Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Evaluating the utility of a geographic information systems-based mobility model in search and rescue operations
(USC Thesis Other)
Evaluating the utility of a geographic information systems-based mobility model in search and rescue operations
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
EVALUATING THE UTILITY OF A
GEOGRAPHIC INFORMATION SYSTEMS-BASED MOBILITY MODEL
IN SEARCH AND RESCUE OPERATIONS
by
Mark Powell Johnson
Presented to the
Faculty of the USC Graduate School
University of Southern California
In Partial Fulfillment of the
Requirements for the Degree
Master of Science
(Geographic Information Science and Technology)
August 2016
Copyright © 2016 by Mark Johnson
I dedicate this to my family and my friends for their patience as I invested many nights and
weekends to follow my passion in geographic information science and technology. I write this
for those that volunteer so many hours to search for and rescue those in need,
hoping to make their efforts even more effective.
iv
Table of Contents
List of Figures ................................................................................................................................ vi
List of Tables ................................................................................................................................ vii
Acknowledgements ...................................................................................................................... viii
Abbreviations ................................................................................................................................. ix
Abstract ......................................................................................................................................... xii
Chapter 1 Introduction .................................................................................................................... 1
1.1 Why Choose Search and Rescue? ........................................................................................1
1.2 Why Study a Mobility Model?.............................................................................................4
1.3 Research Objectives .............................................................................................................5
1.4 Research Overview and Stages ............................................................................................6
1.5 Expected Outcomes .............................................................................................................7
Chapter 2 Existing Research Relating to Search and Rescue ......................................................... 8
2.1 Search and Rescue Science ..................................................................................................8
2.1.1. Early Research ...........................................................................................................9
2.1.2. Search Theory Applied to SAR .................................................................................9
2.1.3. Recent Research .......................................................................................................13
2.1.4. Current Use of SAR Science....................................................................................14
2.2 Mobility Models .................................................................................................................15
2.3 Mapping Tools Used in Search and Rescue.......................................................................16
2.4 Integrated Geospatial Tools for Search and Rescue ..........................................................17
2.4.1. Available Tools ........................................................................................................18
2.4.2. Mobility Model ........................................................................................................22
2.4.3. Documentation .........................................................................................................26
2.5 Summary ............................................................................................................................26
Chapter 3 Data and Methods......................................................................................................... 28
3.1 Research Process ................................................................................................................28
3.2 Software Used ....................................................................................................................30
3.3 Data Required ....................................................................................................................31
3.4 Data Used in the Model .....................................................................................................32
3.4.1. SAR Data Sets..........................................................................................................32
3.4.2. Base data ..................................................................................................................35
v
3.4.3. Conditioning the SAR Data .....................................................................................37
3.5 Variables Required for SAR Mobility Model Analysis .....................................................38
3.6 IST4SAR Model Processes ................................................................................................40
3.6.1. IGT4SAR Mobility Model Workflow .....................................................................41
3.6.2. Changes Made to IGT4SAR for Throughput Improvement ....................................42
3.6.3. Model Outputs Combined ........................................................................................43
3.7 Model Assumptions and Considerations............................................................................46
3.8 Summary ............................................................................................................................47
Chapter 4 Results .......................................................................................................................... 48
4.1 Model Results Findings .....................................................................................................48
4.1.1. Training Required ....................................................................................................49
4.1.2. Computation Time ...................................................................................................49
4.1.3. Success of Modeled Travel Time ............................................................................50
4.2 Lost Subjects ......................................................................................................................51
4.3 Unusual Incidents ...............................................................................................................52
4.4 Summary ............................................................................................................................54
Chapter 5 Discussion and Conclusions ......................................................................................... 56
5.1 The Model and Reality .......................................................................................................58
5.2 Limitations and Hindsight ..................................................................................................60
5.3 Future Work .......................................................................................................................61
5.4 Conclusion .........................................................................................................................62
References ..................................................................................................................................... 63
Appendix A: Oregon’s Incident Data Collection Form ................................................................ 66
Appendix B: Python editing of IGT4SAR .................................................................................... 67
Appendix C: Tool Results from Mobility Model Run .................................................................. 68
Appendix D: Results Sorted by Distance Traveled ...................................................................... 70
vi
List of Figures
Figure 1 Model produced by a classification and regression tree (CART) .................................... 2
Figure 2 Example ICS-based organizational chart for a SAR team ............................................... 3
Figure 3 Study area in blue (Lane and Deschutes Counties) .......................................................... 7
Figure 4 Example probability of area (POA) map ........................................................................ 11
Figure 5 IGT4SAR’s main toolbox............................................................................................... 18
Figure 6 IGT4SAR’s Utilities toolbox .......................................................................................... 19
Figure 7 IGT4SAR’s UpdateParameters toolbox ......................................................................... 19
Figure 8 IGT4SAR’s UpdateParameters toolbox ......................................................................... 19
Figure 9 IGT4SAR’s Logistics toolbox ........................................................................................ 20
Figure 10 IGT4SAR’s Operations toolbox ................................................................................... 21
Figure 11 IGT4SAR’s Planning toolbox ...................................................................................... 22
Figure 12 Example of terrain’s effect on isochrones and the Theoretical Search Area ............... 25
Figure 13 Overall Research Process ............................................................................................. 30
Figure 14 Oregon SAR Find Points .............................................................................................. 34
Figure 15 Base data for Lane and Deschutes Counties................................................................. 36
Figure 16 Map showing the mobility model output...................................................................... 40
Figure 17 IGT4SAR Theoretical Search Area results for all 51 incidents ................................... 44
Figure 18 Using the XY to Line tool to associate IPPs with Find Points ..................................... 45
Figure 19 IGT4SAR Theoretical Search Area results, IPPs, and Find Points .............................. 48
Figure 20 Theoretical Search Area for extraordinary trail runner ................................................ 53
Figure 21 Theoretical Search Area for hiker ................................................................................ 54
Figure 22 Comparing National Geographic Society topographic map to the IGT4SAR ............. 57
vii
List of Tables
Table 1 Simple consensus matrix ................................................................................................. 13
Table 2 Land cover impedance values .......................................................................................... 23
Table 3 Examples of the federal codes for roads .......................................................................... 24
Table 4 Example trails impedance values ..................................................................................... 24
Table 5 Comparison of SAR data sets .......................................................................................... 33
Table 6 Oregon database fields and example data ........................................................................ 35
Table 7 Example results ................................................................................................................ 50
Table 8 Success of results from the Theoretical Search Area tool ............................................... 50
Table 9 Aggregated success of results from the Theoretical Search Area tool ............................ 51
Table 10 Subject characteristics .................................................................................................... 52
viii
Acknowledgements
I am grateful to Professor Karen Kemp for helping me develop my thoughts, capture them, and
convey them in a meaningful way. I also appreciate Dr. Su Jin Lee, Dr. Laura Loyola, as well as
my other Spatial Sciences Institute professors for all their instruction and support. I thank my
supervisors, Tom Ducker, Karl Johnson, and Ted Ross for their flexibility with my schedule. I
would like to also thank the Oregon Office of Emergency Management and the Nevada County
Sheriff’s Search and Rescue unit for sharing their data and experiences, as well as Paul Doherty
PhD and Don Ferguson PhD for sharing their thoughts and research on this important subject.
ix
Abbreviations
ABM Agent-based models
BYU Brigham Young University
CART Classification and regression tree
DEM Digital elevation model
FEMA Federal Emergency Management Agency
GDB Geodatabase
GIS Geographic information systems
GISci Geographic information science
GIST Geographic information science and technology
GMU George Mason University
GNSS Global navigation satellite systems
GPS Geospatial positioning system
IC Incident command
ICS Incident command system
IGT4SAR Integrated Geospatial Tools for Search and Rescue
IHOG Interagency Helicopter Operations Guide
IPP Initial planning point
ISRID International Search and Rescue Incident Database
LKP Last known point
LPB Lost person behavior
LZ Landing zone
MRA Mountain Rescue Association
x
MXD Map file
NAD North American datum
NAPSG National Alliance for Public Safety GIS
NASAR National Association for Search and Rescue
NHD National Hydrological Dataset
NHTSA National Highway Traffic Safety Administration
NLCD National Land Cover Dataset
PDF Portable document format
PLBs Personal locating beacons
PLS Point last seen
POA Probability of area
POC Probability of containment
POD Probability of detection
POS Probability of success
PSR Probability of success rate
ROW Rest of the world
SAR Search and rescue
SSARCC State Search and Rescue Coordinators Council
SSI Spatial Sciences Institute
TIGER Topologically integrated geographic encoding and referencing
TNP Terrain Navigator Pro
UAS Unmanned aerial systems
USC University of Southern California
xi
USGS United States Geological Survey
USNG United States National Grid
UTM Universal Transverse Mercator
WiSAR Wilderness search and rescue
YOSAR Yosemite Search and Rescue
xii
Abstract
Every year thousands of people become lost or injured to the extent that a search and rescue
(SAR) unit needs to step in and help. Through the ages, we have needed to look for people and
things yet the theory behind searching goes back less than 75 years to World War II. The main
idea is that to be successful, searchers need to search the right area, and be able to detect the
person or thing. This research explored the utility of using a GIS-based mobility model to assist
search planners in developing their search areas. A mobility model incorporates consideration of
the speed with which a person can move across the landscape. The tool used here is an Esri
ArcGIS template called Integrated Geospatial Tools for Search and Rescue (IGT4SAR). While it
includes many SAR tools, this research focused on the mobility analysis component. This study
specifically assessed IGT4SAR’s ease of use, speed, and success rate at determining how far a
person can travel in a given time. Nevada County provided detailed information on a few
incidents used to gain familiarization with IGT4SAR and the state of Oregon provided a large
database of historical and diverse SAR events that allowed for broader testing of the model.
Ultimately, 44 incidents were used to test the model. The model itself is easy to use, but the
template is complex. With preloaded data, the model creates a product in less than 15 minutes.
Starting with an unrealistic assumption that the incident start time recorded in the database
represented the time when the subject left the last known location, test runs resulted in a 30%
success rate where the found location fell in a time band that was less than the amount of time
between the start time and the found time recorded in the database. After adding a estimated
three-hour delay in reporting time to the SAR notification times the model had a 75% success
rate. These results suggest that IGT4SAR can assist in defining a containment area to limit a
search radius and is worthy of continued development.
1
Chapter 1 Introduction
“I cannot begin to tell you how grateful I am that Chris is alive because of you. The response to
my call was phenomenal … thank you for all that you do to help people in trouble. I cannot
picture a braver set of people … nothing can ever repay what you have done. Thank you, Sara
Ray” (Ray 2015).
The Deschutes County Search and Rescue Team received this note in 2015 after a successful
search and rescue. It provides a brief example of the importance of effective search and rescue
operations. This research strives to make a small contribution to further develop search and
rescue (SAR) operations and its use of geographic information science and technology (GIST). A
number of organizations in the SAR community currently use geographic information systems
(GIS) but primarily as a tool to produce simple maps. Further leveraging GIS data and analysis
in order to find missing persons provides endless opportunities for growth in the field. This paper
specifically looks at the effectiveness of using a mobility model that considers travel cost path
analysis techniques.
1.1 Why Choose Search and Rescue?
SAR is naturally a geospatial problem (Doherty et al. 2014). People can become lost,
hurt, trapped, despondent, or even abducted. Something along these lines happens daily across
the world, totaling in the thousands as evidenced by the International Search and Rescue Incident
Database (ISRID) (Koester 2008). These people need help, and as illustrated in Figure 1, they
need it fast (Adams et al. 2007). Even with our advanced technologies such as global navigation
satellite systems (GNSS), personal locating beacons (PLBs), and mobile phones, this is still
simply the reality we live in. There are hundreds of SAR units across the United States and many
others across the world that are there to help. Volunteers usually staff these units, regularly
investing their time-off to support others. Reducing the time spent searching for these people in
2
need of aid is to the benefit of both the lost and the searching. Techniques that assist the planners
in defining the search area will likely lead to finding the lost subject in less time, which
dramatically increases chances for survival (Cooper, Frost, and Robe 2003). GIS in SAR is not a
replacement for other SAR tools, such as paper maps and expert planners, but can be an
enhancement working to supplement and validate other capabilities.
Figure 1 Model produced by a classification and regression tree (CART)
(Source: Adams et al. 2007)
Search and rescue teams often organize at the county level and can vary greatly from one
jurisdiction to another. In California there is typically one Sheriff’s Deputy assigned to the SAR
team. The team has a SAR Volunteer Coordinator that works issues often at the strategic level
and various team leaders working at the tactical level. The broad structural goal is for the group
to organize according to the Federal Emergency Management Agency’s (FEMA) Incident
Command System (ICS) as shown in Figure 2. This system allows for command and control, is
3
familiar to the emergency management community, and can scale from a small incident to one
that requires aid from other nearby teams or potentially those from across the state.
Figure 2 Example ICS-based organizational chart for a SAR team
Often volunteers arrive ready to search yet the planning team has not had enough time to
divide the area into efficient search segments for the three to five person search teams. Getting
the search teams out initially can be a bottleneck. Being able to use a mobility model to assist in
the decision process may help speed the process. Thankfully, 50% of searches end with the
subject found in three hours or less (Koester 2008). When the subject remains missing after a
number of hours, using the search teams effectively becomes a major concern because the area
4
that a missing person may possibly be located in becomes exponentially larger as time continues.
Having a good estimation of how far a person potentially could travel in the various directions,
given the relevant information, would be valuable. With this information, the planners would be
less apt to send search teams out into areas where the subject is unlikely to be. This is where a
GIS mobility model with the capability to adjust different variables may prove useful.
1.2 Why Study a Mobility Model?
GIS has been somewhat involved in the search and rescue field for many years and even
more so today as technologies in software, hardware, and data acquisition advance. To address
the administrative aspects of a SAR incident, developers created a few GIS software solutions,
such as MapSAR and Terrain Navigator Pro that focus on facilitating operations rather than
informing operations. Over the last few years, analyzing the potential of GIS to identify a
missing person’s location more effectively and efficiently is receiving more attention (Doherty et
al. 2014). While there has been a fair amount of study on mobility models in general, also
referred to as a motion or cost-distance model (Vogt, Nikolaidis, and Gburzynski 2012), Sava
points out that very few researchers write with regard to search and rescue operations (Sava et al.
2015; Doherty 2013).
With today’s remote sensing, digital storage, and computer processing capabilities,
coupled with the accumulating historical SAR incidence data, geographic information science
(GISci) should be able to contribute to search and rescue endeavors to a greater degree. A GIS
can accurately and efficiently evaluate geographic data such as streams and watersheds,
elevation, vegetation, geology, transportation routes, etcetera, to develop a travel cost layer. This
layer, combined with information regarding the mental and physical characteristics of the lost
person, then can calculate and illustrate the areas a person could likely travel to in a given
5
amount of time. This method will not lead the searchers directly to the missing person, but if it
can more accurately determine the search area, it is a step in the right direction.
Lately, a few people active in the SAR community started rigorously examining what
more GISci can bring to the search and rescue field (Doherty et al. 2014; Durkee and Glynn-
Linaris 2012; Ferguson 2008). Each paper contributes in its unique way, with some addressing
the methodology and others looking more at the data collection and its conditioning. While
working with data is a foundational part of any project, the focus of this research is evaluating
the usefulness of a GIS-based mobility model in search and rescue operations. As our society
encourages people to get out and enjoy the outdoors, equipping SAR teams with better tools such
as mobility models may be significant, especially to those that have been lost and found or
injured and rescued.
1.3 Research Objectives
The main objective that this research addressed was to evaluate whether a GIS-based
mobility model tool can be useful in search and rescue operations. Three underlying sub-
objectives support this overall objective. The first is assessing if using a mobility model GIS tool
is feasible without having extensive training. The second is considering if the process is
executable in a short timeframe with limited computing resources. The last underlying objective
is determining if the model can reliably give the incident command (IC) planners an effective
outer limit for their search areas. Ultimately, checking this tool in real-world situations will
decide its true utility. In order to arrive at the overall research objective, various areas were
investigated including GIS analysis techniques currently used in SAR, search theory, lost person
behavior, search operations, wilderness search and rescue, and travel cost modeling. These
6
subjects either led to the current line of study or supported the investigation into the utility of
mobility models in the search and rescue context.
1.4 Research Overview and Stages
This research took a number of different turns following the discoveries of work already
completed coupled with a better understanding of the data. Initially, the thought was to create a
model that would display the greatest likely distance that a lost subject would travel in a given
amount of time. Both static and dynamic environmental conditions would have been the basis for
the calculations as well as the subject’s particular relevant characteristics. During the research for
this approach, a tool that executed some of these very tasks was discovered. Ferguson developed
a model in an Esri ArcGIS template called Integrated Geospatial Tools for Search and Rescue
(IGT4SAR), which also includes many other SAR tools (Ferguson 2012). Exploration of
different avenues occurred upon that discovery. Examining the impacts of using different land
cover data sets, or using different resolution elevation data sets were considered as potential
points of focus. Ultimately, determining the usefulness of the existing mobility model became
the main goal. This effort broadly included completing initial research to define the appropriate
research objectives, deciding on a methodology, searching for applicable data sets of actual SAR
incidents, executing the method, and evaluating the results.
The specific study area, illustrated in Figure 3, consists of two Oregon counties, Lane and
Deschutes. The decision on the area came following an evaluation of availability and
characteristics of a number of different SAR incident data sources. Balancing the number of
incidents, computational time required, and diversity of subjects and topography, all were part of
the decision process.
7
Figure 3 Study area in blue (Lane and Deschutes Counties)
1.5 Expected Outcomes
Expected outcomes developed from the beginning of the study and continued throughout
the rest of the process. Before learning of IGT4SAR, there was an expectation that a GIS-based
SAR mobility model could be learned rather easily. Another was that the anticipated calculation
time for developing a furthest extent search area or containment distance would take 10 to 20
minutes with a reasonably capable computer. Having had discussions with experienced SAR
personnel and initially working with IGT4SAR on two SAR incidents, the major expected
outcome was that 100% of the lost subjects would be found within the model’s containment
distance assessment. There was also an expectation that the model would greatly overestimate
the furthest point a subject could travel in a given amount of time.
8
Chapter 2 Existing Research Relating to Search and Rescue
This chapter is based on an extensive literature review coupled with discussions with
experienced SAR volunteers. Those contacted included people from both the west and east coast.
Some came from an academic perspective but all were actively serving in the SAR community.
The literature review was ongoing and included sources written specifically for SAR as well as
those that examined associated or supporting topics. Much of the related research was located in
journals although a number of books and websites were found to be relevant.
In November 2015, the National Alliance for Public Safety GIS (NAPSG) Foundation
sponsored an excellent workshop held near Yosemite National Park called SARGIS7 that drew
more than 50 people with interests in SAR, GIS, or both fields to discuss advancing the
connection between the two (Doherty 2015). This provided an opportunity for some important
dialogues with leaders in this area of study. At the workshop, Donald Ferguson, PhD traveled
from the east coast and presented his extensive work on the only publically known GIS-based
mobility model for SAR (Doherty et al. 2014). The model is one tool among a multitude that are
part of an impressive template for ArcMap called Integrated Geospatial Tools for Search and
Rescue (IGT4SAR).
2.1 Search and Rescue Science
It is crucial to remember that a rescue cannot happen without first finding the subject.
Many people become lost each year, and this has been the case for countless years. On the other
side of these situations is that fact that, usually, there are people out there searching for these
people. Beyond just lost subjects there are also other things people search for, such as animals,
plants, gold, and even one’s military enemies. Surprisingly, the task of searching for people and
things had not been scientifically addressed until World War II.
9
2.1.1. Early Research
Bernard Koopman was the first to address search theory in literature after working for the
US Navy during the Second World War (Koopman 1946). He worked on a team that was
assembled to conduct research that could help military operations improve, it was called
operations research. This discipline is still alive and well although it is applied to much more
than just military operations today. Koopman’s original charge was to locate enemy ships and
submarines. As promise was seen in the work, it was also applied to searching for friendly pilots
that had been shot down over the Pacific Ocean. While there are vast differences between
searching on land and searching on the water, which has no slope to speak of, roads, or surface
differences, the general theory is still quite applicable. It was somewhat unanticipated to see how
much high-level math is involved in what appears to be a relatively simple concept.
In search theory there are basically two costs that need to be considered. The most
obvious is the cost of the search, which can be measured in terms of time, money, effort,
etcetera. The other is the cost of the subject being searched for, which can also be measured in a
number of different ways including money, inconvenience, or even lives. These two costs need
to be balanced. For example, if a person loses a pencil between their home and a destination five
miles away, the cost of the search would far outweigh the cost of the pencil. The opposite is also
true and in the case of searching for a missing person, it is likely the only cost considered too
high would be if someone on the SAR team had his or her life placed in grave danger or if the
subject cannot be found and the exposure time overwhelmingly indicates death.
2.1.2. Search Theory Applied to SAR
Put simply, the search theory idea is that in order to find something, searchers need to be
looking in the right area and be observant enough to detect the object when the correct area is
10
searched. The right area is represented by the term probability of area (POA) and is calculated
partly from the subject’s point last seen (PLS) or their last known point (LKP). Either of these
points can be used by the team as their initial planning point (IPP). The likelihood that the
searcher would detect the object being looked for is referred to as the probability of detection
(POD). Both the POA and POD are expressed as percentages. The POA is the percentage chance
that what is being looked for is in the area being searched. There can be some confusion here as
some in the SAR community, primarily in the maritime environment, refer to the probability of
area concept as the probability of containment (POC) (Koester 2008; Lovelock 2008).
In SAR operations, one team cannot search the entire area in which the person is likely to
be found, so the search manager divides the area into smaller segments as seen in Figure 4. The
whole search area’s POA should approach a value of 100% when adding all the search segments
together but some portion should be allocated to the area outside the designated search area. This
area outside the search area is termed the rest of the world (ROW) (Phillips et al. 2014). The
larger the planners draw the search area or POA, the smaller the ROW percentage will be. The
goal is to get the POA to be as small a geographical area as possible while keeping the ROW
percentage as small as possible. Each team’s smaller search segment will potentially be in the 5-
25% range.
11
Figure 4 Example probability of area (POA) map divided into search segments with red
indicating a higher POA and dark green a lower POA (Source: Ferguson 2008)
The POD is how well the search team searches their segment. It is considered physically
impossible to search any area to 100% coverage so, on extended searches, a segment may be
searched more than once in order to increase the percentage covered for that segment. Key
variables influencing POD are the team’s competence, speed, spacing, as well as the conditions
such as vegetation, topography, lighting, and so forth (Phillips et al. 2014). There are different
ways this variable is determined. One of the simplest is when a team returns from searching their
assigned segment, they are asked something along the lines of “if 10 subjects were lost in your
segment, how many would you have found?” A more technical means is downloading the track
12
log from their global positioning system (GPS) device and then buffering the track to the
distance the team was spaced, also called an effective sweep width (ESW), and then finally
calculating the percentage of their search area that is intersected by the buffered area. Combining
both answers to establish a number is also used. The result expressed as a percentage, using
whichever method, is the POD.
The probability of success (POS) in finding the subject equates to the probability of area
multiplied by the probability of detection. This is illustrated in Equation (1).
(1)
Lost person behavior (LPB) is studying what people do when they become disoriented in
the wilderness in order to try to better estimate where a lost person might be. This is helpful
because a typical person can walk three miles in an hour, which quickly makes for a large search
area. Syrotuck completed seminal work in this area in his examination of 229 incidents, mainly
from the states of Washington and New York (Syrotuck 1976). He divided lost people into six
categories: small children, children, hunters, hikers, miscellaneous outdoors, and elderly people.
He calculated and averaged the distances they traveled in flat and hilly terrain. He then
determined the distance ranges these people, by quartile, were found from the place they were
last seen. This information assists in keeping the search area to a more manageable size.
Combined with these statistical aids, decision by consensus is often used either
informally or formally. This is done by having experts evaluate the possible distance and
direction the subject would take, ranking their choices privately so as to reduce bias, and then
combining the evaluations to determine the group’s decision. The associated output can be seen
in Table 1 (Mattson 1980).
13
Table 1 Simple consensus matrix (Source: Mattson 1980)
2.1.3. Recent Research
More in depth LPB research has recently been done by Koester. Through a USDA grant,
his dbS Productions LLC has gathered data sets from across the United States and around the
world accumulating more than 50,000 SAR incidents. The International Search and Rescue
Incident Database (ISRID) has allowed Koester to further divide lost people into 41 categories
and the incidents into multiple ecoregions. The larger sample sizes provide greater confidence in
his statistical analysis, which considers distances from their point last seen, track offset, and
mobility hours, among others. Koester’s book, Lost Person Behavior, is known throughout the
SAR community and seen as an essential resource by those that study search and rescue
operations (Koester 2008). Work has also been done on developing a Bayesian approach to
modeling LPB with the ability to incorporate SAR planners’ expert opinion and terrain data (Lin
and Goodrich 2010)
More recent advances in SAR science have come from Doherty’s work in Yosemite
National Park (Doherty 2013). He looked at a number of search and rescue tasks and connected
them with GIS capabilities. The three broad areas investigated were prevention, search, and
rescue. The preventative search and rescue (PSAR) aspect discussed georeferencing text-based
data from previously written reports into a format appropriate for analysis. The searching section
14
investigated ways GIScience can assist the established statistical models to better determine a
search area. This work was done in conjunction with Ferguson (Doherty et al. 2014).
The rescue component compared two models, one expert-based and one GIS-based, that
were used to select subject extraction points that are accessible by helicopters (Doherty, Guo,
and Alvarez 2011). Jacobs has also recently contributed to the field in his report for the National
Association for Search and Rescue (NASAR) which identifies what geographic features lost
subjects are often found by and their frequency (Jacobs 2015). This analysis was done
independent of the person’s last known point partly because those points are of limited number
and quality.
2.1.4. Current Use of SAR Science
Even with the work done in search and rescue science, the adoption of new ideas is slow.
Syrotuck (1976) and Koester (2008) are frequently referred to in SAR trainings, more often with
training for incident command team members rather than the searchers themselves. Rose found
in her 2015 study that while GIS had some of the most advantages of the mapping technologies
used by SAR personnel, it is not used as often as other traditional methods because it is
considered difficult to learn (Rose 2015). Many training courses are still using Syrotuck’s
probability zones and probability tables from the 1970s, which come from a much smaller set of
data than the ISRID.
A large part of SAR science uptake involves the collection and storing of incident data.
The ISRID is the largest database as it compiles data from around the world. A challenge there is
the consistency of the data with some groups tracking all fields with a great deal of accuracy to
groups at the other end of the spectrum reporting very little information. Access to data is
another limiting factor to the adoption of the science. The case is likely that most of the SAR
15
teams are not recording their search data and uploading it to a larger collection effort. States like
Oregon, New York, Washington, and a few others are collecting data but many are not.
California, often an environmental and technical leader, is just starting to revive theirs. The 2015
State Search and Rescue Coordinators Council (SSARCC), which is where those state employees
that oversee their state’s SAR operation meet, had 11 states attend. Between this number, which
is an indicator of the level of many states’ engagement, and the fact that there is no standardized
data collection tool or mechanism listed at their website, highlights the lack of collective effort at
establishing a larger SAR data gathering mechanism (SSARCC 2016).
Even with the somewhat slow uptake, the future of SAR science with its use of
GIScience is bright. More people are conducting research in the field and data is becoming
easier to collect, share, and analyze. Other areas that are specifically being investigated that have
a GIS connection at the practical level are the use of unmanned aerial systems (UAS) and mobile
phone triangulation (Durkee and Glynn-Linaris 2012; Goodrich et al. 2008). Areas that have
promise of being practical but are more theoretical are agent-based models (ABM) and Bayesian
models (Doherty et al. 2014; Lin and Goodrich 2010).
2.2 Mobility Models
Mobility models have been used for a number of years and look at how animals, vehicles,
people, etcetera move over time. These models primarily look at nodes within a network and as
of late, particularly wireless devices on a network (Vogt, Nikolaidis, and Gburzynski 2012). This
work deals with advanced math and is moving forward as data from wireless networks become
more available. Mobility models can be generic or they can be developed for specific situations
such as visitors at theme parks (Solmaz, Akbaş, and Turgut 2015). These models are often
evaluated by comparing their results to GPS track data. Mobility models have been used in urban
16
planning, anthropology, and wildlife studies (Doherty 2013). In these models, costs are
calculated by applying a least-cost path algorithm to a source raster and a resistance raster
(Adriaensen et al. 2003).
Often times, in search and rescue events, the subject either has no phone or there is not a
sufficient network signal. It would be an interesting study to see how many GPS tracks could be
accumulated from lost subjects that actually had connected devices. This would further connect
mobility models and search and rescue operations.
Tobler's Hiking Function is foundational to mobility modeling in search and rescue
(Pingel 2010). It is an exponential function that determines hiker’s speeds. The function
estimates how fast a hiker will travel based on the slope angles encountered. This function
should be an important component in calculating potential distance covered.
A joint study between George Mason University (GMU) and Brigham Young University
(BYU) produced an online tool called MapScore that can be used to evaluate models using the
ISRID although at this point, it appears MapScore is not capable of evaluating a full GIS-based
mobility model (Twardy et al. 2011). In Ferguson’s IGT4SAR template, he has incorporated
multiple variables in his Theoretical Search Area model. This particular model receives more
attention below.
2.3 Mapping Tools Used in Search and Rescue
Some SAR units have been using GIS extensively yet it is not commonly applied to
search theory (Rose 2015). A few teams use ArcGIS for Desktop but of those that use some sort
of GIS tool, more use a software package designed specifically for search and rescue operations.
There are several mapping tools available that cover tasks search and rescue teams frequently
execute (Pfau 2013; Rose 2015). The oldest and most often used is the paper map and a pencil.
17
This tool is low cost, uses no batteries, and is familiar to most people. Although, as the next
generation moves to a greater extent into SAR operations, fewer people may have a history with
paper maps, having relied on digital devices for their navigation.
There are a few different options for digital mapping available to planners. Google Earth
is an intuitive tool used by many organizations that gives more up-to-date information than paper
maps, can show imagery, can store location information, and has many other strengths. SARsoft,
also referred to as SARtopo, is a tool that is browser-based that can also be used offline. The
offline version can be operated off a team’s own remote server so multiple users can work with
the same data on the same event. It has many excellent tools to assist in the administration of the
search.
MapSAR is an ArcGIS template and was a seminal project that made using a fully
capable GIS more accessible to the SAR community (Pedder 2012). An updated version is
currently being developed using ArcGIS Pro and its expanded capabilities. Terrain Navigator Pro
(TNP), OziExplorer, Mission Manager, and National Geographic Topo! offer other SAR
mapping solutions (Rose 2015). Of the various offerings, IGT4SAR has the most advanced
collection of analysis tools available and is further detailed in the next section.
2.4 Integrated Geospatial Tools for Search and Rescue
Integrated Geospatial Tools for Search and Rescue (IGT4SAR) is an ArcGIS template
that had its beginnings in MapSAR. It is a free, open-source template licensed under the terms
of the GNU General Public License as published by the Free Software Foundation and is
available on GitHub (https://github.com/dferguso/MapSAR_Ex). This is the most advanced GIS
analysis package for SAR operations. It offers not only multiple tools supporting administrative
18
functions but also may be the only one that offers multiple analysis tools. Specifically, IGT4SAR
incorporates a mobility model called the Theoretical Search Area model.
2.4.1. Available Tools
To work with the IGT4SAR, a new incident must be created based off the original
downloaded template. This will bring into ArcMap a customized IGT4SAR toolbar and in the
catalog window a home folder including an Esri geodatabase (GDB) and map file (MXD). In the
toolbar, there is a palette with six groups of tools as seen in Figure 5. Having a palette is helpful
considering the number of tools available. These tools are viewable in the Catalog under the
SAR_Toolbox10b toolbox and are typically well explained in the help section. All but two of the
tools are written in Python and the others were created using Esri’s ModelBuilder. One important
point is that most of these tools require input or uploading of other supporting data sets in order
to be effective. The template does help with team effectiveness and efficiency but it does not
magically create the inputs.
Figure 5 IGT4SAR’s main toolbox
The Utilities group has six tools within it. This is where the user finds the Create New
Incident tool. Here, a user finds tools to export both the feature layers in to Esri or Google Earth
formats, as well as exporting table data to Microsoft Excel. Other tools include Plot Point
Locations for adding points and Select Features by Distance from IPP (Initial Planning Point)
19
which is the primary point a SAR team uses to plan their activities. As can been observed in
Figure 6, printing maps using a Map Book is the sixth tool available.
Figure 6 IGT4SAR’s Utilities toolbox
The next group of tools is called UpdateParameters as seen in Figure 7. This is also a
grouping of utility-like tools. Search Area Names and Update Domains updates the respective
items in the user-defined folder. Update Map Layout refreshes the Universal Transverse
Mercator (UTM) zone, United States National Grid (USNG), and magnetic declination based on
the Incident Information Layer.
Figure 7 IGT4SAR’s UpdateParameters toolbox
In the Clues group, there are two tools. Both tools use ModelBuilder, Esri’s internal
graphical modeling tool. Figure 8 shows the first is a ClueHotSpot tool that uses Getis-Ord G
statistic to help planners assess the clustering of clues. The second is Hot Spot with Rendering.
Figure 8 IGT4SAR’s UpdateParameters toolbox
20
Logistics is another grouping in the SAR toolbox, and is shown in Figure 9. The
Estimated Radio Coverage tool uses a Digital Elevation Model (DEM) and the radio’s sensitivity
to provide an estimate of signal strength for the team’s radios. It considers tower power,
transmitter and receiver gain, and other factors specified in the user-entered radio tower feature
class. If coverage is not complete in the search area, the Repeater Locations tool will analyze the
optimal locations to place radio repeater stations. To facilitate the extraction of the subject after
they have been found, there is a Helicopter LZ (landing zone) tool. This tool uses Interagency
Helicopter Operations Guide (IHOG) standards to select safe landing areas. This tool uses a
given extent, DEM, National Land Cover Database (NLCD), LZ size, and slope to determine
potential areas. A machine learning LZ model has been examined against expert opinion in the
Yosemite National Park and the two methods matched 90% of the time with the model correctly
identifying 95% of the points in the test location data set (Doherty, Guo, and Alvarez 2011).
Figure 9 IGT4SAR’s Logistics toolbox
The Operations group of tools includes eight different options. Two create forms for the
missing person and the search team’s assignment as can be seen in Figure 10. There is a
Coverage tool that analyzes the search team’s global position system (GPS) device data to
determine their POD. The GPXtoFeatures tool uploads information such as clues from a GPS
into the GDB. Two tools are available to create a pre-defined search pattern for piloted aircraft or
UAS given the pattern type and sweep width. If an internet connection is available, there is a tool
that will retrieve the weather forecast. The last tool exports the post-incident data into a file that
can be shared using the ISRID data format.
21
Figure 10 IGT4SAR’s Operations toolbox
The final group in the SAR_Toolbox10b is Planning. This group contains 18 unique
tools. Seven of the tools use some method to assist planners in approximating the subject’s
location. These include Koester’s Statistical Search Area, Track Offset Model, Find Locations,
and Elevation Model (Koester 2008). One is from Doke’s Watershed Model (Doke 2012) and
another is Jacob’s Stream Interface finding (Jacobs 2015). The last model is Ferguson’s
Theoretical Search Area, which uses two tools. The first is a Cost Distance model that creates an
impedance or friction surface. This surface is then used by the Theoretical Search Area tool to
develop a least cost path distance surface from a given point grouped by time units. Other tools
include the Segment Search Speed tool, which estimates the team’s speed based on the friction
surface, a PSR (Probability of Success Rate) Estimate, Slope Analysis, Cell Coverage based on
entered cell tower data, and a Resource Estimation tool that approximates how many searchers
will be needed. Other features include QR code generation, many other forms, multiple pre-
established map formats, radio log, and personnel tracker to list some as captured in Figure 11
below.
22
Figure 11 IGT4SAR’s Planning toolbox
2.4.2. Mobility Model
The Cost Distance model and resulting Theoretical Search Area model are of primary
interest here. The Cost Distance model is basically an impedance layer calculated using multiple
underlying layers and map algebra. The map algebra adds overlapping cell values from input
layers to determine a cumulative impedance score. Cell values range from zero to 99, which
indicates no friction or impedance like a level road (0) to something that is impassable such as a
lake (99). The elevation surface is calculated by applying Tobler’s Hiking Function (Tobler
1991) to the DEM created slopes, both uphill and downhill, in order to obtain the raster’s
impedance values. Excessive slope is determined when it is greater than 60% and the impedance
is then valued at 99 or impassable. As seen in Table 2, a land cover impedance value is attributed
to each of the land cover types in the NLCD with open water deemed impassable at 99. Table 3
lists some examples of the federal codes for roads but the walking impedance is at zero for each,
although there could be impedance associated with a road due to its slope. The values for trails
23
are seen in Table 4. Streams and rivers are given a value based on the specified or calculated size
of the feature.
Table 2 Land cover impedance values
LAND COVER
CODE
DESCRIPTION IMPEDANCE
11 Open Water 99
12 Perennial Ice/Snow 80
21 Developed, Open Space 20
22 Developed, Low Intensity 20
23 Developed, Medium Intensity 30
24 Developed, High Intensity 40
31 Barren Land (Rock/Sand/Clay) 60
32 Unconsolidated Shore 70
41 Deciduous Forest 50
42 Evergreen Forest 50
43 Mixed Forest 50
51 Dwarf Scrub 75
52 Shrub/Scrub 75
71 Grassland/Herbaceous 50
72 Sedge/Herbaceous 50
73 Lichens 20
74 Moss 20
81 Pasture/Hay 20
82 Cultivated Crops 30
90 Woody Wetlands 80
91 Palustrine Forested Wetland 80
92 Palustrine Scrub/Shrub Wetland 80
93 Estuarine Forested Wetland 80
94 Estuarine Scrub/Shrub Wetland 80
95 Emergent Herbaceous Wetlands 80
96 Palustrine Emergent Wetland (Persistent) 80
97 Estuarine Emergent Wetland 80
98 Palustrine Aquatic Bed 99
99 Estuarine Aquatic Bed 99
24
Table 3 Examples of the federal codes for roads. CFCC is a Census Feature Class Code, MTFCC
is a 5-digit code assigned by the Census Bureau intended to classify and describe geographic
objects or features
CFCC MTFCC MTFCC_FC SPEED_LIMIT Walk_Imped
A00 S1100 Primary Road 40 0
A01 S1100 Primary Road 40 0
A02 S1100 Primary Road 40 0
A03 S1100 Primary Road 40 0
A04 S1100 Primary Road 40 0
A05 S1100 Primary Road 40 0
A06 S1100 Primary Road 40 0
A07 S1100 Primary Road 40 0
A08 S1100 Primary Road 40 0
A09 S1100 Primary Road 0 0
A10 S1100 Primary Road 65 0
A11 S1100 Primary Road 65 0
A12 S1100 Primary Road 65 0
Table 4 Example trails impedance values
TRAIL_CLASS DESCRIPTION IMPED
1 Minimal/Undeveloped Trail 25
2 Simple/Minor Development Trail 20
3 Developed/Improved Trail 15
4 Highly Developed Trail 5
5 Fully Developed Trail 0
The next step is creating the mobility model, specifically the Theoretical Search Area
model. The new piece of information needed here is the estimate of the subject’s expected
nominal walking speed. This is assuming the subject is walking across a flat surface with no
slope. The default Walking Speed entry is approximately three miles per hour. The number can
be adjusted for the environmental situation and the subject’s characteristics. The harsher the
conditions and less mobile the person is, the lower the Walking Speed entry. The estimate
entered will be the maximum speed used by the model with impeding terrain lowering the speed.
25
Part of the output from this model is a set of continuous points that all take the same
amount of travel time from the IPP. These points create a line, also known as isochrone rings,
surrounding the initial planning point (IPP). These rings are the polygon boundaries of what are
termed isochrone polygons. When visualizing the results, the impact on the isochrone polygons
of roads, streams, water bodies, and slope is evident, ultimately enlarging or reducing the
potential search area such as is shown in Figure 12. In the northwest portion of this map the
influence is demonstrated by the isochrones extending along roads and generally shrinking back
from increased slopes.
Figure 12 Example of terrain’s effect on isochrones and the Theoretical Search Area
26
2.4.3. Documentation
Although there has been substantial effort into documenting IGT4SAR, the template still
has room for improvement. Most tools have comments in the help area that are helpful.
Ferguson also attends major SAR events and offers free training where people can ask specific
questions and really see how the tools work although only so many trainings can be given each
year. There are 17 different portable document format (PDF) files and one README file that
are included in a documentation folder that is created when IGT4SAR is initially copied. These
documents cover detailed installation, administration, and explanation of various tools and
workflows. There are also six YouTube videos created by Ferguson totaling approximately one
and one-half hours (https://www.youtube.com/channel/UCrWNjhnpNOiEAATDzNw3lFg).
Even with all this information, the complexity of IGT4SAR is such that more references
would be helpful. An example of help information that is missing is how to save the initial base
data for a team’s area of responsibility into IGT4SAR. Simply having the information that is
currently available, systematically organized in one file, or at least linked would help in
understanding IGT4SAR.
2.5 Summary
This chapter reviewed the previous research related to search and rescue operations.
Koopman’s early work in search theory as well as the research by Syrotuck, Koester, and
Doherty have made major contributions to SAR operations. Search theory has been applied to
SAR operations although adoption is slow and data collection needs to be improved. While there
have been a number of studies done on mobility models in general, more work needs to be done
specific to SAR. A few mapping applications have been developed for search and rescue but
IGT4SAR really uses the power of GIS with its multitude of tools, specifically analysis tools.
27
The mobility model in IGT4SAR appears to be the first of its kind. The next chapter examines
the research process, different aspects of the data, and the procedures used in this research.
28
Chapter 3 Data and Methods
This chapter has four sections. Discussing the research process for the project is first. The second
part is an exploration of the data requirements. Data sourcing and condition is next with the final
section being about the project procedures and expected outcomes.
3.1 Research Process
As stated in the first chapter, the main objective of this research was to evaluate whether
a GIS-based mobility model tool is useful in search and rescue operations. In order to address
this, three sub-objectives were identified. Stated as questions, these are:
Is using a GIS-based SAR mobility model tool possible without having extensive training
in both GIS and SAR? This was designed to be a qualitative assessment. The availability
and usefulness of in-person training, tutorial videos, documentation, and hands-on time
with the tools were all part of this assessment.
Is the modeling process executable in a short timeframe with limited computing
resources? This was assessed throughout the process, noting the time the various tools
took to set up and execute.
Is the output from the mobility model accurate enough that it can assist the search
manager or incident command (IC) planners in determining an effective outer limit for
their search areas? Most of the work was involved in addressing this question.
The research strategy began in the extensive literature review coupled with discussions
with experienced SAR volunteers. It was at this stage that the first draft of the research objective
and questions formed. More detailed consultations with local search and rescue experts took
place following the initial literature review. Local SAR team members provided specific incident
29
information for evaluation. Familiarity with the developed SAR mobility model started during
the examination of these local incidents and gave an opportunity to begin to assess the level of
knowledge needed to operate such a model. Collection of a good deal of local GIS data sets
began as well as the conditioning of the data. After executing the initial runs of the model, the
results were reviewed. During this time, larger data sets were being researched and requested in
order to run the model on a larger sampling of incidents.
Following the investigation of the local incident data, a larger data set was chosen for the
main project. For computational reasons, reduction of its number of incidents occurred. Once this
paring down was complete, the next step was collecting the additional data sets required by the
IGT4SAR mobility model such as elevation, land cover, hydrology, and transportation lines.
More clipping and conditioning of the data occurred. To reduce the computational time, much
effort was put into the editing of Python scripts used in IGT4SAR tools. The model processed
through the selected incidents. Formatting and examination followed the exporting of the results
from the ArcMap table into Microsoft Excel. Figure 13 illustrates the overall project process.
30
Figure 13 Overall Research Process
3.2 Software Used
The software primarily used in this thesis was Esri’s ArcGIS 10.4 for Desktop, Microsoft
Excel, and PyScripter. Within the ArcGIS platform, ArcMap and its Spatial Analyst extension,
received the lion’s share of the work. The IGT4SAR template, built using Python scripting, was
used extensively. The more common tools used were the clip, project, merge, select by attribute
and location, some mosaicking, and XY to line. There was much less spreadsheet work done
compared to the amount done in ArcGIS as Microsoft Excel was mainly used to organize, sort,
31
format, and otherwise manipulate and evaluate the finished model output data. A few very basic
averaging and summing formulas were also used. The free PyScripter was used to edit IGT4SAR
to save processing time during the actual process of running the model on multiple incidents.
3.3 Data Required
The data used in this research fall into two major categories. The first is the underlying
environmental or base data sets that are used in the model. The other major data set was the
compilation of actual search and rescue incidents to be used to test the accuracy of the model.
Without either one of these data sets, the model cannot be properly evaluated. The more base
data sets that are incorporated, the more accurate the model will be, assuming the data sets are
accurate.
The foundational or base data sets that a comprehensive GIS-based mobility model for
SAR should heavily rely on are elevation, land cover, lines of travel, streams, and bodies of
water. Not only do these data need to be accurate but they also need to be available for large
areas in order to complete an analysis that covers a variety of topography. Lastly, for the
purposes of this study, the data sets need to be consistent, allowing comparison of an incident in
one area to another incident a great distance away. Information on administrative boundaries,
power lines, fence lines, and watersheds are also useful in IGT4SAR although only the power
and fence lines are used in its Theoretical Search model.
The SAR incident data sets are not well developed but they are starting to move in that
direction with low cost online portal solutions. Many other fields such as demography have a
strong coordinating body at the federal level and they appear to conduct more studies that
examine the field as a whole compared to SAR that has little research with unconsolidated data.
The US Census Bureau, an authoritative source, collects, conditions, and distributes key data that
32
demography and many other fields study. Those interested in traffic safety can search the
National Highway Traffic Safety Administration (NHTSA) or the Insurance Institute for
Highway Safety and find statistics. At this point, the SAR community does not have that sort of
national aggregator of data except for Koester’s ISRID, which is not publically available. The
Mountain Rescue Association (MRA) is also collecting data from their teams in a standardized
way at the national level. SAR data sets often rely on a volunteer entering the information about
each incident. This volunteer may not have been involved in the search and may not understand
the relevance of the data. These data sets can advance the science behind search and rescue and
provide justification for the continued and expanded resourcing of SAR efforts.
3.4 Data Used in the Model
Sourcing the data for the environmental side of the process went fairly well whereas the
search and rescue data set was somewhat more challenging. All the data used in this research
was free. All were available online except the SAR data set, which required a special request.
3.4.1. SAR Data Sets
Choosing the SAR incident data set was required before any base data could be collected.
Although there are not many different databases that could be researched, there are some. The
three sources considered were Yosemite Search and Rescue (YOSAR) data, Oregon’s Office of
Emergency Management (OEM) SAR database, and the ISRID as listed in Table 5. Four criteria
were used to select the data set were whether there was a sufficient number of incidents,
diversity in terrain, variety of subjects, and if it was previously studied.
33
Table 5 Comparison of SAR data sets
Data Set Number of
Incidents
Diverse
Terrain
Diverse
Subjects
Other Academic
Studies
Oregon state 4,447 Yes Yes Some
ISRID >50,000 Yes Yes Many
Yosemite National Park 2,308 Somewhat Somewhat Multiple
The YOSAR information was recently used in at least two graduate course projects and
the custodians of the data were very open to sharing their information. There is a good amount of
variety in the terrain although the extreme landscape of the Yosemite area made it less attractive.
Also, the park attracts more active users which led to less diversity in the subjects. The ISRID,
which is the largest consolidated SAR database of them all and has been the subject of a number
of studies, was developed and is owned by dbs Productions LLC and, while it is certainly shared,
it is not as accessible as other governmental or non-profit databases. In the end, the Oregon
database was chosen for further research since it best met the criteria. Incident points from this
database can be seen in Figure 14. It appears to have been studied in Jacobs’ work but otherwise
it has not been used as much as the ISRID or Yosemite data.
34
Figure 14 Oregon SAR Find Points, including out-of-state mutual aid operations
The Oregon database is extensive. It has a very large number of incidents recorded from
1998 to 2015 with most incidents recorded after 2010. There are a total of 4,477 subjects
recorded although multiple subjects can be associated with one SAR event. There are 52
different fields stored in their database as seen in Table 6. Oregon has a form that SAR teams
are required to fill out and it can be viewed in Appendix A.
35
Table 6 Oregon database fields and example data
Field Name Field Value Field Name Field Value
OEMNumber 2014-2267 HuntingGame No
IPPLat 45.6542 HuntingBirds No
IPPLong -121.64105 NordicSkier No
IPPAlt 2395 OtherSkier No
FindLat 45.6636072 Snowboarder No
FindLong -121.6458139 HorseRiding No
FindAlt 2633 Swimming No
IPPDist 3640 Powerboat No
MissingLand YES OtherWaterAct No
MissingWater No NonPowerboat No
RescueLand No MotorVehicle No
RescueWater No OtherAct No
Water No Suicide No
FalseInc No Wandering No
BodyRecovery No Criminal No
Other No VehicleTravel No
Snowmobile No Recovered Alive
Bicycle No Condition Well
Climbing No Sex Female
Hiking No Age 22
OtherWork No Group Female solo
Unknown No AdultChild No
MushroomPicker No ReportedDate 10/19/14 18:19
OtherPicker No FoundDate 10/19/14 18:52
OtherForestAct No Duration 0:33:00
Fishing No Narrative [See report for details.]
3.4.2. Base data
Following the SAR database selection, collection of base data sets began. Because the
data needed at least to span multiple counties, so that a large sampling was possible, sourcing
was somewhat limited to the state and federal levels in order to ensure compatibility throughout.
An important element is the quality of environmental GIS data covering the area of interest. The
USGS was the source for the elevation, which is a complete and accurate data set at the 10-meter
36
resolution. USGS also provided the land cover and hydrology layers. Both of these are the best,
freely available data sets for this area and are considered complete and accurate. The lines of
transportation, which run the range from major highways and unnamed dirt roads, was obtained
from the GIS Unit in the Oregon Department of Transportation. They compile files from the
many entities responsible for the roads in their particular jurisdiction. US Census Topologically
Integrated Geographic Encoding and Referencing (TIGER) data was considered as a source for
the transportation lines but it had much less detail. These data sets can be seen in Figure 15. All
datasets were projected to NAD 1983 Oregon Statewide Lambert Feet Intl and clipped to the
administrative boundaries of the two counties.
Figure 15 Base data for Lane and Deschutes Counties including roads and trails, streams and
water bodies, land cover, and elevation
Streams/Water Bodies
Elevation
Roads and Trails
Land Cover
37
Considerable conditioning of the base data took place. In order for IGT4SAR to function
properly, GIS data needed to be loaded into existing templated files so the model would know
where to pull its data from, exactly which fields would be there, and the fields’ names. This
required field matching. When the attributes or their data types did not match, creating new
columns and dictating attribute values using the field calculator occurred.
3.4.3. Conditioning the SAR Data
Conditioning of the Oregon SAR database took some more effort. Oregon OEM collects
their data in Microsoft Access. They exported the information to Microsoft Excel and emailed it
with the OEMNumber being the unique ID, or key, for each row of information. There were
4,447 rows of data, each representing one lost person. Not all of these rows contained the needed
data points so some sorting and cleaning was needed. The first sort removed all rows that did not
each have a value for IPP, Find Points, or recorded time. Also removed were incidents that were
not land-based and those that were marked as being vehicle travel. That brought the number of
rows down to 656. Further sorting led to selection of only incidents that had a time of 10 hours or
less. This was done in order to keep the extent of computed search areas from becoming too
large. Another deletion included removing multiple subjects involved in a single SAR incident.
This reduced group of incidents of 657 was then loaded into ArcMap to examine them
with regard to concentrations and diversity of topography. Two contiguous counties in western
and central Oregon contained a large number of incidents and have varying terrain. The Select by
Location geoprocessing tool was used to choose SAR incidents in Lane and Deschutes Counties,
creating a data set of 102. Finally, and importantly, to test the limits of the mobility model, the
incidents were sorted by distance from the IPP to the Find Points. The 51 incidents with the
38
greatest distance were then selected to challenge the model. The group was quickly reviewed to
ensure there was still diversity in terrain and subjects.
3.5 Variables Required for SAR Mobility Model Analysis
There were three major variables and a number of other significant variables involved in
the testing of the model in this thesis. The first major variable is the initial planning point (IPP)
which is the primary point the SAR team uses to plan their activities. The IPP is usually either
the last known point (LKP) of the subject or their point last seen (PLS). During an actual SAR
incident, only the IPP is required for the model to produce an output.
The next variable used in this analysis was the point where the subject was found. In
Oregon’s database they are stored in latitude and longitude in WGS84. Capturing this point
could be done any number of ways from a GPS reading to a post-operation map check so the
recorded Find Point is likely accurate to within 5-100 meters. Interestingly, this location does not
have a universal name but often is called the Find Point, Find Coordinates, Found Point, etc. In
this thesis, Find Point is the terminology used throughout. The IGT4SAR template can
incorporate these spatial variables in a variety of different geographic or projected coordinate
systems using formats from decimal degrees to military grid reference system.
The last major variable is the amount of time the subject was lost. There is an expectation
that the location data is accurate but the recorded time data does not reflect the time lost and can
be misleading. For these data sets, usually the clock starts upon notification of the authorities or
the SAR team, not when the person initially became lost. This is an important point because this
delay gives the subject more time to travel and if it is not accounted for, the search area may not
accurately represent the possible distance covered. The amount of time delay will vary greatly
39
depending on the subject’s typical pattern of life along with the reporting party’s level of
concern.
To evaluate the mobility model, the bare minimum attributes for these data sets are the
IPP (LKP or PLS), the Find Points, and time lost. If one of these three variables was missing,
evaluation of a SAR mobility model would be problematic at best. In addition, having good
information about the subject’s characteristics that relate to their speed traveled over a level
surface will make the model’s output more accurate and ultimately will add to the data set’s
usefulness in SAR research and resourcing.
Before beginning work with the Oregon incident database, two incidents from Nevada
County, California using local base data readily available were explored to understand the
problem set and gain familiarity with IGT4SAR. These incidents were used in the assessment of
required training and the processing time that was required. These were both real situations that
had fairly detailed information available. The first involved a person that was camping in an area
known to the subject. After entering the variables, the model’s outcome successfully gave an
outer limit of the search area with the subject found within the distance IGT4SAR modeled.
The second case was a day hiker separated from the group during the return portion.
Again, the model was successful as illustrated in Figure 16. The problem with these two
incidents was that their circumstances did not challenge the model. The SAR team ultimately
found the camper deceased close to the IPP. Nobody heard from the camper for a couple of days
so the model created a very large search area since the actual time lost time was unknown. In the
second incident the team found the person safe and the times recorded were generally accurate
but the distance the subject traveled was quite short. Thankfully, this second situation is
frequently the case but again, it did not push the model and, because of the variables in these
40
cases, the model output was much larger than many planners would find useful. These incidents
were important in gaining understanding of the data sets needed to run the model and their
required conditioning as well as the operation and capabilities of IGT4SAR itself.
Figure 16 Map showing the mobility model output. The day hiker’s IPP is in the center and
found location is the black diamond in a blue circle to the northeast. The hiker had been lost for
1.5 hours and the search area generated is about 15,000 acres. The team found him within the 0.5
hour polygon, which is just over 1,700 acres. This was a positive result as the subject was found
within the range of the model’s estimation.
3.6 IST4SAR Model Processes
After developing the research process, determining data requirements, sourcing the data,
and conditioning it, the process of running the model on the Oregon SAR data began. This
section describes the model’s typical workflow, changes that were made to the model for the
research, and the output.
41
3.6.1. IGT4SAR Mobility Model Workflow
IGT4SAR uses three Python scripted tools to create its mobility model. The first tool,
called Create New Incident, generates an incident folder that contains a geodatabase (GDB), map
file (MXD), and seven subfolders with various files that IGT4SAR uses to plan for and execute
search and rescue operations and administrative procedures. Variables required by this tool are
the folder’s location and name, coordinate system to use, and the type of form desired for task
assignments. Teams plan their search using these forms. Entering other information such as the
subject’s name, incident’s name and number, lead agency, and IPP information is optional. If it
were not entered here, it would have needed to be entered before the next tool is run. After
running this tool and opening the new MXD, the IGT4SAR template will be in place.
The Cost Distance model is the next tool needed in the procedure. Required fields allow
the user to identify the file workspace, choose the subject, choose the IPP, select the distance
units, and identify the DEM and NLCD data files. This tool creates an impedance raster, which is
the beginning of a least cost path analysis, for the area around the IPP. It takes into account
slope, land cover, transportation routes, streams, and water bodies (and fence and power lines if
available).
The last tool is the Theoretical Search model. This tool needs the workspace, subject,
IPP, distance units, subject’s estimated walking speed, and the DEM. The unique variable here is
the walking speed. This speed is assuming the subject is walking on unobstructed level ground at
their normal pace. The average adult generally walks at three miles per hour (Tobler 1991). This
variable is adjustable depending on environmental conditions or subject characteristics. The tool
creates polygons that are bounded by isochrones at ½ and one hour steps expanding from the
IPP. An isochrone is a line of points all having the same traveling time from a given location. At
this point, the model output is complete.
42
3.6.2. Changes Made to IGT4SAR for Throughput Improvement
With the data sets and variables used in this project, the process of running the three tools
and the three associated variable entries took almost 15 minutes to complete analysis for a single
incident. This timing is fine for a single incident but the computing time needed to run the large
number of sample incidents required for this project would be nontrivial. With this in mind,
effort went into combining the mobility model processes using Python coding so that incidents
could be run with minimal input. Because Ferguson naturally designed IGT4SAR to work one
incident at a time, there were challenges.
The Cost Distance tool had 670 lines of code and the Theoretical Search Area has 275
lines of code. Ultimately, it was possible to combine the two tools by editing the Python to
remove duplicate scripting and ensure the variable matched from one tool to the next. Another
piece in the process was to write Python script that hard-coded all of the required input variables,
except for the IPP. A portion of this revised script for variables is shown in Appendix B.
Working with the data began with an .xlsx file exported from Oregon’s OEM Microsoft
Access SAR database. This file contained critical incident data such as the IPP, Find Point, and
the search duration for multiple subjects. The Create New Incident tool was run once and it
produced a home folder containing six subfolders, an .mxd map file, and a .gdb geodatabase to
house the model output. This home folder stored all 51 of the model runs. The subject
information was loaded into the subject table in the home folder so they could individually be
selected to run through the model. Running the model required entering the subject’s unique
identification number into the newly combined tool in order to generate both Cost Distance and
Theoretical Search Area output files for the incident associated with that particular subject. The
file names needed to be changed so they were not overwritten during the next model run. By
using one incident home folder, combining the Cost Distance and Theoretical Search Area tools,
43
and hard-coding all variables but one, the process went from doing 3-4 incidents per hour and
having 9-12 interactions with the machine to processing 7-8 incidents per hour and only having
7-8 interactions.
3.6.3. Model Outputs Combined
After all 51 incident IPPs, Find Points, and mobility model outputs were on the map, one
could begin to see the success of the model. Unfortunately, after culling the data set down to 51
incidents and running them through the previously defined procedure, seven incidents needed
removal because there were database classification errors or the recorded numbers were very
improbable. The final number of incidents used in this analysis is 44.
The result of running all 51 incidents though the combined model tool was 51 impedance
rasters and 51 polygon shapefiles saved in the geodatabase. The impedance rasters buffered 10
miles out from the IPP or if not hard-coded, the distance specified by the user. The values range
from zero to 99 designating the calculated impedance. This raster was not placed in the map
document but was used to calculate the Theoretical Search Area. The shapefile generated by the
Theoretical Search Area was named MobilityModel and contained polygons bounded by
isochrones measured from the subject’s IPP. These isochrones begin in ½-hour increments and
increase to one-hour increments extending out to 12 hours. As shown in Figure 17, a number of
the events overlap because of people becoming lost in the same area over the more than 10 years
that the database has been housing SAR incidents.
44
Figure 17 IGT4SAR Theoretical Search Area results for all 51 incidents
One aspect that made visual interpretation of the combined model output difficult was the
number of data points in view and the overlapping nature of the paired locations. To improve the
visualization, the IPP and Find Points feature classes were used in the XY to Line tool in
ArcMap. This associated each IPP, or subject location, with its Find Point, using their
OEMNumber as the key. A sample is seen in Figure 18.
45
Figure 18 Using the XY to Line tool to associate IPPs with Find Points
Next, since the iterative process of the combined model could not label each set of output
isochrone polygons with the related incident number, it was necessary to devise a method to
extract the relevant isochrone polygon in which each Find Point fell. While it is possible to
conceive of an automated method, given the relatively small number to be extracted and the large
number of search area overlaps, it seemed wisest just to examine the map and visually identify
which isochrone polygon contained the Find Point for each incident.
The isochrone polygons were grouped in ½ hour bands up to the three-hour point and
then into one hour polygons. Because of this level of precision, a Find Point that plotted just
inside the polygon labeled five hours would be given the same value as a Find Point that plotted
at the outer edge of the five-hour polygon. To address this problem a precision adjustment,
subtracting half the polygon’s time equivalency, either 15 or 30 minutes, was introduced into the
46
table. The resulting polygon values were recorded in a spreadsheet that was initially populated
by exporting the table storing the Find Points feature class data from ArcMap and are seen in
Appendix B.
3.7 Model Assumptions and Considerations
While modifying and running the combined model with real incident data, a number of
assumptions were required in order to produce effective results. While the more general of these
are addressed in the last chapter, it is important at this point to highlight some of the key
assumptions and considerations that affect the results of this model and the subsequent analysis.
When people get lost, they usually wait to report their situation so they can try to find
their way. Likewise, when someone calls to report a friend or family member lost, there is
always some delay either because of lack of awareness or attempting to find the person. The
SAR database tracks the time of notification as the beginning of the incident time. To account for
this unknown delay time, three hours were added to each of the incidents’ time in order to
produce an estimate of each subject’s actual time of travel. This necessary adjustment will be too
long for some and not long enough for others but the expectation is that it will produce results
that are generally in the range of the true travel time. In Doke’s research on Yosemite SAR
incidents, he found that the median time missing was nine hours while the median time searching
was two hours (Doke 2012).
Another assumption used is that each of the subjects had an average walking speed of
three miles per hour. While this is the average determined by Tobler, IGT4SAR’s design allows
for variability in this number based on weather and other environmental conditions as well as the
age and other subject characteristics when performing the analysis on an individual incident.
Since this information was difficult to extract automatically from the source data, in order to
47
process the incidents in batch mode, it was necessary to assume the default speed across the set
of samples analyzed.
3.8 Summary
This chapter explained the research process and the overarching research objective of
determining whether a GIS-based mobility model could be useful in SAR operations. To carry
out this task Esri’s ArcGIS 10.4 was used as well as some Microsoft Excel. The data analyzed by
this software was divided into two groups, search and rescue incidents and base environmental
data. Both groups required conditioning and cleaning. To run an incident through IGT4SAR’s
mobility model the only incident variable required is the IPP. To assess the model’s utility other
variables beyond the IPP were needed, namely the Find Point and the time the subject was lost.
The template is designed to run at least three tools with different input in order to create the
model’s output. To reduce processing time for the assessment One incident folder was used to
hold all 51 incidents, the second two tools were combined, and most variables were hard-coded.
Adding a three-hour reporting delay time and defaulting all subjects to a three miles per hour
normal walking speed were two of the biggest assumptions made in the analysis.
In the next chapter, the results are explored in light of the expected outcomes and the
model’s overall success, general findings, and some unusual incidents are discussed.
48
Chapter 4 Results
This chapter explores the results and findings regarding all three of the sub-objectives.
Discussion is also included regarding the lost subjects from the study. Finally, some of the more
unusual subjects are addressed. Figure 19 shows the model results for the 51 incidents from the
original runs of the combined model.
Figure 19 IGT4SAR Theoretical Search Area results, IPPs, and Find Points for all 51 incidents
4.1 Model Results Findings
This section reviews the findings for each of the three sub-objectives. The first was
assessing if using a mobility model GIS tool is feasible without having extensive training. The
second was considering if the process is executable in a short timeframe with limited computing
resources. The last underlying sub-objective was determining if the model can reliably give the
incident command (IC) planners an effective outer limit for their search areas.
49
4.1.1. Training Required
The assessment for training needs was made after working with IGT4SAR on the two
Nevada County incidents. This was important because the tool was designed to work through
one incident at a time without having the Python script manipulated. The template has many
different capabilities beyond that of the mobility model. These other tools often require various
other data points which can make a user unsure of what different types of data are needed and
where they should be uploaded or input. Because of these types of capabilities and complexities
and the fact that it is a template within ArcGIS which is known as a challenging program,
IGT4SAR is not a solution that a SAR unit can start using day one. As with most software,
getting in and working through a problem is one of the best ways to learn the application. The
documentation, while extensive, still could be expanded. The training offered was invaluable as
well as the YouTube videos. After one user has learned IGT4SAR, that person would be capable
of teaching others.
4.1.2. Computation Time
The typical workflow to generate a mobility model in IGT4SAR included approximately
two minutes to create a new incident, four minutes for the impedance layer, and another five
minutes for the Theoretical Search Area. That was a total of approximately 11 minutes of
computational time with a minute for variable entry and general pause between tool execution.
That brought the total average time to about 14 minutes. This is a reasonable computational time,
especially considering the laptop used was 2 ½ years old. The planner can also accomplish other
tasks while the tools are running. The results dialogue boxes for the tools can be seen in
Appendix C.
50
4.1.3. Success of Modeled Travel Time
The complete table of individual results can be found in Appendix D. A sample of the
aggregated results is shown in Table 7. The success of the model is summarized in Table 8. Of
the incidents in the sample group and based on the unadjusted SAR response time, 30% were
found within the isochrones equaling the search time or less. When the generalized delay time
was incorporated, 75% of the Find Points were observed in any polygon whose time is
equivalent to or less than the search time.
Table 7 Example results. Green indicates modeled times that are greater than actual search time
and pink indicates modeled times that are less than actual.
Calculated results
Subject Number
1 2 3 4
Find to IPP Distance (feet) 26,232 20,809 18,098 21,811
Modeled Time Band 9:00:00 7:00:00 9:00:00 5:00:00
Center of Band Offset 0:30:00 0:30:00 0:30:00 0:30:00
Modeled Time (Center of Band) 8:30:00 6:30:00 8:30:00 4:30:00
Recorded Time 6:42:00 6:26:00 2:47:00 1:40:00
(Recorded Time) - (Modeled Time
[Center of Band])
-1:48:00 -0:04:00 -5:43:00 -2:50:00
Accounting for Measurement
Precision (1/2 of Band Value)
-1:18:00 0:26:00 -5:13:00 -2:20:00
Generalized Delay 3:00:00 3:00:00 3:00:00 3:00:00
(Recorded Time) - (Adjusted
Modeled Time) + (Gen Delay Time)
1:42:00 3:26:00 -2:13:00 0:40:00
Table 8 Success of results from the Theoretical Search Area tool
Scenario
Within Model
Area
Outside Model
Area
Recorded Time 30% 70%
Recorded Time Plus 3 hr Delay 75% 25%
51
Table 9 shows a summary of the results aggregated into quartiles according to distance
traveled. It can be seen that model more successfully covered the lost subjects who traveled
shorter distances.
Table 9 Aggregated success of results from the Theoretical Search Area tool. Success measured
using times including 3 hour delay.
Quartile of result set ordered by
distance traveled
Within Model
Area
Outside Model
Area
Quartile with the Longest Distance 45% 55%
Second Quartile 64% 36%
Third Quartile 91% 9%
Quartile with the Shortest Distance 100% 0%
4.2 Lost Subjects
On examination of the detailed information in the full database about these sampled
incidents, some interesting facts about the subjects and the performance of the model can be
detected. It is important to note that some SAR incident narratives have sensitive information
that is not publically released so the narrative column giving specific details and names was not
included. Those subjects that traveled five miles or more, the top seven, were quite diverse
subjects with ages ranging from 13 to 64, men accounting for 63% of the subjects, activities
varying from mushroom picking to running away from home, and different mental states from
dementia to intoxicated to suicidal. This variety in subject characteristics is seen throughout the
51 selected incidents. The average distance traveled was 3.3 miles, the average age 45, and 64%
of those lost were male. Alzheimer’s or dementia accounted for 18% of the incidents. Data on
the subjects can be found in Table 9. Thankfully, no deaths were recorded in the studied set of
Oregon data and only one had injuries.
52
Table 10 Subject characteristics
Age Average 45
Range 2 to 88
Gender Males 61%
Females 39%
Activity Hikers 31%
Pickers (mushroom, antlers) 13%
Hunters 7%
Other 9%
Wanderers 32%
Suicide 2%
Not Recorded 6%
Mental State Dementia, alcohol, suicide 27%
Distance Traveled Average 3.33 miles
Range 0.8 - 9.1 miles
Condition Found Well 98%
Group Solo 75%
Single Gender Group 9%
Mixed Group 16%
4.3 Unusual Incidents
A couple of subjects stood out from the rest. As addressed earlier, the subjects varied
greatly, as did their travel, and so is the case for the outliers. The person that moved the second
greatest distance of the 51 incidents selected and portrayed in Figure 20 was a trail runner.
Because he was a runner instead of someone walking and the time and distance attributed to him
were most unusual, he was not part of the final selection. The report stated he had been running
to a location about 3.5 miles away. He missed his destination and continued to run the trails
through the evergreen forest, as classified by the NLCD. The DEM indicates that he was
generally running downhill although there were many streams along his path. His run ended at a
bridge approximately 22 miles from his starting point. The 22 miles is a straight-line distance so
he likely ran a bit farther. The report stated that it took him four hours to run the distance and
while that is fast for most, if this 25-year-old was serious about running it is certainly possible.
53
Although a marathon is farther, just for reference, the median 2014 time for U.S. men’s
marathon finishers was 4 hours and 19 minutes (Running USA 2015).
Figure 20 Theoretical Search Area for extraordinary trail runner
Another interesting case was that of the person that went the third farthest in the group,
illustrated in Figure 21. This person was hiking, wandered off the trail, and became disoriented.
She was found five-and-one-half hours later walking down a road, having crossed at least three
other roads to get there. It was in December and she was found at 8:30 pm, 12 miles from her
starting point. Being 64 did not seem to slow her down. Fortunately she was found before the
colder hours since the temperature dropped down to 35F later that night (Weather Underground
2014).
54
Figure 21 Theoretical Search Area for hiker
4.4 Summary
This chapter covered the assessed results from the analysis and how they fit in with the
stated research objectives, information on subjects generally, and information of a few specific
unusual cases. It was determined that a fair amount of upfront training was required in order to
use the template effectively. The model was able to execute in a short enough time to keep it
from being a concern. As computing speed increases, it will be even less of a problem. The
finding with regard to the model’s success rate was that it performed well with the delay factor
included and as time continues, it will be even more successful as it moves down the list to the
people that have subjects that did not travel very far.
55
The last chapter concludes this report with a discussion of the model compared to the real
world. Limitations of this study as well as future work in the area are considered.
56
Chapter 5 Discussion and Conclusions
The first research question was whether learning how to effectively use this SAR mobility model
would require extensive training. After attending the SARGIS workshop, watching videos, and
working with the model, this question’s answer is generally no. Extensive training is needed to
get the system installed, loaded with base data, and knowing how to use it. Once that is done,
training others on a team would not be too difficult. The classroom training and online videos are
excellent, but ultimately understanding of both ArcGIS and SAR operations is needed. With all
this said, once the organization’s base data such as elevation, land cover, transportation lines,
streams, water bodies, fence and power lines have been projected and conditioned so they can be
loaded into the template, using IGT4SAR is much less difficult. This loading should be a one-
time event with information copied to subsequent computers as needed. Although learning how
to create the Theoretical Search Area using IGT4SAR is a challenge, learning the procedures is
not overly complicated and simply takes time and effort.
The next expected outcome was that the mobility model could be executed in 10 to 20
minutes with adequate computing capability. With the study complete, the finding is that this
outcome was met. As long as the area’s base data is already loaded and information on the
incident is known and available, an output of isochrone polygons can be available in less than 15
minutes. This is assuming the use of a computer with at least a processing speed of 2.5Mhz and
8GB of RAM. With a higher performance computer, times will be faster. Just because the model
can be produced that quickly does not mean that it will. The search manager may want the
planner to use IGT4SAR for other tasks such as creating quick initial search assignments prior to
generating the model. However, assignment of search areas is often the first priority so that
search teams can move to their search segment and begin looking for the subject.
57
An important finding is that time data in SAR is uncertain. It is an important variable to
understand in search and rescue. Accurately portraying time is useful in both advancing SAR
science as well as justifying resourcing for search and rescue efforts. The results from this study
definitely would have a higher degree of certainty if the time lost were recorded along with the
time that authorities were involved.
The last finding came from an interview with an experienced SAR planner. The
perception is that the mobility model gives too large of a search area for it to be useful, which is
connected to the last expected outcome. The intermediate stage of an impedance layer was of
interest to the planner though, see Figure 22. This would give the planners some indication of
which direction the subject could not go without running into difficult or which direction a
subject may find easier to travel. With this information, the team could create search segments
accordingly. Some areas are very dense or steep and therefore would be divided into smaller
segments as opposed to a flat open area that could be searched quickly allowing the team to be
assigned to a larger area. It could also provide potential directions of travel if the impedance
layer showed a high value that might channel the lost subject’s route choice.
Figure 22 Comparing National Geographic Society topographic map to the IGT4SAR generated
impedance layer for a portion of Deschutes County
58
The major expected outcome is somewhat predicated on the first two. If learning how to
work with the model was too difficult or it took too much computing power or time to be useful,
knowing whether it was effective or not would not be as important. Although the learning curve
was steeper than expected, the process can be mastered and the times are reasonable. The
expectation was that all of the search and rescue incidents would be found within the model’s
appropriate isochrone polygon. This did not turn out to be the finding. Importantly, Ferguson
points out that he did not intend the output to dictate where the team should look but that it
would be an aid to search manager and planners in the decision-making process. The broader
impact discussion takes place in the last chapter. The final expected outcome was not met either.
Interestingly, of the three expected outcomes, only one was realized.
5.1 The Model and Reality
Determining what is reality is the biggest cause for uncertainty when comparing the
model to real-world events. The biggest question is how long the person has been lost. SAR
databases often just record the time authorities were notified and the time the search was
successful or suspended. The other important factors in correlating the model and reality are the
initial planning point used and the find point. The IPP, point last seen or last known point, may
have a greater level of uncertainty because the person reporting them missing may only have a
educated guess as to the point to start the search from. The find point typically involves SAR
personnel on scene that will record the location while standing there. These locations are
generally recorded using a GPS devise but if not, SAR personnel tend to have good recall of their
physical surroundings and landmarks.
The model performed differently than expected with regard to reality, even when times
were adjusted by a set delay constant. One of the expected outcomes was that the model was
59
going to greatly overestimate the furthest extent of the lost person’s travels. This was not the
case as one-quarter of the Find Points fell outside the model’s prediction. This was a positive
result. If the model search areas contained all find points, including incidents that were above the
median distance travelled, it would simply be estimating a very large area with all points
enveloped.
The other expected outcome is that all lost subjects would be found within the model’s
estimate. This outcome was not seen, as demonstrated again by the fact that one-quarter of the
lost subjects were beyond the model’s estimate. It is important to remember that these SAR
incidents were selected to challenge the model when testing whether all incidents would be
within the predicted search area. With some of the incidents having greater distances traveled
and not falling within the correct isochrones, it could be argued that the model’s output could be
useful in the determination of a largest POA scenario. On the other hand, if none fell outside the
projected distances, it could not be determined if the model produces a result that was even close
to reality since it could grossly be overestimating the furthest point traveled.
When considering the incidents with shorter distances traveled, the model more
successfully covered the lost subjects. This can be observed when looking at the group of
incidents with the shortest distance traveled listed in Appendix D. This analysis focused on the
44 furthest travelers, 51 original minus misclassified and extreme outliers, in the set of 102 land-
based incidents within Lane and Deschutes Counties contained in the original database obtained
for this project. Having selected the incidents with the furthest distance traveled, it seems likely
that the rest of the land-based Lane and Deschutes Counties incidents would have a much higher
rate of being captured in the correct isochrone.
60
Overall, the assessment is that this model can be helpful to search managers and their
planners. For the majority of the cases, the lost person was found within the modeled isochrone.
That said, the mobility model does not include all sampled incidents demonstrating that is does
not necessarily grossly overestimate a lost person’s distance traveled. Importantly, it must be
remembered that all assessments are based on the critical assumption that the average delay from
when the subject became lost until they were reported lost averaged approximately three hours.
5.2 Limitations and Hindsight
If a model uses poor data, its results will be poor, so the accuracy of data is extremely
important. Models do not reflect reality perfectly, but with a good model and accurate data, they
are useful. The amount of time the subject was lost before he or she was reported lost is the
biggest limitation in assessing the success of this model. As the search and rescue field matures,
hopefully accurate and complete data collection will as well.
It was a little surprising to discover that the IGT4SAR mobility model did not use subject
characteristic information, such as age and gender, nor what has been learned regarding lost
person behavior. These variables are collected and are used in IGT4SAR’s tools incorporating
Koester’s models, but they are not used in the Theoretical Search Area model. Lost person
behavior statistics may be worth incorporating into the mobility model somehow or at least
giving the user an option for such a hybrid.
From the perspective of a non-profit organization as SAR teams are structured, cost of
software might be considered a limitation. Most consider Esri’s ArcGIS to be expensive software
with commercial starting at $1500. However, nonprofits do not pay for the software but only
need to pay a small annual administrative fee (Esri 2016). Microsoft Excel is included on many
computers and can be purchased for $110 (Microsoft 2016), although it was required for the
61
evaluation but not the model generation itself and the same operations could have been done free
on OpenOffice or Google Sheets.
Looking back there are a number of things that could have been done differently in this
study. After doing further research, it appears Syrotuck’s statistical research has not been
integrated into a GIS. Building a tool that incorporates his work could have been a good project
to undertake to advance SAR efforts.
Another option on how this project could have been done differently would be to have
limited the incidents not just to land-based events but to hikers exclusively as Doherty did in his
Yosemite mobility model research (Doherty et al. 2014). If done at a statewide level, this may
have created a more homogeneous data set that may have produced more consistent relationships
between Find Point and theoretical search areas. This would have at least controlled one variable
to give a more precise assessment of the model although it would have required more data and
conditioning.
5.3 Future Work
This area of study is ripe for investigating the connections between search and rescue and
geographic information science. Lin and Goodrich developed a model that allows for expert
opinion as stated in their 2010 article. Because Ferguson has made his template available on
GitHub, someone adding on features that would allow for such input could prove to be very
helpful. Another technical point for future work would be to include not just concepts from Lin
and Goodrich but also Syrotuck, Jacobson and other models and research into IGT4SAR’s
mobility model. This could combine the findings from Jacobs, Syrotuck, and/or Doke to
potentially reduce the POA while increasing the success rate.
62
There are two other, less technical and not necessarily GIS related, areas that could be
explored in the future. Researching how people’s trust of output from computers influence their
decision making could have an impact on SAR operations. Some people are inclined to believe
an estimate generated by a computer more than their own experience. In these cases, it may be
better to have no estimate than a largely incorrect one. A second option would be to work on
adding to the existing documentation for IGT4SAR. There is already much work that has been
done and a potentially quick improvement would be to simply combine the existing help files.
5.4 Conclusion
IGT4SAR is an extremely capable template, full of a multitude of useful tools to plan for
and execute search and rescue operations. The challenge with having so much functionality in an
application is that it can often bring an interface that is complex to use. This creates a
requirement for a fair amount of training and hands-on practice with the template. Once an
individual planner or group of planners have some training and experience with IGT4SAR, the
overall SAR operation and administration should run much more effectively and efficiently. The
Theoretical Search Area model processing time is short enough that it can be used during actual
events. In many situations the mobility model may produce an area too large to be used by itself,
but it can serve as an aid to planners as the operation continues. Ferguson did not intend for his
tool to dictate search decisions but to be an aid to decision makers. Making an impedance layer
covering the entire area the team is responsible for, which is created as an intermediate step,
available independently would be helpful as a planning tool so they can assign search segments
accordingly. Regarding this study’s main objective, IGT4SAR’s mobility model can be useful in
search and rescue operations but should be used as a support and not as a directive.
63
References
Adams, Annette L, Terri A Schmidt, Craig D Newgard, Carol S Federiuk, Michael Christie, Sean
Scorvo, and Melissa DeFreest. 2007. “Search Is a Time-Critical Event: When Search and
Rescue Missions May Become Futile.” Wilderness & Environmental Medicine 18 (2): 95–
101.
Adriaensen, F., J. P. Chardon, G. De Blust, E. Swinnen, S. Villalba, H. Gulinck, and E.
Matthysen. 2003. “The Application of ‘Least-Cost’ Modelling as a Functional Landscape
Model.” Landscape and Urban Planning 64 (4): 233–47. doi:10.1016/S0169-
2046(02)00242-6.
Cooper, Donald C., John R. Frost, and R Quincy Robe. 2003. Compatibility of Land SAR
Procedures with Search Theory. Alexandria: Potomac Management Group, Inc.
http://www.uscg.mil/hq/cg5/cg534/nsarc/LandSearchMethodsReview.pdf.
Doherty, Paul, Qinghua Guo, and Otto Alvarez. 2011. “Expert versus Machine : A Comparison
of Two Suitability in Yosemite National Park.” The Professional Geographer 0124
(November): 1–16. doi:10.1080/00330124.2012.697857.
Doherty, Paul J. 2013. “The Applications of GISystems to Wilderness Search and Rescue, an
Overview within a GIScience Context and Examples from Yosemite National Park.” PhD
diss., University of California Merced.
Doherty, Paul J. 2015. “SARGIS7 Report.” WiSAR and GIS Blog. November 24, 2015.
http://wisarandgis.blogspot.com/2015/11/sargis7-report.html.
Doherty, Paul J., Qinghua Guo, Wenkai Li, and Jared Doke. 2014. “Space-Time Analyses for
Forecasting and Understanding Future Incident Occurrence: A Case-Study from Yosemite
National Park Using the Presence and Background Learning Algorithm.” International
Journal of Geographical Information Science. 28:5, 910-927, doi:
10.1080/13658816.2014.890202
Doherty, Paul J., Quinghua Guo, Jared Doke, and Don Ferguson. 2014. “An Analysis of
Probability of Area Techniques for Missing Persons in Yosemite National Park.” Applied
Geography 47 (February): 99–110. doi:10.1016/j.apgeog.2013.11.001.
Doke, Jared. 2012. “Analysis of Search Incidents and Lost Person Behavior in Yosemite
National Park.” Masters thesis, University of Kansas.
Durkee, George, and Vanessa Glynn-Linaris. 2012. Using GIS for Wildland Search and Rescue.
Redlands: Esri.
Esri. 2016. “Nonprofit Organization - What’s Included.” Accessed March 21.
http://www.esri.com/nonprofit/whats-included.
Ferguson, Don. 2008. “GIS for Wilderness Search and Rescue.” Presented at the Esri Federal
User Conference, Washington DC. February 20-22.
Goodrich, Michael A., Bryan S. Morse, Damon Gerhardt, Joseph L. Cooper, Morgan Quigley,
Julie A. Adams, and Curtis Humphrey. 2008. “Supporting Wilderness Search and Rescue
Using a Camera-Equipped Mini UAV.” Journal of Field Robotics 25 (1-2): 89–110.
doi:10.1002/rob.20226.
64
Jacobs, Matt. 2015. Terrain Based Probability Models for SAR. San Diego: Mountain Rescue
Association. http://www.mra.org/member-services/grant-information/current-grants#terrain-
models.
Koester, Robert J. 2008. Lost Person Behavior. Charlottesville: dbS Productions LLC.
Koopman, Bernard Osgood. 1946. “Search and Screening.” Report No. 56 of the Operations
Evaluation Group, no. 56: 172.
Lin, Lanny, and Michael A. Goodrich. 2010. “A Bayesian Approach to Modeling Lost Person
Behaviors Based on Terrain Features in Wilderness Search and Rescue.” Computational
and Mathematical Organization Theory 16 (3): 300–323. doi:10.1007/s10588-010-9066-2.
Lovelock, David. 2008. POC Is Not POA. Olympia: International SAR Alliance.
http://www.isaralliance.com/general-library/download/53
Mattson, Robert. 1980. Search Theory and Application. In "NATO Conference Series. 1st ed."
Edited by Brian Haley and Lawrence D. Stone. 87-92. New York: Plenum Press.
Pedder, John. 2012. “MapSAR—Life-Saving Maps.” Presented at Esri 2012 GeoDesign
Summitt. Redlands, California. January 5-6.
Pfau, Loren D. 2013. “Wilderness Search and Rescue: Sources and Uses of Geospatial
Information.” Master's thesis, Pennsylvania State University.
Phillips, Ken, Maura J Longden, Bil Vandergraff, William R Smith, David C Weber, Scott E
McIntosh, and Albert R Wheeler. 2014. “Wilderness Search Strategy and Tactics.”
Wilderness & Environmental Medicine 25 (2). Elsevier: 166–76.
doi:10.1016/j.wem.2014.02.006.
Pingel, Thomas J. 2010. “Modeling Slope as a Contributor to Route Selection in Mountainous
Areas.” Cartography and Geographic Information Science 37 (2): 137–48.
doi:10.1559/152304010791232163.
Ray, Sara. 2015. “Chris and I Are Alive Because of You - Deschutes County Search and
Rescue.” http://deschutessearchandrescue.org/testimonials/chris-and-i-are-alive-because-of-
you/.
Rose, Caroline. 2015. “Mapping Technology in Wilderness Search and Rescue.” Master's thesis,
University of Wisconsin-Madison.
Running USA. 2015. 2014 Running USA Annual Marathon Report. Los Angeles: Running USA.
http://www.runningusa.org/marathon-report-2015.
Sava, Elena, Charles Twardy, Robert Koester, and Mukul Sonwalkar. 2015. “Evaluating Lost
Person Behavior Models.” Transactions in GIS 20: 38–53. doi:10.1111/tgis.12143.
Solmaz, Gürkan, Mustafa Ilhan Akbaş, and Damla Turgut. 2015. “A Mobility Model of Theme
Park Visitors.” IEEE Transactions on Mobile Computing 14 (12): 2406–18.
doi:10.1109/TMC.2015.2400454.
SSARCC. 2016. SAR Coordinators Informational Resources. Centreville: State Search and
Rescue Coordinators Council (SSARCC).
http://www.ssarcc.com/index.php?option=com_content&task=view&id=41&Itemid=1.
Syrotuck, W.G. 1976. Analysis of Lost Person Behavior. 2012 ed. Mechanicsburg: Barkleigh
Productions, Inc.
65
Tobler, Waldo. 1993. “Three Presentations on Geographical Analysis and Modeling: Non-
Isotropic Geographic Modeling; Speculations on the Geometry of Geography; and Global
Spatial Analysis.” Santa Barbara: National Center for Geographic Information and
Analysis.
Twardy, Charles R, Michael Goodrich Byu, Robert J Koester, Eric Cawi, Lanny Lin Byu, Elena
Sava. 2011. “MapScore : A Portal for Scoring Probability Maps.” http://sarbayes.org/wp-
content/uploads/2012/08/MapScorePoster_MORSS2012.pdf.
Vogt, Ryan, Ioanis Nikolaidis, and Pawel Gburzynski. 2012. “A Realistic Outdoor Urban
Pedestrian Mobility Model.” Simulation Modelling Practice and Theory 26. Elsevier B.V.:
113–34. doi:10.1016/j.simpat.2012.04.006.
Weather Underground. 2014. “Weather History for Eugene, OR.” Accessed April 10.
https://www.wunderground.com/history/airport/KEUG/2014/12/7/DailyHistory.html?req_ci
ty=Lorane&req_state=OR&req_statename=Oregon&reqdb.zip=97451&reqdb.magic=1&re
qdb.wmo=99999.
66
Appendix A: Oregon’s Incident Data Collection Form
67
Appendix B: Python editing of IGT4SAR - changing all variables, except
SubNum (Subject Number), from arcpy.GetParameterAsText(0) lines to
hard-coded values
# Main Program starts here
if __name__ == '__main__':
#in_fc - this is a point feature used to get the latitude and longitude of point.
mxd, df = getDataframe()
SubNum = arcpy.GetParameterAsText(0) # Get the subject number
if SubNum == '#' or not SubNum:
SubNum = "1" # provide a default value if unspecified
wrkspc = "C:\Users\CIP Assessor\Desktop\ORData\FinalRun\SAR_Default.gdb"
arcpy.AddMessage("\nCurrent Workspace" + '\n' + wrkspc + '\n')
env.workspace = wrkspc
ippType = "LKP" # Determine to use PLS or LKP
TheoDist = "10"
if TheoDist == '#' or not TheoDist:
TheoDist = "0" # provide a default value if unspecified
bufferUnit = "miles" # Desired units
if bufferUnit == '#' or not bufferUnit:
bufferUnit = "miles" # provide a default value if unspecified
uSeStr= "true" # Desired units
if uSeStr == '#' or not uSeStr:
uSeStr = "true" # provide a default value if unspecified
DEM2 = "C:\Users\CIP Assessor\Desktop\ORData\ORprojected.gdb\DEM10m"
if DEM2 == '#' or not DEM2:
# DEM2 = "DEM" # provide a default value if unspecified
arcpy.AddMessage("You need to provide a valid DEM")
NLCD = "C:\Users\CIP Assessor\Desktop\ORData\ORprojected.gdb\NLCD"
if NLCD == '#' or not NLCD:
# NLCD = "NLCD" # provide a default value if unspecified
NLCD = "empty"
deSiredSpdA = "3" # Nominal walking speed
if deSiredSpdA == '#' or not deSiredSpdA:
deSiredSpdA = "3.0" # provide a default value if unspecified
68
Appendix C: Tool Results from Mobility Model Run
Detailed results window shows 1:50 minutes to process the Create an Incident tool
69
Detailed results window shows 4:05 minutes to process the Cost Distance model tool
Detailed results window shows 4:52 minutes to process the Theoretical Search Area tool
70
Appendix D: Results Sorted by Distance Traveled (green indicates modeled
correctly and pink indicates an incorrectly modeled incident)
Subject Number 1 2 3 4 5 6 7 9 11 12 13
Find to IPP
Distance (feet)
26,232 20,809 18,098 21,811 38,016 18,101 45,443 25,399 24,782 24,054 34,863
Modeled Time
Band
9:00:00 7:00:00 9:00:00 5:00:00 10:00:00 5:00:00 12:00:00 11:00:00 7:00:00 8:00:00 11:00:00
Center of Band
Offset
0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00
Modeled Time
(Center of Band)
8:30:00 6:30:00 8:30:00 4:30:00 9:30:00 4:30:00 11:30:00 10:30:00 6:30:00 7:30:00 10:30:00
Recorded Time 6:42:00 6:26:00 2:47:00 1:40:00 0:50:00 1:08:00 3:15:00 3:16:00 1:58:00 9:41:00 1:49:00
(Recorded Time) -
(Modeled Time
[Center of Band])
-1:48:00 -0:04:00 -5:43:00 -2:50:00 -8:40:00 -3:22:00 -8:15:00 -7:14:00 -4:32:00 2:11:00 -8:41:00
Accounting for
Measurement
Precision (1/2 of
Band Value)
-1:18:00 0:26:00 -5:13:00 -2:20:00 -8:10:00 -2:52:00 -7:45:00 -6:44:00 -4:02:00 2:41:00 -8:11:00
Generalized Delay 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00
(Recorded Time) -
(Adjusted
Modeled Time) +
(Gen Delay Time)
1:42:00 3:26:00 -2:13:00 0:40:00 -5:10:00 0:08:00 -4:45:00 -3:44:00 -1:02:00 5:41:00 -5:11:00
Subject
Recovered
Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive
Subject Condition Well Well Well Well Well Well Well Well Well Well Well
Gender Male Male Female Female Male Male Female Male Female Male Female
Age 80 49 64 50 62 80 47 61 15 67 34
Alcohol Yes
Dementia-
Alzheimer's
Yes Yes
Suicide
Other
71
Subject Number 15 16 18 19 20 21 22 23 24 26 27 28
Find to IPP
Distance (feet)
19,727 28,170 19,287 17,466 33,875 44,356 48,083 19,762 16,207 16,002 15,662 15,513
Modeled Time
Band
5:00:00 7:00:00 8:00:00 4:00:00 9:00:00 14:00:00 13:00:00 5:00:00 7:00:00 10:00:00 9:00:00 6:00:00
Center of Band
Offset
0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00 0:30:00
Modeled Time
(Center of Band)
4:30:00 6:30:00 7:30:00 3:30:00 8:30:00 13:30:00 12:30:00 4:30:00 6:30:00 9:30:00 8:30:00 5:30:00
Recorded Time 3:53:00 6:30:00 8:06:00 3:30:00 2:48:00 4:57:00 5:26:00 9:38:00 4:42:00 8:22:00 3:51:00 3:35:00
(Recorded Time) -
(Modeled Time
[Center of Band])
-0:37:00 0:00:00 0:36:00 0:00:00 -5:42:00 -8:33:00 -7:04:00 5:08:00 -1:48:00 -1:08:00 -4:39:00 -1:55:00
Accounting for
Measurement
Precision (1/2 of
Band Value)
-0:07:00 0:30:00 1:06:00 0:30:00 -5:12:00 -8:03:00 -6:34:00 5:38:00 -1:18:00 -0:38:00 -4:09:00 -1:25:00
Generalized Delay 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00
(Recorded Time) -
(Adjusted
Modeled Time) +
(Gen Delay Time)
2:53:00 3:30:00 4:06:00 3:30:00 -2:12:00 -5:03:00 -3:34:00 8:38:00 1:42:00 2:22:00 -1:09:00 1:35:00
Subject
Recovered
Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive
Subject Condition Well Injured Well Well Well Well Well Well Well Well Well Well
Gender Male Male Male Male Male Male Female Female Male Female Female Female
Age 9 57 48 82 13 37 64 64 44 24 58 16
Alcohol
Dementia-
Alzheimer's
Yes Yes
Suicide Yes
Other
72
Subject Number 29 30 31 32 33 34 35 36 37 38 39 40
Find to IPP
Distance (feet)
13,209 13,111 13,000 11,744 11,098 11,089 10,923 10,643 9,915 9,049 8,278 7,455
Modeled Time
Band
6:00:00 4:00:00 5:00:00 6:00:00 3:00:00 6:00:00 5:00:00 3:00:00 3:00:00 5:00:00 2:00:00 3:00:00
Center of Band
Offset
0:30:00 0:30:00 0:30:00 0:30:00 0:15:00 0:30:00 0:30:00 0:15:00 0:15:00 0:30:00 0:15:00 0:15:00
Modeled Time
(Center of Band)
5:30:00 3:30:00 4:30:00 5:30:00 2:45:00 5:30:00 4:30:00 2:45:00 2:45:00 4:30:00 1:45:00 2:45:00
Recorded Time 1:00:00 1:04:00 1:45:00 3:45:00 2:00:00 5:42:00 1:58:00 1:45:00 3:03:00 2:44:00 0:49:00 2:06:00
(Recorded Time) -
(Modeled Time
[Center of Band])
-4:30:00 -2:26:00 -2:45:00 -1:45:00 -0:45:00 0:12:00 -2:32:00 -1:00:00 0:18:00 -1:46:00 -0:56:00 -0:39:00
Accounting for
Measurement
Precision (1/2 of
Band Value)
-4:00:00 -1:56:00 -2:15:00 -1:15:00 -0:30:00 0:42:00 -2:02:00 -0:45:00 0:33:00 -1:16:00 -0:41:00 -0:24:00
Generalized Delay 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00
(Recorded Time) -
(Adjusted
Modeled Time) +
(Gen Delay Time)
-1:00:00 1:04:00 0:45:00 1:45:00 2:30:00 3:42:00 0:58:00 2:15:00 3:33:00 1:44:00 2:19:00 2:36:00
Subject
Recovered
Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive Alive
Subject Condition Well Well Well Well Well Well Well Well Well Well Well Well
Gender Female Female Male Female Male Female Male Male Female Male Male Male
Age 46 79 72 25 74 36 43 8 13 31 79 88
Alcohol
Dementia-
Alzheimer's
Yes Yes Yes Yes
Suicide
Other
73
Subject Number 41 42 43 44 45 46 47 49 50
Find to IPP
Distance (feet)
6,792 6,722 6,621 6,442 6,328 5,484 5,410 4,700 4,662
Modeled Time
Band
1:30:00 2:00:00 2:30:00 2:30:00 2:30:00 2:00:00 1:30:00 2:00:00 2:00:00
Center of Band
Offset
0:15:00 0:15:00 0:15:00 0:15:00 0:15:00 0:15:00 0:15:00 0:15:00 0:15:00
Modeled Time
(Center of Band)
1:15:00 1:45:00 2:15:00 2:15:00 2:15:00 1:45:00 1:15:00 1:45:00 1:45:00
Recorded Time 0:31:00 0:42:00 7:20:00 0:16:00 1:27:00 1:46:00 2:33:00 6:03:00 7:00:00
(Recorded Time) -
(Modeled Time
[Center of Band])
-0:44:00 -1:03:00 5:05:00 -1:59:00 -0:48:00 0:01:00 1:18:00 4:18:00 5:15:00
Accounting for
Measurement
Precision (1/2 of
Band Value)
-0:29:00 -0:48:00 5:20:00 -1:44:00 -0:33:00 0:16:00 1:33:00 4:33:00 5:30:00
Generalized
Delay
3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00 3:00:00
(Recorded Time) -
(Adjusted
Modeled Time) +
(Gen Delay Time)
2:31:00 2:12:00 8:20:00 1:16:00 2:27:00 3:16:00 4:33:00 7:33:00 8:30:00
Subject
Recovered
Alive Alive Alive Alive Alive Alive Alive Alive Alive
Subject Condition Well Well Well Well Well Well Well Well Well
Gender Male Male Female Male Male Male Female Male Male
Age 9 2 26 76 28 55 7 23 45
Alcohol
Dementia-
Alzheimer's
Yes
Suicide
Other Yes
Abstract (if available)
Abstract
Every year thousands of people become lost or injured to the extent that a search and rescue (SAR) unit needs to step in and help. Through the ages, we have needed to look for people and things yet the theory behind searching goes back less than 75 years to World War II. The main idea is that to be successful, searchers need to search the right area, and be able to detect the person or thing. This research explored the utility of using a GIS-based mobility model to assist search planners in developing their search areas. A mobility model incorporates consideration of the speed with which a person can move across the landscape. The tool used here is an Esri ArcGIS template called Integrated Geospatial Tools for Search and Rescue (IGT4SAR). While it includes many SAR tools, this research focused on the mobility analysis component. This study specifically assessed IGT4SAR’s ease of use, speed, and success rate at determining how far a person can travel in a given time. Nevada County provided detailed information on a few incidents used to gain familiarization with IGT4SAR and the state of Oregon provided a large database of historical and diverse SAR events that allowed for broader testing of the model. Ultimately, 44 incidents were used to test the model. The model itself is easy to use, but the template is complex. With preloaded data, the model creates a product in less than 15 minutes. Starting with an unrealistic assumption that the incident start time recorded in the database represented the time when the subject left the last known location, test runs resulted in a 30% success rate where the found location fell in a time band that was less than the amount of time between the start time and the found time recorded in the database. After adding an estimated three-hour delay in reporting time to the SAR notification times the model had a 75% success rate. These results suggest that IGT4SAR can assist in defining a containment area to limit a search radius and is worthy of continued development.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Applying least cost path analysis to search and rescue data: a case study in Yosemite National Park
PDF
A spatial and temporal exploration of how satellite communication devices impact mountain search and rescue missions in California’s Sierra Nevada mountain range
PDF
The movement of Mexican migration and its impact based on a GIS geospatial database
PDF
Evaluating predator prey dynamics and site utilization patterns of golden eagles using resource selection modeling and spatiotemporal pattern mining
PDF
Development and evaluation of a USV based mapping system for remote sensing of eelgrass extent in southern California
PDF
Visualizing historic space through the integration of geographic information science in secondary school curriculums: a comparison of static versus dynamic methods
PDF
Using building information modeling with augmented reality: visualizing and editing MEP systems with a mobile augmented reality application
Asset Metadata
Creator
Johnson, Mark Powell
(author)
Core Title
Evaluating the utility of a geographic information systems-based mobility model in search and rescue operations
School
College of Letters, Arts and Sciences
Degree
Master of Science
Degree Program
Geographic Information Science and Technology
Publication Date
07/21/2016
Defense Date
05/09/2016
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
digital map,electronic map,friction layer,geographic information systems,geospatial,GIS,GISc,GIScience,GIST,IGT4SAR,impedance layer,Integrated Geospatial Tools for Search and Rescue,lost,lost person behavior,LPB,Map,mapping,mobility model,OAI-PMH Harvest,POA,POC,POS,probability of area,probability of containment,probability of success,SAR,search,search and rescue,Spatial Sciences Institute,SSI,wilderness search and rescue,WiSAR
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Kemp, Karen K. (
committee chair
), Lee, Su Jin (
committee member
), Loyola, Laura C. (
committee member
)
Creator Email
mark.p.johnson@alumni.usc.edu,markpjoh@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-270460
Unique identifier
UC11281474
Identifier
etd-JohnsonMar-4558.pdf (filename),usctheses-c40-270460 (legacy record id)
Legacy Identifier
etd-JohnsonMar-4558.pdf
Dmrecord
270460
Document Type
Thesis
Format
application/pdf (imt)
Rights
Johnson, Mark Powell
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
Tags
digital map
electronic map
friction layer
geographic information systems
geospatial
GIS
GISc
GIScience
GIST
IGT4SAR
impedance layer
Integrated Geospatial Tools for Search and Rescue
lost person behavior
LPB
mapping
mobility model
POA
POC
POS
probability of area
probability of containment
probability of success
SAR
search
search and rescue
Spatial Sciences Institute
SSI
wilderness search and rescue
WiSAR