
mode while italicized scores designate the median. An asterisk (*) indicates a picture
where the median and mode were not the same.
Score Pic 1 Pic 2* Pic 3 Pic 4 Pic 5 Pic 6 Pic 7 Pic 8
1 12 8 6 90 81 94
2 51 8 1 56 38 8 15 3
3 36 45 9 34 50 2 4 3
4 1 46 25 1 6
5 2 65

Score Pic 9 Pic 10 Pic 11 Pic 12 Pic 13* Pic 14 Pic 15 Pic 16
1 88 17 4 21
2 13 57 43 24 10 10
3 26 51 52 42 58 1
4 2 4 47 30 72 23
5 1 2 27 77

Score Pic 17 Pic 18 Pic 19 Pic 20 Pic 21 Pic 22 Pic 23 Pic 24
1 13 9 4 1
2 16 32 60 49 6 23
3 64 1 58 26 42 66 56
4 17 20 43 10 1 24 20
5 83 56

Score Pic 25* Pic 26 Pic 27 Pic 28 Pic 29 Pic 30 Pic 31 Pic 32
1 17 3 2 3
2 36 8 59 17 39
3 46 1 63 26 65 50
4 1 57 27 22 4 16 8 80
5 42 2 78 8 20

Agreement in damage scores was similar between the inexperienced and
experienced groups. The mode of damage scores for each picture is shown in Figure 10 in
three series: as scored by all respondents, by the experienced respondents only, and by
the inexperienced respondents only. The mode of the experienced response disagreed
with the mode of the inexperienced response in three cases.
79

(a)

(b)

Figure 10 Mode of damage scores for the ordered (a) and random picture sets (b) as
scored by Groups A, E-All, and I.

Group E-All was expected to demonstrate more precision in its scoring as a group
than Group I given its prior experience with DA, but this trend was not observed.
Precision was measured by comparing the two groups in terms of how frequently the
mode score was chosen for each picture. For Picture 8, for example, 96% of Group E-All
chose the mode score whereas only 86% of Group I chose it, suggesting greater precision
in terms of scoring by Group E-All (Table 14). The two groups were within 5% points of
0
1
2
3
4
5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Damage Score
Picture
0
1
2
3
4
5
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Damage Score
Picture
80
each other in 11 cases. Of the remaining 21 cases, Group E-All chose its mode score
more frequently than Group I 11 times while the reverse was true in 10 cases. In this
sense there is no compelling evidence that Group E-All demonstrated any more certainty
than Group I. The same conclusion would be reached if the range size of damage classes
were the measure of precision. Between the two groups, the pictures’ range sizes were
nearly identical (six pictures where Group E-All’s range was 1 smaller than Group I, nine
where the reverse was true, and 17 where the range sizes were the same).

81
Table 14 Percentages of scores for each picture as assigned by Group E-All (left; n = 83)
and Group I (right; n = 23). Cases of “no response” are ignored for these calculations.
Bold and shaded values designate the mode while italics designate the median. An
asterisk (*) indicates a picture where the median and mode were not the same.
Group E-All Group I

P01 P02* P03 P04 P05 P06 P07 P08 P01 P02 P03 P04 P05 P06 P07 P08

S1 10 9 6 90 82 96 19 6 5 90 78 86

S2 52 7 1 54 38 10 16 3 43 9 61 33 11 5

S3 38 43 10 37 49 2 1 33 55 5 28 57 10 11 9

S4 48 25 6 5 36 27 6 5

S5 2 64 68

P09 P10 P11* P12 P13 P14 P15 P16 P09 P10 P11 P12 P13 P14 P15 P16

S1 87 18 5 20 89 14 20

S2 13 55 46 27 9 10 11 62 32 15 14 9

S3 27 47 50 48 60 24 64 60 18 55 5

S4 1 3 42 29 73 20 5 5 68 32 68 32

S5 1 1 28 80 5 27 68

P17 P18 P19 P20 P21 P22 P23 P24 P17 P18 P19 P20 P21 P22 P23 P24

S1 13 9 3 14 10 9 5

S2 15 37 63 49 6 22 14 14 48 48 5 24

S3 66 1 53 24 42 67 61 59 76 33 43 64 43

S4 15 19 37 11 24 18 27 27 68 10 5 23 29

S5 85 62 73 32

P25* P26 P27 P28 P29 P30 P31 P32 P25* P26 P27 P28 P29 P30 P31 P32

S1 18 4 1 3 15 5 5

S2 35 10 62 16 40 40 52 23 36

S3 45 1 63 24 68 51 45 62 33 55 45

S4 1 55 26 22 4 15 6 81 64 33 23 5 18 14 73

S5 43 1 78 6 19 36 5 77 10 27

The MWW statistical test was used to test whether or not Group E-All’s scores
differed from Group I’s scores. This test invokes the null hypothesis that two populations
82
are the same while the alternative hypothesis is that they are different. The test results
show that the null hypothesis could only be rejected in the case of Picture 19 (Table 15).
The home in this image was debated by the respondents as either having sustained
“Major” damage or having been completely “Destroyed”.

Table 15 The p-values from MWW tests between Group E-All and Group I’s damage
scores. Only Picture 19, in bold, had a p-value within a 95% confidence limit.
Picture p-value Picture p-value Picture p-value Picture p-value
P01 0.81 P09 0.83 P17 0.17 P25 1.00
P02 0.25 P10 1.00 P18 0.48 P26 0.61
P03 0.59 P11 0.07 P19 0.02 P27 0.15
P04 0.95 P12 0.45 P20 0.12 P28 0.92
P05 0.66 P13 0.13 P21 0.32 P29 0.20
P06 0.97 P14 0.57 P22 0.99 P30 0.61
P07 0.57 P15 0.75 P23 0.64 P31 0.70
P08 0.11 P16 0.26 P24 0.90 P32 0.39

While median damage scores are generally consistent between these two groups
just as the mode is, it should be noted that some scores by a respondent group were
sometimes split between two classes. For example, Picture 11 was close to evenly split
between the scores of “Affected” and “Minor” by Group E-All (Figure 11). This picture
demonstrates a case where using either the median or the mode as a measure of central
tendency is problematic: the median is 2 and the mode is 3, but neither value captures the
bimodal distribution of the data. This home’s ambiguous score is likely due to more
subtle signs of damage such as the tarp on the roof may. The boards over windows may
83
also cause confusion; broken windows are easy to recognize, but may not indicate any
structural damage to the house.

(a)

(b)
Figure 11 Picture 11 (a) and Group E-All's assessment of it (b), split between the
“Affected” and “Minor” damage scores.

In one case, the counts of damage classes chosen were evenly split across three
damage classes. Group E-O gave Picture 21 an equal number of damage scores across the
"Unaffected", "Affected", and "Minor" damage classes (Figure 12). Twelve people were
evenly split between the three scores while one person did not score it. Interestingly, this
picture is a repeat of Picture 10. When it was scored the first time, five people in the
group called it "Unaffected", three people called it "Affected", and four called it "Minor”
damage. As there appears to be no apparent damage to the house other than debris in the
front yard and a tarp on the roof, this disagreement is likely due to assessors’
interpretations of what a tarp meant for damage scoring.

4
34
35
1
0
9
0
10
20
30
40
1 2 3 4 5 No
answer
Respondents
Damage Score
84

(a)

(b)
Figure 12 Picture 21 (a) and Group E-O's assessment of it (b), evenly split between the
“Affected”, “Minor”, and “Major” damage scores.

In most cases, the two most common scores for a picture were sequential (i.e. the
most common scores for a single picture were 1 and 2 as opposed to 1 and 3). This
indicates that split decisions tended to be between one damage class and the next highest
or lowest class). The only cases where the two most common damage scores were not
adjacent were Group I’s assessment of Picture 12 (though the 2
nd
and 3
rd
largest counts
for damage scores were only one respondent apart) and Group E-O’s assessments of
Pictures 5, 10, 12, 22, and 25. It should be noted that Group E-O is the group with the
least amount of group cohesion, as it is composed of a variety of experienced damage
assessors, but with no specified common assessment framework such as the one used by
the American Red Cross.
The Krippendorff’s α tests were initially run for Groups A, E-All, E-M, E-S, E-
SA, E-O, and I three times each, once with all 32 pictures as subjects, once with only the
ordered set of 16 pictures, and once with the random set of 16 pictures (Table 16). The
results indicate that there is general inter-rater agreement in the data. Using
Krippendorff’s (2013) guidelines, the data for all groups and subject sets can be used at
4 4 4
0 0
1
0
1
2
3
4
5
1 2 3 4 5 No
answer
# of Respondents
Damage Score
85
least for drawing tentative conclusions, with the exception of the inexperienced group
rating the random picture set where α = 0.639. Each group of experienced assessors
demonstrated more agreement over all three sets than the inexperienced group, with the
American Red Cross DA supervisors showing the most agreement. Each group also
demonstrated more agreement in the set of ordered pictures than in the set of random
pictures.

Table 16 Krippendorff's α values for IRR and the damage scores assigned by different
respondent groups to multiple subject sets. Green values are reliable, black values are
acceptable for tentative conclusions, and red values should not be accepted as indicators
of agreement.
Group Total Raters All 32 Pictures 16 Ordered
Pictures
16 Random
Pictures
12 Ordered &
Damaged
Pictures
A 108 0.754 0.787 0.691 0.669
E-All 83 0.764 0.795 0.706 0.676
E-M 12 0.739 0.754 0.701 0.619
E-S 13 0.833 0.864 0.785 0.783
E-SA 37 0.770 0.812 0.703 0.672
E-O 13 0.757 0.774 0.701 0.695
I 23 0.712 0.751 0.639 0.632

Each group observed the highest α values for the 16 ordered pictures rather than
the 16 random pictures. At first, the likely explanation for this difference in IRR is that
there actually is some bias when the pictures are of houses near each other, leading to
more convergence toward agreement. However, this result is contrary to the hypothesis
that responses begin to converge after respondents get practice with assessing pictures; in
86
this case, IRR should be higher in the later picture set. The more likely explanation for
the higher α values is that the ordered picture set contained several homes that were
completely unaffected and were easy to evaluate. The random picture set only had images
of damaged homes. A fourth run of Krippendorff’s α was conducted on the ordered
picture set data with the four unaffected homes removed, with the results indicating
smaller α values (Table 16).

4.4.3 Image Quality
Similar to the case with the damage classes, the null hypothesis from the MWW
on image quality scores could not be rejected for any picture (Table 17). The conclusion
drawn from these tests is that the image quality scores given by the experienced and
inexperienced groups cannot be proven to be dissimilar. The ways that the groups rated
image quality were not necessarily different. This does not imply, however, that IRR is
strong in this group, as IRR measures the variation in multiple raters’ responses within a
single group for all pictures while MWW measures two groups’ responses for each
picture.
87
Table 17 The p-values from MWW tests between Group E-All and Group I’s image
quality scores resulted in no cases where the null hypothesis could be rejected.
Picture p-value Picture p-value Picture p-value Picture p-value
P01 0.57 P09 0.73 P17 0.12 P25 0.32
P02 0.20 P10 0.80 P18 0.52 P26 0.99
P03 0.98 P11 0.35 P19 0.35 P27 0.98
P04 0.12 P12 0.96 P20 0.66 P28 0.20
P05 0.46 P13 0.53 P22 0.35 P29 0.41
P06 0.70 P14 0.53 P23 0.44 P30 0.67
P07 0.72 P15 0.61 P24 0.70 P31 0.53
P08 0.77 P16 0.54 P25 0.72 P32 0.72

Even though the MWW tests suggest that Group E-All and Group I converged on
similar image quality scores for each picture, this does not mean that the respondents
showed agreement on image quality. Krippendorff’s α was also calculated for image
quality scores in the same manner as it was calculated for damage scores and the results
show no agreement within any group (Table 18). A simple way of illustrating the
meaning of the MWW and α results is to draw an analogy of two groups of people who
will average the same number if asked to score something (the result of MWW). But if
either group is examined for their members’ individual scores, no strong group agreement
will be found in those assessments (the result of α).

88
Table 18 Krippendorff's α values for IRR and the image quality scores assigned by
different respondent groups to multiple subject sets. All values are red, indicating that
they should not be accepted as indicators of agreement.
Group Total Raters All 32 Pictures 16 Ordered
Pictures
16 Random
Pictures
A 108 0.111 0.108 0.082
E-All 83 0.113 0.100 0.094
E-M 12 -0.006 -0.011 -0.033
E-S 13 0.078 0.040 0.108
E-SA 37 0.163 0.146 0.127
E-O 13 0.111 0.115 0.074
I 23 0.098 0.130 0.020

When comparing the image quality score distributions by each group, Group E-M
stood out because they assigned the lowest image quality scores to the pictures. This
could be due to the experience of DA managers, suggesting that this group felt that a
complete disaster assessment cannot be conducted with these pictures alone. Other
groups may have been more accepting of these images for DA.
The 32 pictures that each of the 108 raters viewed resulted in 3,456 “picture
responses”, and each picture response possesses a combination of an assigned damage
class and an image quality score. These combinations were tabulated for all picture
responses and the resulting table provides a compact overview of the distributions of
responses for the damage and quality variables as well as their intersection (Table 19).
The two most commonly chosen image quality scores were “Fair” and “Good”,
totaling 64.47% of all image ratings. Only 3.18% were rated as “Excellent” while 19.24%
of all pictures were rated as “Poor”. Of the images that were categorized as having
89
“Poor” image quality, 31.4% were not assigned a damage score. This makes sense, as
people are probably less likely to assign a damage score to a picture if they think the
picture is of low quality.

Table 19 Percentages of pictures with different combinations of damage scores (rows)
and image quality scores (columns) out of 3,456 total picture responses. One rater’s
response to one picture accounts for 0.03% in the table. Cells are colored on a gradient
from red to green corresponding with smallest to largest values (0.09% to 10.73%).
Poor Fair Good Very
Good
Excellent No
answer
Total
Unaffected 2.75 4.02 4.11 0.98 0.26 0.09 12.21
Affected 4.17 7.61 5.96 0.93 0.09 0.14 18.89
Minor 3.04 10.71 10.73 2.98 0.23 0.14 27.84
Major 2.49 5.56 7.12 2.31 0.46 0.09 18.03
Destroyed 0.75 1.97 5.67 3.82 1.71 0.12 14.03
No answer 6.05 0.81 0.20 0.14 0.43 1.36 9.00
Total 19.24 30.67 33.80 11.17 3.18 1.94 100.00

Better image quality is expected to correspond with a higher percentage of
pictures being rated, and this trend is observed from the “Poor” to “Very Good” pictures.
The trend diverges with the “Excellent” pictures, which is surprising. Of the images that
were categorized as having “Excellent” image quality, 13.6% were not rated. Still, of the
images rated “Fair” to “Excellent” in terms of image quality, only 2.0% were not rated.
The majority of “Excellent” quality pictures were expected to be of houses that
were either “Unaffected” or “Destroyed”. However, this was only true for the pictures
rated “Destroyed”. Of the images that were categorized as having “Excellent” image
90
quality, 53.6% were rated as “Destroyed”. For homes marked as “Destroyed”, there was a
trend of more such pictures for each better image quality class. The opposite trend was
seen with homes marked “Unaffected”.

4.4.4 Repeated Images
The repeated images were inserted into the survey to explore whether or not
respondents could reliably repeat their ratings. Exploring consistency in rating is
independent of a group’s ability to agree on its ratings. In this sense, if a rater looked at a
picture of a 4 and assigned it a 2 the first time and then a 4 the second time, that is a
worse result for consistency than assigning it incorrectly as a 2 both times.
The mode damage scores were the same within each pair of repeated images.
While these results at a group level may suggest that the group consistently rated the
pairs of pictures, an examination of how individuals did at repeatability was showed less
encouraging results. More often than not, a picture that was repeated within the survey
was given the same score, but there were still many who changed their minds the second
time around. Of the people who changed their minds, there was no particular trend in
whether that change had a specific direction. In most cases, that change was only up or
down by one damage class, and few cases showed a change in two damage classes or
more.
Contingency tables were calculated for each repeated picture, with the score
assigned the first time recorded as rows and the second score recorded as columns
(Figure 13). In each contingency table, the diagonal series of boxes from (1, 1) to (5, 5)
represent repeat damage scores. Boxes above the diagonal represent pictures that were
assessed the second time as being more damaged. Boxes below the diagonal represent
91
pictures that were assessed the second time as being less damaged. The repeat ratings
ranged from 60% (P05 and P31) to 84% (P16 and P28) and the unequal sums between
tables are due to missing data.

1 2 3 4 5
1 6 5 1
2 2 31 14
3 9 26
4 1
5

(a) P01 & P22
1 2 3 4 5
1 2 3 1
2 23 13 1
3 14 32 3
4 4 2
5

(b) P05 & P31
1 2 3 4 5
1 9 7
2 2 45 8
3 2 4 17 1
4
5

(c) P10 & P21
1 2 3 4 5
1 10 3 2
2 2 13 4
3 14 30
4 1 1
5

(d) P12 & P25
1 2 3 4 5
1
2
3 1
4 54 19
5 1 2 25

(e) P15 & P26
1 2 3 4 5
1
2
3
4 15 9
5 8 73

(f) P16 & P28
Figure 13 Contingency tables for each pair of repeated pictures as rated by Group A. The
rows in the tables represent scores the first time the picture appeared while the columns
represents scores the second time.

4.4.5 Confidence
A set of questions immediately following the damage assessment portion of the
survey dealt with the confidence that users felt when assigning damage scores and their
92
feelings regarding their ability to work with the images. Approximately 78% of
respondents in Group I rated their confidence with the imagery as “Fair” or better, with
17% indicating “Poor” or “Very Poor” confidence (Figure 14). In contrast, 75% of Group
E-All described their confidence as “Fair” or better, but a greater percentage of that group
fell into the “Good” confidence class. This likely reflects the experience levels of the
respondents and how that experience informs their DA decisions.
The group of inexperienced respondents displayed more agreement that the
images were sufficient for DA than did their experienced counterparts. This is
unsurprising given that on-the-ground DA experience has likely taught this group lessons
in assessment that relate to factors that cannot be seen in the imagery. The same can be
said of these groups’ opinions on how their on-the-ground assessments might compare to
this web-based experience; nearly half of the inexperienced respondents thought their
assessments would have been the same on the ground whereas just one third of the
experienced group felt the same way.
93

(a)

(b)

Figure 14 Confidence related responses for Group I and Group E-All, with general
confidence on the left (a) and assessments of image quality and likelihood of on-the-
ground reproducibility on the right (b).

4.4.6 Respondents’ Comments
The survey provided an option to include comments on the survey itself or on the
images that were used. The majority of feedback given in this section regarded the
concept of a 3-point view method of damage assessment. A total of 34 people (41% of all
experienced respondents) suggested that multiple points of view were needed for a single
house in order to determine damage: one point of view for the front of the house and two
for each of the adjacent sides. Nearly half of the experienced assessors mentioned it, and
there were surely some others who thought it but just did not leave a comment. The
inclusion of side pictures was considered during the survey design, but these pictures
were intentionally left out to simplify the questions and manage the length of time needed
4%
2%
13%
22%
65%
40%
9%
31%
4%
4% 4%
1%
I: Confidence E-All: Confidence
% of Respondents
V. Poor Poor
Fair Good
V. Good No Response
9% 7% 9%
5%
22%
29% 26%
25%
13%
24%
13%
34%
43%
39%
48%
33%
4%
1%
9%
1%
4%
2%
I: Images E-All:
Images
I: Same E-All:
Same
% of Respondents
S. Disagree Disagree
Neither Agree
S. Agree No Response
94
by respondents to provide answers. In hindsight, the inclusion of these additional views
may have reduced the number of respondents willing to sit through the survey, but it
would have likely added confidence to peoples’ assessments.
In a similar vein, 13 respondents commented that they felt they needed to see
more to make better assessments. Suggestions included the option to zoom into a picture
(one respondent), see images from a higher vantage point (e.g. a camera tower mounted
to a car; one respondent), or employ image processing to remove shadows (four
respondents). Five respondents recognized the importance of seeing the roof, and would
have liked an aerial image to complement the street-level image. Two respondents
commented that scores were difficult to determine where blue tarps concealed roof
damage, but this limitation would also be observed by someone at the damage site since
most levels of damage assessment would disallow the removal of protective tarps. Only
one respondent observed that these pictures did not give a person the same amount of
detail that he or she would get in person, but no explicit explanation for that observation
was provided.

4.5 Discussion and Conclusions
Images of houses that were clearly “Unaffected” or “Destroyed” were expected to
be rated with an image quality of “Excellent” and the moderately damaged houses were
expected to be given lower image quality scores. This was not observed in the results, but
may be due to respondents’ understanding of the image quality questions. The questions
specifically asked how good the image quality was for the purpose of determining a
damage score, but respondents may have interpreted it with shadows, sharpness, contrast,
and detail level as the questions’ main meaning.
95
Another outcome that was expected but not observed was that the variation in
scores would shrink as the survey progressed. The rationale is that the more experience
people had with the survey, the better they would become at rating pictures. One way this
could be tested is by examining the respondent population’s variation in scoring over the
course of the damage assessment. If people got better at assessing the pictures, the entire
group might start to trend to scoring with the same values (i.e. there would be less
variation in scoring). The results of this study, however, suggest this did not happen.
One of the most promising features of crowdsourcing is the ability to collect
accurate information by aggregating responses from a large number of people. This study
showed that similar results were achieved using inexperienced and experienced damage
assessors. This result does not suggest that experience adds little value to the
assessments, but rather that inexperienced assessors can be tapped for this work if needed
and the content they generate can be trusted as tentative or preliminary results.
One of the more noticeable challenges faced in this study was the fact that
definitive and/or objective damage classifications are hard to come by. Multimodal score
distributions present challenges, and this research demonstrated that such results are
possible. A potential correction for this problem may be to use arithmetic means as
measures of central tendency, despite the data being ordinal. In most cases, an average
will likely approximate a median or a mode, and averages that are caught between two
discrete scores can be quickly identified. A more thorough method of dealing with
multimodality is to incorporate thresholds for the distribution of scores for each picture,
where a flag is raised if the distribution shows no strict majority score.
96
Even the most experienced experts can disagree on how a house should be scored,
especially if a more refined damage scheme with more classes is utilized. For this reason,
no analysis was performed to learn how “right” or “wrong” any of the groups were in
their assessments; hence, there was no answer key or baseline to compare these responses
with. However, an organization that participates in damage classification that wishes to
employ a similar crowdsourcing platform may benefit from creating a tutorial showing
visual examples of the damage classes used. Such a tutorial would surely improve IRR.
Similarly, such a tutorial could be used for disaster assessment training and
testing. The results of this study do not imply that field crews are not necessary. Rather,
the results emphasize that such rapid information collection can be deemed tentatively
reliable. A thorough damage assessment might employ a number of techniques, such as
field crews who assess houses on the ground while capturing imagery and video which
can be referenced at a later time. Aerial imagery can also aid in validating information
and aiding damage classification.
Performing similar surveys on images collected from other tornado events would
further refine the definitions of each class. Images of houses affected by more extreme
tornadoes may show a greater variety of damage. Such data has been collected for the
EF4 2011 Tuscaloosa–Birmingham tornado, EF5 2011 Joplin tornado, and the EF5 2013
Moore tornado. Expanding this survey to include a larger set of images would improve
confidence in the determinations of damage score.
This survey offered just one possible presentation of Spatial Video. For the sake
of simplicity, the video aspect of the data was removed and still frames were used. Those
who found value in the still frames would surely find value in the video. The most
97
common negative feedback was the lack of a three-point view, and access to the source
video would abate that concern. The spatial component of the data adds possibility to the
way that the imagery to be assessed is visually presented to the assessor.
Further research in validating user generated content can employ datasets where
houses have already been assessed in the field. A crowdsourced assessment similar to the
one conducted in this study can then be performed to compare its results to the field
assessment. Of the respondents in this study who had prior DA experience, 30% felt that
their assessments would have differed if they had been in the field. Studies utilizing
existing field assessments can further explore such claims.

98
CHAPTER FIVE: SYNTHESIS AND CONCLUSIONS
While there are many clear roles for GIS to play in emergency management, the technical
and relatively recondite nature of traditional GIS still acts as a major impediment to
implementation for many managers and first responders who would benefit from using
geospatial tools. Chapter 2 explored some of those impediments, and a major conclusion
found that GIS adoption has been stifled due to the unavailability of either software,
hardware, or a GIS champion or technician. This is unsurprising, as all three need to exist
in concert for a GIS program to develop successfully (Obermeyer and Pinto 2007,
Tomlinson 2007, Croswell 2009). However, the start-up cost of software has been abated
through developments brought on by others, such as the creation of tools such as SitCell
or the distribution of GIS software licenses through NHQ. Further, the GIS technician
can now be virtually available through arrangements such as Virtual Disaster Assessment
Team (VDAT). Other lessons learned from early-adopting chapters such as standards and
workflows can simply help fill in details once a chapter begins spatially enabling their
data and analysis.
Chapter 3 showcased an example of GIS analysis at a local level using datasets
that were identified as highly important to the Red Cross in Chapter 2. This disaster
planning-based analysis is a likely early step in utilizing Red Cross GIS data after
spatially enabling it. It also represents the use of GIS for local chapters: understanding
their communities, as Red Cross chapters differ across the country in terms of their needs,
the disasters they respond to, and the people they serve. Though these differences are
significant, this particular analysis can be nationally applicable and can provide a strong
use case for GIS as more than just a map.
99
The crowdsourcing model as proposed in Chapter 4 leverages large numbers of
volunteers to rapidly create user generated content. The proposed technique for gathering
damage assessment information is easily relatable to VDAT and the human capital
mentioned in Chapter 2; in fact, several VDAT members actually partook in the study.
While an ideal Spatial Video damage assessment tool is intentionally technically simple
to use, the creation of the tool and spatial data collection require expertise that teams such
as VDAT can provide.
GIS technology is becoming more and more available to Red Cross chapters and a
unifying thread across the preceding chapters relating to increased technology adoption is
user friendliness and accessibility. A user friendly interface has previously been
identified in other research as a necessity for putting GIS into the hands of those working
in disasters (Gunes and Kovel 2000). The designs of websites and web-based tools are
critical for their use, especially when designing tools that are intended to be used by all
audiences. In particular, design issues surrounding older end-users can include
consideration of vision problems, literacy, cognition of web content, and the motor skills
needed to navigate computer applications with a mouse or keyboard (Becker 2004). Due
to the diversity of the Red Cross workforce which includes older volunteers, tools for
deployment in disaster operations must not be limited in their technically difficulty.
The most important product to come out of the Red Cross GIS Project in Los
Angeles is the web-based map that enables the visualization of Red Cross data, including
historical records on Red Cross disaster response and the placement of logistical supplies
across the Los Angeles Region (Figure 15). In addition to use during relief operations,
this map is often used as a story-telling tool for personnel in any activity, and it impacts
100
the way that disaster relief and the Red Cross are perceived. The map has been used to
convey the mission of the Red Cross to policy makers, disaster response officials, and
donors. Its importance lies in its usability and accessibility: it is easy to access from a
web browser and it requires no technical training to operate and understand. It was
created by GIS technicians, but it is intended for end-users with no technical savvy.

Figure 15 Screenshot of the Red Cross GIS Project website showing preparedness
education events layered over a choropleth map of disaster events (e.g. house fires).

Issues of accessibility have been addressed with concepts taken from Web 2.0,
many of which are relevant to spatial decision making (Rinner et al. 2008). These
concepts involve the progression from a world wide web where data is being read to one
where data is also created, providing for an interactive and collaborative experience. The
intersection of geography and Web 2.0 has been referred to as GIS 2.0 (Miller 2006,
101
Elwood 2009) or neogeography (Turner 2006, Hudson-Smith et al. 2009). Initiatives that
crowdsource content are enhanced by that intersection, and have been referred to as
Public Participation GIS (PPGIS; Ghose and Elwood 2003; Laituri 2003; Merrick 2003)
and Volunteered Geographic Information (VGI; Goodchild 2007a, 2007b).
While many of these terms differ in the emphases, audiences, and approaches they
describe, some central tenets underlie them all: geospatial technologies made easily
available to non-geographers who have the ability to participate in defining geographic
information. They all also suggest a move away or an evolution of traditional GIS. PPGIS
has been viewed in part as a response to some classic critiques of GIS (Elwood 2006).
VGI expands diversity and contradicts paradigms of consensus (Elwood 2008b). More
traditional spatial data infrastructures (SDI) are created by formal organizations and data
are distributed to passive users, which is fundamentally at odds with the "active
participant" element of VGI (Budhathoki et al. 2008).
The different names given to them correlate to the different characteristics of
each, such as VGI for the processes and relationships that produce geographic data, GIS
2.0 to highlight technology and the interactive Web 2.0, and neogeography for the idea of
a new perspective on a conventional discipline (Elwood 2009). An umbrella term for all
these may be the Geospatial Web or GeoWeb (Elwood 2008a, 2008b). Through the
GeoWeb, maps become easy. Through both PPGIS and VGI, the untechnical end-user
becomes an important content creator, similar to the survey takers discussed in Chapter 4.
A related concept is Ambient Geographic Information (AGI), which describes
data that is harvested from sources such as social media (Stefanidis et al. 2011, Crooks et
al. 2013, Rice et al. 2013). This information may still be user-generated content, but not
102
with the intention of contributing to a project or goal. Twitter is a common example of a
source of AGI, as tweets are often geotagged and therefore capable of being displayed or
analyzed spatially. Twitter users may not intend to have their content analyzed, but
tweets are public and may contain useful information, particularly when examined in
bulk with spatial patterns emerging. Such content could be extremely helpful following a
disaster, as the producers of this information may be communicating on-the-ground
information such as obstructed roads or collapsed buildings.
Given the open nature of the GeoWeb, security, privacy, and confidentiality have
become an important issue (Batty et al. 2010). These concerns arise shortly after initial
introduction of GIS technology. Other needs include better data, models, and
infrastructure for GIS to prepare for disaster rather than make GIS just an ad hoc tool
cobbled together when disaster strikes (Cutter 2003). To this end, standards in software
and hardware have long been identified as necessary (e.g. Ilmavirta 1995). Even
standards regarding issues that superficially seem insignificant such as symbolization or
iconography can greatly help an organization have a thoroughly prepared geographic
system for disaster response (Robinson et al. 2011). Community-based disaster
preparedness, for example, has been described as requiring appropriate information
technology for success (Troy et al. 2008).
While the Red Cross GIS Project in the Los Angeles Region has been underway
for five years and standard operating procedures are in place, there are still elements of
the project that are evolving, particularly in how it (and VDAT) can effectively expand to
other chapters. Future research in this field can explore exactly how these technologies
and guiding documents can be disseminated for adoption. Part of the answer to that
103
question lies in more explicitly defining the end goals of a potentially adopting chapter or
organization. As mentioned in Chapter 2, using GIS can simply be an exploratory
exercise for some. Exemplars of goals informed by current disaster research can provide
the insight needed to decide how much GIS is needed (e.g. how big a team or how much
money to invest in hardware, software, and training) and what kind (e.g. points on a map,
field collection, complex spatial analysis). Much of what determines chapter goals
depends on current capabilities and deficiencies, which can be studied in more depth.
Analyses similar to those conducted in Chapter 2 can be applied to datasets such
as shelter locations, which is rich and geographically comprehensive for the country
because it is centralized in a national database. When more Red Cross data becomes
spatially enabled, the potential for future research in Red Cross operations grows. For
example, volunteer datasets have previously been sparsely populated and central lists
were often kept on local office computers. New initiatives to centralize such data could
enable research into the activities of volunteers, who make up the majority of the Red
Cross workforce. More can be done to better understand active volunteers as well as
those who simply sign up to volunteer but never return (usually after a major disaster that
receives news coverage). Initial analysis of volunteer data in Los Angeles indicates that
the volunteers in the region generally come from the communities that experience
relatively few disasters such as single-family fires. This also means that there are
relatively few volunteers in communities that have a high rate of receiving Red Cross aid.
Much more research can also be done in the field crowdsourced user generated
content. As the field grows and technological resources such as Ushahidi become more
commonplace than not following disaster events, the need to research the content grows.
104
This particular field is trending toward a model of citizen scientists who look to provide
some measure of aid from their computers, and the quality of those contributions needs to
be better understood to maximize how that information is utilized. As the field becomes
saturated with applications with good intentions, there may be a tipping point where good
applications become lost in a sea of useless ones. This will be particularly true for data
that may come from sources such as social media, where misinformation can be easily
ingested along with the valuable information.
Spatial Video data has provided valuable post-disaster imagery for damage
assessment, but much attention is being focused at present on unmanned aerial vehicles
(UAVs), or drones. These devices are decreasing in price and are widely available for
purchase by the general public. They are an inexpensive solution to rapidly creating high-
resolution aerial images and they require little training to operate. However, much debate
currently exists over how these vehicles should be operated with respect to privacy
protection. No-fly zones targeted specifically at UAVs have been established. Future
research regarding this data collection technique will be needed, as its use in disaster
response will likely increase in frequency.
The age of the digital humanitarian responding to disasters has been in progress
for years, but much opportunity still exists to improve understanding of how aid can be
best provided. Geographic information will remain a vital component of aid in the digital
space, as location provides context for safety, danger, and relief en route. The American
Red Cross continues to carve out its own place in crisis mapping, and the lessons learned
from this process can help inform local chapters, other organizations, and the budding
field of digital humanitarianism in general.
105
REFERENCES
Adger W N 2006 Vulnerability. Global Environmental Change 16: 268–281
Adger W N and Kelly P M 1999 Social vulnerability to climate change and the
architecture of entitlements. Mitigation and Adaptation Strategies for Global
Change 4: 253–266
Ahrens M, Stewart S, and Cooke P L 2003 Fire data collection and databases. In A E
Cote (ed) Organizing For Fire and Rescue Services. Sudbury, MA, Jones & Bartlett
Learning: 35–52
Alonso O 2012 Implementing crowdsourcing-based relevance experimentation: An
industrial perspective. Information Retrieval 16: 101–120
Barrington L, Ghosh S, Greene M, Har-Noy S, Berger J, Gill S, Lin A Y-M, and Huyck
C 2011 Crowdsourcing earthquake damage assessment using remote sensing
imagery. Annals of Geophysics 54: 680–687
Batty M, Hudson-Smith A, Milton R, and Crooks A 2010 Map mashups, Web 2.0 and the
GIS revolution. Annals of GIS 16: 1–13
Becker S A 2004 A study of web usability for older adults seeking online health
resources. ACM Transactions on Computer-Human Interaction 11: 387–406
Blaikie P, Cannon T, Davis I, and Wisner B 1994 At Risk: Natural Hazards, People’s
Vulnerability and Disasters (1st Edition). London, Routledge
106
Borden K A, Schmidtlein M C, Emrich C T, Piegorsch W W, and Cutter S L 2007
Vulnerability of U.S. cities to environmental hazards. Journal of Homeland Security
and Emergency Management 4: 22
Brenkert A L and Malone E L 2005 Modeling vulnerability and resilience to climate
change: A case study of India and Indian States. Climatic Change 72: 57–102
Brown S, Archer P, Kruger E, and Mallonee S 2002 Tornado-related deaths and injuries
in Oklahoma due to the 3 May 1999 tornadoes. Weather and Forecasting 17: 343–
353
Budhathoki N R, Bruce B, and Nedovic-Budic Z 2008 Reconceptualizing the role of the
user of spatial data infrastructure. GeoJournal 72: 149–160
Budić Z D and Godschalk D R 1994 Implementation and management effectiveness in
adoption of GIS technology in local governments. Computers, Environment and
Urban Systems 18: 285–304
Campbell H 1996 Theoretical perspectives of the diffusion of GIS. In I Masser, H
Campbell, and M Craglia (eds) GIS Diffusion: The Adoption and Use of
Geographical Information Systems in Local Government in Europe. London, Taylor
& Francis: 23–45
Chakraborty J, Tobin G A, and Montz B E 2005 Population evacuation: Assessing spatial
variability in geophysical risk and social vulnerability to natural hazards. Natural
Hazards Review 6: 23–33
107
Chan T O and Williamson I P 1999 A model of the decision process for GIS adoption
and diffusion in a government environment. URISA Journal 11: 7–16
Collins T and Bolin B 2007 Characterizing vulnerability to water scarcity: The case of a
groundwater-dependent, rapidly urbanizing region. Environmental Hazards 7: 399–
418
Collins T W 2005 Households, forests, and fire hazard vulnerability in the American
West: A case study of a California community. Global Environmental Change Part
B: Environmental Hazards 6: 23–37
Cova T J 1999 GIS in emergency management. In P A Longley, M F Goodchild, D J
Maguire, and D W Rhind (eds) Geographical Information Systems: Principles,
Techniques, Applications, and Management. New York, John Wiley & Sons: 845–
858
Crampton J W 2009 Cartography: Maps 2.0. Progress in Human Geography 33: 91–100
Crooks A, Croitoru A, Stefanidis A, and Radzikowski J 2013 #Earthquake: Twitter as a
Distributed Sensor System. Transactions in GIS 17: 124–147
Croswell P L 2009 The GIS Management Handbook. Frankfort, Kentucky, Kessey Dewitt
Publications
Curtis A 2008 From healthy start to Hurricane Katrina: Using GIS to eliminate disparities
in perinatal health. Statistics in Medicine 27: 3984–3997
108
Curtis A and Fagan W F 2013 Capturing damage assessment with a spatial video: An
example of a building and street-scale analysis of tornado-related mortality in Joplin,
Missouri, 2011. Annals of the Association of American Geographers 103: 1522–
1538
Curtis A, Mills J, Blackburn J K, Pine J C, and Kennedy B 2006 Louisiana State
University geographic information system support of Hurricane Katrina recovery
operations. International Journal of Mass Emergencies and Disasters 24: 203–221
Curtis A and Mills J W 2012 Spatial video data collection in a post-disaster landscape:
The Tuscaloosa Tornado of April 27th 2011. Applied Geography 32: 393–400
Curtis A, Mills J W, Kennedy B, Fotheringham S, and McCarthy T 2007 Understanding
the geography of post-traumatic stress: An academic justification for using a spatial
video acquisition system in the response to Hurricane Katrina. Journal of
Contingencies and Crisis Management 15: 208–219
Curtis A, Mills J W, McCarthy T, Fotheringham A S, and Fagan W F 2009 Space and
Time Changes in Neighborhood Recovery after a Disaster Using a Spatial Video
Acquisition System. In P Showalter and Y Lu (eds) Geospatial Technologies in
Urban Hazard and Disaster Analysis. Springer: 373–392
Cutter S L 2003 GI Science, disasters, and emergency management. Transactions in GIS
7: 439– 445
109
Cutter S L 1996 Vulnerability to environmental hazards. Progress in Human Geography
20: 529–539
Cutter S L, Boruff B J, and Shirley W L 2003 Social vulnerability to environmental
hazards. Social Science Quarterly 84: 242–261
Cutter S L and Finch C 2008 Temporal and spatial changes in social vulnerability to
natural hazards. Proceedings of the National Academy of Sciences of the United
States of America 105: 2301–2306
De Silva F N and Eglese R W 2000 Integrating simulation modelling and GIS: Spatial
decision support systems for evacuation planning. Journal of the Operational
Research Society 51: 423–430
DiGuiseppi C, Edwards P, Godward C, Roberts I, and Wade A 2000 Urban residential
fire and flame injuries: A population based study. Injury Prevention 6: 250–4
Elwood S 2006 Critical issues in participatory GIS: Deconstructions, reconstructions, and
new research directions. Transactions in GIS 10: 693–708
Elwood S 2009 Geographic Information Science: Emerging research on the societal
implications of the geospatial web. Progress in Human Geography 34: 349–357
Elwood S 2008a Geographic Information Science: New geovisualization technologies -
emerging questions and linkages with GIScience research. Progress in Human
Geography 33: 256–263
110
Elwood S 2008b Volunteered Geographic Information: Future research directions
motivated by critical, participatory, and feminist GIS. GeoJournal 72: 173–183
Flanagan B E, Gregory E W, Hallisey E J, Heitgerd J L, and Lewis B 2011 A social
vulnerability index for disaster management. Journal of Homeland Security and
Emergency Management 8: 22
Folz D, Shults C, Meyers M, and Adams-O’Brien F 2011 An Analysis of Civilian
Residential Fire Deaths in Tennessee, 2002-2010. Nashville, TN
Fothergill A and Peek L A 2004 Poverty and disasters in the United States: a review of
recent sociological findings. Natural Hazards 32: 89–110
Füssel H-M 2007 Vulnerability: A generally applicable conceptual framework for climate
change research. Global Environmental Change 17: 155–167
Gaither C, Poudyal N, Goodrick S, Bowker J, Malone S, and Gan J 2011 Wildland fire
risk and social vulnerability in the southeastern United States: An exploratory spatial
data analysis approach. Forest Policy and Economics 13: 24–36
Gao H, Barbier G, and Goolsby R 2011 Harnessing the crowdsourcing power of social
media for disaster relief. IEEE Intelligent Systems 26: 10–14
Gerke M and Kerle N 2011 Automatic structural seismic damage assessment with
airborne oblique Pictometry imagery. Photogrammetric Engineering and Remote
Sensing 77: 885–898
111
Ghose R and Elwood S 2003 Public participation GIS and local political context:
Propositions and research directions. URISA Journal 15: 17–24
Godschalk D R 1991 Disaster mitigation and hazard management. In T Drabek and G
Hoetmer (eds) Emergency Management: Principles and Practice for Local
Government. Washington DC, International City Management Association: 131–160
Goodchild M 2009 NeoGeography and the nature of geographic expertise. Journal of
Location Based Services 3: 82–96
Goodchild M F 2007a Citizens as sensors: The world of volunteered geography.
GeoJournal 69: 211–221
Goodchild M F 2007b Citizens as voluntary sensors: Spatial data infrastructure in the
World of Web 2.0. International Journal of Spatial Data Infrastructures Research
2: 24–32
Goodchild M F and Glennon J A 2010 Crowdsourcing geographic information for
disaster response: A research frontier. International Journal of Digital Earth 3: 231–
241
Greenwald M J 2000 Beyond city limits: The multi-jurisdictional applications of GIS.
URISA Journal 12: 31–43
Greiving S, Fleischhauer M, and Lückenkötter J 2006 A methodology for an integrated
risk assessment of spatially relevant hazards. Journal of Environmental Planning
and Management 49: 1–19
112
Gunes A E and Kovel J P 2000 Using GIS in emergency management operations.
Journal of Urban Planning and Development 126: 136–149
Haklay M, Singleton A, and Parker C 2008 Web mapping 2.0: The neogeography of the
GeoWeb. Geography Compass 2: 2011–2039
Hall G B, Chipeniuk R, Feick R D, Leahy M G, and Deparday V 2010 Community-based
production of geographic information using open source software and Web 2.0.
International Journal of Geographical Information Science 24: 761–781
Heinzelman J and Waters C 2010 Crowdsourcing Crisis Information in Disaster-affected
Haiti. Washington, DC
Holub M and Fuchs S 2009 Mitigating mountain hazards in Austria - Legislation, risk
transfer, and awareness building. Natural Hazards and Earth System Science 9:
523–537
Hudson-Smith A, Crooks A, Gibin M, Milton R, and Batty M 2009 NeoGeography and
Web 2.0: Concepts, tools and applications. Journal of Location Based Services 3:
118–145
Ilmavirta A 1995 The use of GIS-system in catastrophe and emergency management in
Finnish municipalities. Computers, Environment and Urban Systems 19: 171–178
Jedlovec G J, Nair U, and Haines S L 2006 Detection of storm damage tracks with EOS
data. Weather and Forecasting 21: 249–267
113
Jennings C 2013 Social and economic characteristics as determinants of residential fire
risk in urban neighborhoods: A review of the literature. Fire Safety Journal 62: 13–
19
Johnson D, Zagorecki A, Gelman J M, and Comfort L K 2011 Improved situational
awareness in emergency management through automated data analysis and
modeling. Journal of Homeland Security and Emergency Management 8: 17
Joyce K E, Belliss S E, Samsonov S V., McNeill S J, and Glassey P J 2009 A review of
the status of satellite remote sensing and image processing techniques for mapping
natural hazards and disasters. Progress in Physical Geography 33: 183–207
Kevany M J 2003 GIS in the World Trade Center attack—trial by fire. Computers,
Environment and Urban Systems 27: 571–583
Kiltz L and Smith R 2011 Experimenting with GIS in doing damage assessments: A trial
run at Disaster City. Journal of Homeland Security and Emergency Management 8:
21
King D and MacGregor C 2000 Using social indicators to measure community
vulnerability to natural hazards. Australian Journal of Emergency Management 15:
52–57
Krippendorff K 2013 Content Analysis: An Introduction to Its Methodology (3rd
Edition). Thousand Oaks, CA, SAGE Publications
114
Laituri M 2003 The issue of access: An assessment guide for evaluating public
participation geographic information science case studies. URISA Journal 15: 25–32
Liu S B and Palen L 2010 The new cartographers: Crisis map mashups and the
emergence of neogeographic practice. Cartography and Geographic Information
Science 37: 69–90
McCarthy T, Farrell R, Curtis A, and Fotheringham A S 2008 Integrated remotely sensed
datasets for disaster management. In U Michel, D L Civco, M Ehlers, and H J
Kaufmann (eds) SPIE 7110, Remote Sensing for Environmental Monitoring, GIS
Applications, and Geology VIII.
Merrick M 2003 Reflections on PPGIS: A view from the trenches. URISA Journal 15:
33–39
Miller C C 2006 A beast in the field: The Google Maps mashup as GIS/2. Cartographica
41: 187–199
Mills J W 2008 Understanding disaster: GIScience contributions in the ongoing recovery
from Katrina. Transactions in GIS 12: 1–4
Mills J W, Curtis A, Kennedy B, Kennedy S W, and Edwards J D 2010 Geospatial video
for field data collection. Applied Geography 30: 533–547
Mills J W, Curtis A, Pine J C, Kennedy B, Jones F, Ramani R, and Bausch D 2008 The
clearinghouse concept: A model for geospatial data centralization and dissemination
in a disaster. Disasters 32: 467–479
115
Montoya L 2003 Geo-data acquisition through mobile GIS and digital video: An urban
disaster management perspective. Environmental Modelling & Software 18: 869–
876
Moore G C 1993 Implications from MIS research for the study of GIS diffusion: Some
initial evidence. In I Masser and H J Onsrud (eds) Diffusion and Use of Geographic
Information Technologies. Dordrecht, Kluwer Academic Publishers: 77–94
Morrow B H 1999 Identifying and mapping community vulnerability. Disasters 23: 1–18
Nedović-Budić Z and Godschalk D R 1996 Human factors in adoption of Geographic
Information Systems: A local government case study. Public Administration Review
56: 554–567
Obermeyer N J and Pinto J K 2007 Managing Geographic Information Systems (2nd
Edition). New York, Guilford Press
Okolloh O 2009 Ushahidi, or’testimony': Web 2.0 tools for crowdsourcing crisis
information. Participatory Learning and Action 59: 65–70
Onsrud H J and Pinto J K 1991 Diffusion of geographic information innovations.
International Journal of Geographical Information Systems 5: 447–467
Onsrud H and Pinto J 1993 Evaluating correlates of GIS adoption success and the
decision process of GIS acquisition. URISA Journal 5: 18–39
116
Pelling M 2007 Learning from others: The scope and challenges for participatory disaster
risk assessment. Disasters 31: 373–385
Pidd M, de Silva F N, and Eglese R W 1996 A simulation model for emergency
evacuation. European Journal of Operational Research 2217: 413–419
Rice M T, Jacobson R D, Caldwell D R, McDermott S D, Paez F I, Aburizaiza A O,
Curtin K M, Stefanidis A, and Qin H 2013 Crowdsourcing techniques for
augmenting traditional accessibility maps with transitory obstacle information.
Cartography and Geographic Information Science 40: 210–219
Rinner C, Kesler C, and Andrulis S 2008 The use of Web 2.0 concepts to support
deliberation in spatial decision-making. Computers, Environment and Urban
Systems 32: 386–395
Robinson A C, Roth R E, and MacEachren A M 2011 Understanding user needs for map
symbol standards in emergency management. Journal of Homeland Security and
Emergency Management 8: 14
Roche S, Propeck-Zimmermann E, and Mericskay B 2013 GeoWeb and crisis
management: Issues and perspectives of volunteered geographic information.
GeoJournal 78: 21–40
Rogers E M 1976 New product adoption and diffusion. Journal of Consumer Research 2:
290–301
117
Rogers E M 1993 The diffusion of innovations model. In I Masser and H J Onsrud (eds)
Diffusion and Use of Geographic Information Technologies. Dordrecht, Kluwer
Academic Publishers: 9–26
Runyan C, Bangdiwala S, Linzer M, Sacks J, and Butts J 1992 Risk factors for fatal
residential fires. New England Journal of Medicine 327: 859–863
Rygel L, O’Sullivan D, and Yarnal B 2006 A method for constructing a social
vulnerability index: An application to hurricane storm surges in a developed
country. Mitigation and Adaptation Strategies for Global Change 11: 741–764
Shahabi K and Wilson J P 2014 CASPER: Intelligent capacity-aware evacuation routing.
Computers, Environment and Urban Systems 46: 12–24
Sieber R E 2000 GIS implementation in the grassroots. URISA Journal 12: 15–29
Speheger D A, Doswell III C A, and Stumpf G J 2002 The tornadoes of 3 May 1999:
Event verification in central Oklahoma and related issues. Weather and Forecasting
17: 362–381
Stefanidis A, Crooks A, and Radzikowski J 2011 Harvesting ambient geospatial
information from social media feeds. GeoJournal 78: 319–338
Tierney K J, Lindell M K, and Perry R W 2001 Facing the Unexpected: Disaster
Preparedness and Response in the United States. Washington, DC, Joseph Henry
Press
118
Tomlinson R F 2007 Thinking About GIS: Geographic Information System Planning for
Managers (5th Edition). Redlands, CA, ESRI Press
Turner A J 2006 Introduction to Neogeography. Sebastopol, CA, O’Reilly Media
Turner II B L, Kasperson R E, Matson P A, McCarthy J J, Corell R W, Christensen L,
Eckley N, Kasperson J X, Luers A, Martello M L, Polsky C, Pulsipher A, and
Schiller A 2003 A framework for vulnerability analysis in sustainability science.
Proceedings of the National Academy of Sciences of the United States of America
100: 8074–8079
Vaughan E 1995 The significance of socioeconomic and ethnic diversity for the risk
communication process. Risk Analysis 15: 169–180
Warda L, Tenenbein M, and Moffatt M 1999 House fire injury prevention update: Part I.
A review of risk factors for fatal and non-fatal house fire injury. Injury Prevention 5:
145–50
Weichselgartner J 2001 Disaster mitigation: The concept of vulnerability revisited.
Disaster Prevention and Management 10: 85–95
Yates D and Paquette S 2011 Emergency knowledge management and social media
technologies: A case study of the 2010 Haitian earthquake. International Journal of
Information Management 31: 6–13
119
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 17: 382–398
Zerger A 2002 Examining GIS decision utility for natural hazard risk modelling.
Environmental Modelling & Software 17: 287–294
Zhou H, Wang J, Wan J, and Jia H 2009 Resilience to natural hazards: A geographic
perspective. Natural Hazards 53: 21–41
Zook M, Graham M, Shelton T, and Gorman S 2010 Volunteered geographic information
and crowdsourcing disaster relief: A case study of the Haitian earthquake. World
Medical and Health Policy 2: 7–33

120
APPENDIX A: CHAPTER TWO INTERVIEW GUIDING DOCUMENT

AMERICAN RED CROSS MAPPERS INTERVIEWS

Purpose
The purpose of this study is to explore how local chapters of the American Red Cross
have engaged in spatial visualizations and analyses through the use of maps. A handful of
chapters have implemented map making technology, but most do not. The majority of the
chapters that do map began their efforts independent of one another and of their National
Headquarters (NHQ). Structured interviews with the map leads of various chapters will
offer insight into the experiences of those in the Red Cross who have attempted
geospatial technology adoption. The opportunities and obstacles encountered in the
adoption of this technology can help determine how other chapters can engage
themselves. Additionally, these interviews will determine what ideas brought on the
venture of mapping and what the individual mapping needs were. By its nature,
geospatial technology use is a horizontally integrated activity that has the potential to
touch every part of an organization, and this study will determine which of those parts
were most prominently in the minds of the initiators.

Research Questions
Questions Importance (red is for Red Cross, blue is
broad, black for both)
What are the differences in map
interests between chapters? Why have
some chapters attempted mapping
while others haven’t?
Explains how interests may vary depending
on geography, potential hazards they face,
and servicing population and area.
What are the most common map
interests?
Tells the Red Cross what is needed and can
inform the development of national tools.
Tells us what the Red Cross finds valuable
in mapping.
121
What benefits can maps provide to
organizational awareness?
Reinforces the value of maps as a tool for
organizational awareness.
How many chapters have attempted
tech adoption? When? How was the
process initiated? Who was involved?
How successful have they been?
Reinforces the need for a national map
program.
Reinforces concepts of tech diffusion
processes.
What are the differences in the
approaches to map use? Can these
differences be explained?
Provides insight into how tech adoption is
approached dependent on factors such as
presence of volunteers who use maps.

Approach
In early 2013, an Enterprise License Agreement (ELA) between the American Red Cross
and Esri was announced. This national site license provided some chapters with software
that they were either independently paying for or using by proxy through an external
partner or volunteer. Other chapters were able to start mapping for the first time with the
software. I will conduct structured interviews with the map leads (aka GIS champions) in
the various chapters who have requested licenses for the software, as well as other map
leads who are not using the licenses but have been referred to me by others.
The difficulty in reaching general conclusions from this study is that the Red Cross is an
incredibly large organization made up of many chapters with different interests. In
addition to different chapters’ individual challenges, each chapter also consists of
multiple departments focusing on very different things. Further, there are few mandatory
consistencies in organizational structure between chapters. For example, the preparedness
education activities are housed in the disaster response department of some chapters, but
may be housed in the health and safety department of others. For the purposes of this
study, we will use the term Area of Involvement (AOI) to describe the concept of a
department or activity. “AOI” may be used interchangeably with “department,” but the
122
former is more accurate given the organizational structure of the Red Cross. The AOI’s
that we have identified are:
 Client Services
 Development and Fundraising
 Logistics
 Partners
 Preparedness Education
 Sheltering
 Volunteer Services
This study will demonstrate how map representations of data can provide greater
understanding of an organization’s services as well as an increased ability to understand
inter-department awareness. Additionally, an organization similar to a local chapter of the
American Red Cross can use this information in understanding the value of maps and
how to attempt mapping adoption. There will be many elements of this story that can be
relevant to a wide variety of organizations, regardless of their business focus.

123
University of Southern California
3616 Trousdale Parkway, AHF B55
Los Angeles, CA 90089-0374

INFORMATION/FACTS SHEET FOR NON-MEDICAL
RESEARCH

Geospatial technology implementation at local Red Cross chapters

PURPOSE
The purpose of this study is to explore how local chapters of the American Red Cross
have engaged in spatial visualizations and analyses through the use of maps. Structured
interviews with the map leads of various chapters will offer insight into the experiences
of those in the Red Cross who have attempted geospatial technology adoption.

PARTICIPANT INVOLVEMENT
You will be asked questions regarding your role in implementing geospatial technology
or geographic information management during a phone interview. The interview will be
structured with a document guiding the questions asked. The interviews should last about
30 minutes on average.

CONFIDENTIALITY
Your name and chapter affiliation will be recorded in connection with this study.
However, no identifiable information other than your chapter information will be
reported in subsequent reports and publications.

The members of the research team and the University of Southern California’s Human
Subjects Protection Program (HSPP) may access the data. The HSPP reviews and
monitors research studies to protect the rights and welfare of research subjects.

INVESTIGATOR CONTACT INFORMATION
This survey is a joint study between the Spatial Sciences Institute at USC and the Los
Angeles Chapter of the American Red Cross. For more information, please contact Evan
Lue at evan.lue@redcross.org.

IRB CONTACT INFORMATION

University Park IRB, 3720 South Flower Street #301, Los Angeles, CA 90089-
0702, (213) 821-5272 or upirb@usc.edu

124
Interview Questions
This survey is part of a joint study by the Spatial Sciences Institute at the University of
Southern California and the American Red Cross Los Angeles Region to better
understand how different chapters have approached the adoption of digital maps to meet
their mission.
Mapping is defined here as any activity where a computer-based map or geographic data
is created.
Please direct any questions to Evan Lue (GIS Project Lead, Los Angeles Region) at
Evan.Lue@redcross.org. Thanks for your participation!
Tell Us About Yourself
This section is for basic background information on you as a leader in mapping.
American Red Cross Chapter
Your name
Employee or volunteer? Employee / Volunteer / Other
Role in the chapter
On average, how many hours a week do
you spend doing work for the Red Cross?
○ 0-8 (1 day or less)
○ 8-24 (1-3 days)
○ 24-40 (3-5 days)
○ 40+ (more than 5 days)
What single area of involvement with the
Red Cross are you most knowledgeable
about?
○ Client Services
○ Development and Fundraising
○ Disaster Services Technology
○ Information & Planning
○ DA
○ International Services
○ Logistics
○ Mass Care
○ Partners
○ Preparedness Education
○ Sheltering
○ Volunteer Services
125
○ Other _____________________
How did you hear about the Red
Cross/Esri ELA? If you heard from
another person, please include their name,
title, and chapter in "Other"
○ Saw it in a CrossConnection
○ Saw it while browsing CrossNet
○ Other______________________

Describe your level activity in mapping Red Cross data
○ It’s not my primary activity, and I only do it a little
○ It’s not my primary activity, but I do it a lot.
○ It is my primary activity

Describe your background in mapping or GIS:

Do you commonly create maps that others ask for?
○ No
○ Yes, less than once a month
○ Yes, about once a month
○ Yes, more than once a month

Are your maps used for planning or response?
○
Exclusively
planning
○
Primarily
planning, some
response
○
Both equally
○
Primarily
response, some
planning
○
Exclusively
response

Current Efforts
This section is about your current mapping efforts.
Does your chapter have a well-defined mapping effort, or does it partner with an
organization that assists with your chapter’s mapping needs?
○ Yes, we do our own mapping
○ Yes, another organization does our mapping ____________________
○ Yes, we do our own mapping and another organization assists us as
well_______________
○ No, we have tried to do mapping but have been unsuccessful
○ No, we have never attempted to do mapping
○ I don't know

What software have you tried and what do you currently use?
Software Tried (and not using because…) Currently Use
ArcGIS Desktop
ArcGIS Server
126
ArcGIS.com
Google Earth
Google Maps API
Maps.google.com
Depiction
Other

Please describe your mapping efforts:
________________________________________________________________________
________________________________________________________________________
________________________________________________________________________

How many people in your chapter actively participate in mapping activities?
__________

How much do you believe the mapping has improved:
Layer 1
Not at all
2 3 4 5
Greatly
Your ability to do your job or task?
Your understanding of how much and
where services are provided in your
own area of involvement (AOI)?

Your understanding of how much and
where services are provided by other
areas of involvement (AOI)?

Your understanding of how much and
where services are provided by the Red
Cross as a whole?

Your CHAPTER’S overall ability to
plan and prepare for disasters?

Your CHAPTER’S overall ability to
respond to disasters?

Adoption Process
This section describes how the technology was adopted and what investments were made.
127
Describe the steps taken to accomplish your goals
Tell us how you started
When?
Who initiated it?
Who was involved (employees, volunteers, managers)?
How did the efforts change over time?

If you paid for any software, please describe:
The vendor
How many licenses
How much you paid per license or in total
How long the license term is
If you plan to renew
Other notes (if you paid for other software, please describe the same as above here):

If the software was donated, please describe:
The vendor
How many licenses
How much you paid per license or in total
How long the license term is
If you plan to renew
Other notes (if you paid for other software, please describe the same as above here):

Tell us about your team and your operations
Number of employees that do mapping:
Number of volunteers:
Grants:
Ballpark Expenses:
Collaborations:
Amount of total person-hours spent per week:

Personal Mapping Interests
In this section we ask about the maps you want to see and the data that should be on
these maps.
Rate the value of having the following layers on a map:
Layer 1
None
2 3 4 5
Very
High
Hazard/Risk Areas
Census/Demographics
Critical Infrastructure (e.g. public utility
facilities)

128
Political Boundaries
Response Boundaries
ARC Volunteers*
ARC Donors*
ARC Disaster Responses*
ARC Outreach Events (e.g. CDE,
Preparedness Education)

ARC Shelters
Red Cross Partner Locations
Red Cross Logistical Supplies
*These layers are to be mapped with no personally identifiable information (PII)
associated, such as name and address; this layer would only display locations randomized
up to 500 feet.

Benefits and Return on Investment
This section is about how mapping has enriched your ability to meet the Red Cross
mission.
What were the expected benefits when you started?

What are the benefits you see now?

Challenges Faced and Overcome
This section asks what challenges were faced in getting where you are.
How do you rank the importance of the following factors in a successful mapping
effort?
1 Not at
all
2 3 4 5
Extremely
Presence of a map project lead (GIS
champion)

Tech savvy volunteers
Hardware availability
129
Software availability
Well-defined workflows
General interest of chapter
General interest of senior management
Access to data internally from various
sources

Access to data from partners
Standards for data security/privacy

Where have you had or expect to have the most difficulty?
1 Not at
a
problem
2 3 4 5 Still a
big
problem
Presence of a map project lead (GIS
champion)

Tech savvy volunteers
Hardware availability
Software availability
Well-defined workflows
General interest of chapter
General interest of senior management
Access to data internally from various
sources

Access to data from partners
Standards for data security/privacy

What is keeping you from reaching the following goals?
Goal Not
desired
Short on
Personnel/
Time
Lacking
training or
knowledg
Lacking
resources
Privacy
concerns
N/A,
we’ve
130
e of the
subject
already
done it
Authoring
a web
map

Using
maps for
planning

Using
maps for
operations



What’s Next
This section asks what you’d like to see happen next with mapping.
Please describe your wishlist for Red Cross mapping.

Is there anyone else you think I should talk to?

131
APPENDIX B: CHAPTER FOUR SURVEY (ABRIDGED)

132

133

134

135

136

Abstract (if available)

Abstract The field of study lying at the intersection of geospatial technology and emergency management has grown significantly in recent years and continues to rapidly evolve. Developments are largely the result of advancing technology, making display and analysis of geographic information easier and faster. Collaborative capabilities of technology have also improved, allowing a skilled volunteer to contribute to a disaster response operation taking place thousands of miles away. These developments in “crisis mapping” enable individuals to participate in providing aid and relief through information creation, dissemination, and analysis. While this is happening, an opportunity exists to ensure that long-established organizations keep pace with technology and can adopted to properly accommodate new tools for rapid response.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Developing, maintaining, and employing crowd-sourced geospatial data in support of helicopter landing zone surveys for disaster response operations

PDF

Geospatial web application development to access irrigation asset data: Veterans Affairs Palo Alto Health Care System

PDF

Environmental justice: geospatial impacts of hazardous materials spills

PDF

Preparing for natural disasters: investigating the effects of gain-loss framing on individual choices

PDF

Spatiotemporal visualization and analysis as a policy support tool: a case study of the economic geography of tobacco farming in the Philippines

PDF

Geospatial analysis of the Round Fire: a replication of burn severity analyses in the Sierra Nevada

PDF

The impact of social capital: a case study on the role of social capital in the restoration and recovery of communities after disasters

PDF

Site location suitability analysis for a smart grid network

PDF

Mapping firing ranges as social capital generators in Houston, Texas

PDF

Cal ToxTrack: a full stack Web GIS for mapping pollution in California

PDF

Explorations of American churchscape diversity

PDF

Using pattern oriented modeling to design and validate spatial models: a case study in agent-based modeling

PDF

The role of pesticide exposure in breast cancer

PDF

Evaluating spatial changes in the rate of insurgency‐violence in Central Africa: the Lord's Resistance Army 2008-2012

PDF

Exploring urban change using historical maps: the industrialization of Long Island City (LIC), New York

PDF

Assessing the impact of a web-based GIS application to promote earthquake preparation on the University of Southern California University Park Campus

PDF

The role of GIS in asset management: integration at the Otay Water District

PDF

Integrating GIS into farm operations at the Homer C. Thompson Research Farm in Freeville, New York

PDF

Water Rights Permit System (WRPS): a GIS‐based tool for the Umpqua Drainage Basin

PDF

Eye.Earth Pro (Beta v1.0): application development and spatial financial analysis utilizing the PESTELM framework

Asset Metadata

Creator Lue, Evan (author)

Core Title Defining the geospatial needs of a ubiquitous disaster relief organization

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Geography

Defense Date 08/26/2014

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag American Red Cross,disaster response,emergency management,geospatial,GIS,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wilson, John P. (committee chair), Franklin, Meredith (committee member), Vos, Robert O. (committee member)

Creator Email elue@dornsife.usc.edu,evan.lue@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-469845

Unique identifier UC11287189

Identifier etd-LueEvan-2886.pdf (filename),usctheses-c3-469845 (legacy record id)

Legacy Identifier etd-LueEvan-2886.pdf

Dmrecord 469845

Document Type Dissertation

Format application/pdf (imt)

Rights Lue, Evan

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA