University of Southern California Dissertations and Theses

A spatial and temporal exploration of how satellite communication devices impact mountain search and rescue missions in California’s Sierra Nevada mountain range

(USC Thesis Other)

A spatial and temporal exploration of how satellite communication devices impact mountain search and rescue missions in California’s Sierra Nevada mountain range

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

A Spatial and Temporal Exploration of How Satellite Communication Devices Impact Mountain
Search and Rescue Missions in California’s Sierra Nevada Mountain Range

by

Sophia M. Recca

A Thesis Presented to the
FACULTY OF THE USC DORNSIFE COLLEGE OF LETTERS, ARTS AND SCIENCES
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

August 2023

Copyright 2023 Sophia M. Recca
ii
Dedication

To my husband, Martin Doerr, for his endless support

iii
Acknowledgements
This paper would have been challenging without the guidance and support I received from the
University faculty. I would like to thank the members of my thesis committee, Dr. An-Min Wu
and Dr. Diana Ter-Ghazaryan, for encouraging meaningful and accessible research; my thesis
advisor, Dr. Elisabeth Sedano, for her support that made this paper possible; and Dr. Vanessa
Osborne, for her editing prowess. I am grateful to LtCol Ryan Sealy, Mike Myers, and Danny
Conley at the Air Force Rescue Coordination Center for preparing and sending data on personal
locator beacon activations. I am likewise indebted to Lovell Hopper, Meg Wilson, Eric Howard,
and Monty Bell at the California Office of Emergency Services, who connected me with a
statewide Search and Rescue dataset and humored my flood of questions related to the data. I
would be remiss if I failed to mention Doug Samp, the search and rescue program manager for
the United States Coast Guard (USCG) Pacific Area, and LCDR Zach “Texican” Geyer, a USCG
helicopter pilot responsible for coordinating aviation resources during an emergency incident, for
their support, encouragement, brainstorming, and connection to USCG resources. I also could
not have arrived at the starting line of this project if it were not for the backing by CAPT Jose
“Pepe” Arana and CDR Tom “Dirty” McCurdy, who supported my continuing education while
active duty. Thank you, everyone, for your time and consideration.

iv
Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Abbreviations ................................................................................................................................. ix
Abstract .......................................................................................................................................... xi
Chapter 1 Introduction .................................................................................................................... 1
1.1 Search and Rescue .............................................................................................................. 3
1.1.1 Governing Publications .............................................................................................. 3
1.1.2 Mountain SAR Incident Definition ............................................................................ 4
1.1.3 Five Stages of SAR .................................................................................................... 5
1.2 Satellite Communication Devices ....................................................................................... 6
1.2.1 Personal Locator Beacons .......................................................................................... 8
1.2.1 Satellite Emergency Notification Devices ................................................................. 9
1.3 Study Area ........................................................................................................................ 10
1.4 Search and Rescue Data .................................................................................................... 12
1.4.1 Air Force Rescue Coordination Center PLB Dataset .............................................. 14
1.4.2 California Office of Emergency Services Dataset ................................................... 14
1.5 Motivation ......................................................................................................................... 15
1.6 Thesis Overview ............................................................................................................... 16
1.7 Methodological Overview ................................................................................................ 17
Chapter 2 Related Work ................................................................................................................ 18
2.1 Spatial and Temporal Analysis of SAR Incidents ............................................................ 19
2.1.1 Analysis of Maritime SAR Incidents ....................................................................... 20
2.1.1.1 Spatial analysis of maritime SAR incidents .................................................... 21
2.1.1.2 Temporal analysis of maritime SAR incidents ............................................... 24
2.1.2 Analysis of Mountain SAR Incidents ...................................................................... 27
2.1.2.1 Aspatial statistical analysis of mountain SAR incidents ................................. 27
2.1.2.2 GIS and mountain SAR .................................................................................. 31
2.2 Spatial and Spatiotemporal Analysis ................................................................................ 35
2.2.1 Point Pattern Analysis .............................................................................................. 36
2.2.2 Global Spatial Statistics ........................................................................................... 40
2.2.3 Local Spatial Statistics ............................................................................................. 41
2.2.4 Spatiotemporal Analysis .......................................................................................... 45
2.3 Summary ........................................................................................................................... 48
Chapter 3 Methods ........................................................................................................................ 50
v
3.1 Data ................................................................................................................................... 51
3.1.1 Dataset Acquisition .................................................................................................. 53
3.2 Dataset Preparation ........................................................................................................... 56
3.2.1 Dataset Preparation in Excel .................................................................................... 56
3.2.2 Dataset Preparation in a GIS .................................................................................... 58
3.3 Spatial Analysis of the CALOES Dataset ......................................................................... 63
3.3.1 Point Pattern Analysis .............................................................................................. 65
3.3.2 Incident Aggregation and Neighborhood Structure ................................................. 66
3.3.3 Hot Spot, Cluster, and Outlier Analysis ................................................................... 68
3.4 Spatiotemporal Analysis of the CALOES Dataset ........................................................... 69
3.5 Comparing the AFRCC and CALOES Datasets ............................................................... 71
3.5.1 Comparing Spatial Relationships and Attributes ..................................................... 72
3.5.2 Assessing Accidental Activations ............................................................................ 73
3.6 Summary ........................................................................................................................... 73
Chapter 4 Results .......................................................................................................................... 75
4.1 Spatial Analysis of the CALOES Dataset ......................................................................... 76
4.1.1 Point Pattern Analysis .............................................................................................. 77
4.1.2 Incident Aggregation and Neighborhood Structure ................................................. 82
4.1.3 Hot Spot Analysis .................................................................................................... 85
4.1.4 Cluster and Outlier Analysis .................................................................................... 87
4.2 Spatiotemporal Analysis of the CALOES Dataset ........................................................... 91
4.3 Comparing the AFRCC and CALOES Datasets ............................................................... 98
4.3.1 Attribute Comparison ............................................................................................. 100
4.3.2 Accidental Activations ........................................................................................... 106
Chapter 5 Discussion and Conclusions ....................................................................................... 110
5.1 Project Findings .............................................................................................................. 111
5.1.1 Spatial Analysis Results ......................................................................................... 111
5.1.2 Spatiotemporal Analysis Results ........................................................................... 113
5.1.3 Attribute Analysis Results ..................................................................................... 113
5.1.4 Accidental Sat-Comm Device Activations ............................................................ 114
5.2 Limitations ...................................................................................................................... 114
5.3 Recommendations ........................................................................................................... 117
5.3.1 Recommendations for Future Research ................................................................. 118
5.3.2 Recommendations for SAR Agencies ................................................................... 120
5.4 Conclusion ...................................................................................................................... 121
References ................................................................................................................................... 122

vi
List of Tables
Table 1. The attributes associated with each dataset available for subsequent analysis ............... 63
Table 2. Average nearest neighbor results .................................................................................... 79
Table 3. Distance relationships between actual SAR incidents in the CALOES dataset ............. 80
Table 4. Distribution of actual mountain SAR incidents per hot spot trend type ......................... 98
Table 5. Distances of sat-comm device activations from the area of analysis boundary ........... 109

vii
List of Figures
Figure 1. The placement of mountain SAR within the SAR typological structure ........................ 4
Figure 2. The five stages of SAR, as adapted from the Land SAR Addendum .............................. 5
Figure 3. The relay of information from sat-comm device activation to rescue team launch ........ 7
Figure 4. A map of the study area, detailing land divisions by jurisdiction ................................. 11
Figure 5. An overview of the study's methodology ...................................................................... 51
Figure 6. A map of the AFRCC dataset records ........................................................................... 54
Figure 7. A map of the CALOES dataset records ......................................................................... 55
Figure 8. Initial dataset preparation and organization in Microsoft Excel ................................... 58
Figure 9. Depiction of incidents that were removed or retained ................................................... 60
Figure 10. Example of similar incidents that were retained ......................................................... 62
Figure 11. Overview of the methods and corresponding ArcGIS tools used for analysis ............ 64
Figure 12. A graphical representation of a Space-Time Cube; adapted from Esri (n.d.) ............. 70
Figure 13. Distribution of SAR incidents by means of origination .............................................. 76
Figure 14. KDE results using a 500 m grid cell and a 2,500 m search radius .............................. 81
Figure 15. Mountain SAR incident distribution within the 500 m hexagonal grid ...................... 83
Figure 16. Global Moran’s I distances for maximum spatial autocorrelation .............................. 84
Figure 17. Mountain SAR incident hot spot distribution and overlap .......................................... 87
Figure 18. Results of local spatial statistics .................................................................................. 89
Figure 19. Anselin Local Moran’s I results .................................................................................. 91
Figure 20. Incidents by notification method over time, CALOES dataset ................................... 93
Figure 21. Emerging Hot Spot results on Ritter Range ................................................................ 96
Figure 22. Emerging Hot Spot results near Mt Whitney .............................................................. 97
viii
Figure 23. Data clocks of the AFRCC (left) and CALOES (right) datasets, 2018-2022 ........... 101
Figure 24. Incidents by month and method of notification, CALOES dataset ........................... 102
Figure 25. Incidents originating with a PLB activation per day of the week, AFRCC dataset .. 102
Figure 26. Incidents by day of the week and method of notification, CALOES dataset ............ 103
Figure 27. Incidents by time of day and method of notification, CALOES dataset ................... 104
Figure 28. Incident elevations by PLB activation; AFRCC dataset ........................................... 105
Figure 29. Incident elevations by method of notification, CALOES dataset ............................. 106
Figure 30. Accidental activations of a sat-comm device by day of the week, CALOES dataset 108

ix
Abbreviations
AFRCC Air Force Rescue Coordination Center
ART Alpine Rescue Team
BLM Bureau of Land Management
CALOES California Office of Emergency Services
CCG Canadian Coast Guard
CDFW California Department of Fish and Wildlife
DEM Digital elevation model
ELT Emergency locator transmitter
EPIRB Emergency position-indicating radio beacon
ESDA Exploratory spatial data analysis
FOIA Freedom of Information Act
FS Forest Service
GIS Geographic information system
HEMS Helicopter-based emergency medical services
IAMSAR International Aeronautical and Maritime Search and Rescue Manual
ICAO International Civil Aviation Organization
IERCC International Emergency Response Coordination Center
IMO International Maritime Organization
INSARAG International Search and Rescue Advisory Group
IR Infrared
ISRID International Search and Rescue Incident Database
JMT John Muir trail
x
KDE Kernel density estimation
MODIS Moderate Resolution Imaging Spectrometer
MRA Mountain Rescue Association
NAPSG National Alliance for Public Safety GIS
NPS National Park Service
NSS National SAR Supplement to the IAMSAR Manual
NVD Night vision device
OHV Off-highway vehicle
PLB Personal locator beacon
RCC Rescue Coordination Center
SAR Search and rescue
SASH Spatial association of scalable hexagons
SEND Satellite emergency notification device
SSI Spatial Sciences Institute
STC Space-time cube
USC University of Southern California
USCG US Coast Guard
USDA US Department of Agriculture

xi
Abstract
Mountain search and rescue (SAR) incidents are high risk events that consume time and money,
often placing the lives of rescuers and subjects alike in precarious situations. The increasing
accessibility of satellite communication (sat-comm) devices for outdoor recreation may change
when and where mountain rescue teams are tasked, and California’s SAR agencies need to
understand the implications of emerging sat-comm device usage on SAR requirements to
mitigate future risks caused by resource and training shortfalls. To date, no academic studies
have conducted a holistic assessment of SAR incidents in the Sierra Nevada mountains or
considered the impacts of sat-comm device usage on the SAR caseload. Such a knowledge gap
impairs the ability of federal, state, and local agencies to anticipate costs and adequately train
rescue teams to respond to mountain SAR incidents. This research explores the spatial and
temporal patterns of historical mountain SAR incidents in the Sierra Nevada wilderness areas to
understand how sat-comm devices impact SAR services in one of the most visited mountain
regions in the continental United States. The results of this study suggest sat-comm devices are
replacing traditional methods of notification that alert authorities to an emergency. Incidents
where the subject communicates using a sat-comm device occur at sites of historical SAR
activity where traditional methods of communication are dominant, as well as at new – and more
isolated – locations. A lack of confidence in data quality, however, means this study primarily
serves to demonstrate spatial and spatiotemporal analysis methods that SAR agencies may adopt
to explore historical mountain SAR incidents at a regional scale.

1
Chapter 1 Introduction
Nature cares little for the boundaries built by humans to define dominion and stewardship.
People who venture into the wild and encounter emergency situations likewise request aid
irrespective of jurisdictional lines. Administrative boundaries continue to blur thanks to
technological advances in portable satellite communication (sat-comm) devices. Sat-comm
devices have near-global coverage areas, and they enable users to call for help anytime,
anywhere. More traditional methods of calling for help have limited capabilities compared to sat-
comm devices: cellular network antennas do not provide universal coverage; and word-of-mouth
relay of an accident is limited by human mobility. In theory, increased accessibility to rescue
services could mean an increased level of demand without a matching increase in supply.
Furthermore, should sat-comm devices enable connectivity to communications infrastructure in
areas that previously lacked access to human or cellular services, then the spatial distribution of
emergencies might broaden in addition to increasing numbers of requests for rescue services.
Activating a sat-comm device sets in motion search and rescue (SAR) efforts that are
ultimately executed by the emergency response agency with jurisdiction over the activation site.
Private and public organizations who monitor sat-comm device activations and coordinate the
response often maintain separate SAR incident datasets that adhere to different reporting
requirements. Similarly, the local SAR agencies who execute the response to all SAR incidents
within their jurisdiction, regardless of the method of notification, frequently keep records that are
not held to a state or national standard. This isolation of SAR incident records contributes to a
general lack of awareness of how trends play out across a geographic region. Emergency
response agencies would benefit from an analysis of cross-jurisdictional datasets in order to
improve their SAR response and determine to what extent new technologies like sat-comm
2
devices alter the SAR landscape. The optimal datasets for research therefore lie with state-,
regional-, national-, or international-scale agencies responsible for collecting and standardizing
records.
The intent of this research is to take two, cross-jurisdictional datasets and examine how
the spatial and temporal patterns of mountain SAR incidents originating with a sat-comm device
activation compare with the traditional means of distress notification (e.g., in-person notification,
cell phone, etc.) over time. The study area encompasses the wilderness areas of California’s
Sierra Nevada mountain range due to their extreme topography, relative inaccessibility, multiple
SAR controlling agencies, and high visitor numbers – factors which increase the risk in the SAR
process and complicate post-SAR analysis. To date, there are no academic studies that have
analyzed mountain SAR patterns at this scale in California nor considered the influence of sat-
comm devices on when and where rescue teams might be tasked. The goal of this research is
therefore to remedy this gap and determine the impact of sat-comm devices on the spatial and
temporal distribution of mountain SAR incidents. To meet this goal, this study presents
methodology that may be adapted by SAR agencies to continuously evaluate their local SAR
landscape. In this way, SAR agencies responsible for coordinating rescue teams might be better
prepared to respond to future mountain SAR incidents.
This chapter begins with a definition of the terms used throughout this study. This is
followed by an overview of sat-comm device types and services. The chapter then goes over the
study area and describes what SAR datasets are available for the study area. The chapter
concludes with a statement on the motivation behind the development of this project and a
review of the methods employed to advance the research objectives.
3
1.1 Search and Rescue
Search and rescue efforts involve locating people in potential or actual distress and
delivering them to safety. The goal of SAR agencies is to shorten the time from distress
notification to resolution without compromising safety or mission success. The framework for
operational success is laid out in regulatory publications and ultimately achieved by the real-time
execution and sound judgment of rescue coordinators and rescue teams. International SAR
organizations that fall under the United Nations, like the International Search and Rescue
Advisory Group (INSARAG), International Maritime Organization (IMO), and International
Civil Aviation Organization (ICAO), publish manuals that standardize procedures and articulate
rescue responsibilities on a global scale. National- and state-level guidance builds off these
documents to fit the needs of SAR operations in their respective coverage areas. This section
describes the domestic SAR structure including the chain of responsibility for SAR response and
the management of historical SAR data.
1.1.1 Governing Publications
The federal SAR agencies in the United States lean most heavily on the International
Aeronautical and Maritime Search and Rescue (IAMSAR) Manual – which is the joint work of
the IMO and the ICAO – to refine domestic procedures and offer structure to the civil SAR
process. The National Search and Rescue Committee (NSARC) is the federal organization
responsible for coordinating procedures for interagency standardization, and they publish the
National Search and Rescue Supplement (NSS) to the IAMSAR Manual (NSARC 2016) and the
Land SAR Addendum to the NSS (NSARC 2011). These two documents set out the terminology
and organizational relationships used in this research.
4
1.1.2 Mountain SAR Incident Definition
Unlike most classifications of SAR operations, a mountain SAR incident is not explicitly
defined in the Land SAR Addendum, though it is alluded to as a subset of land SAR, which is a
subset of civil SAR. Figure 1 offers a visual breakdown of how mountain SAR is categorized
within SAR terminology. Civil SAR efforts are defined as those that occur in a non-hostile
environment, and they range from aeronautical and maritime emergencies to catastrophic
disasters. Land SAR refers to SAR incidents that occur on land, outside of urban areas, and are
generally not associated with natural disasters. The definition of mountain SAR used for the
purpose of this research refers to a land SAR event which occurs in mountainous terrain away
from the built environment, where the subject is participating in outdoor recreation, and which
requires the assistance of specialty-trained SAR assets. Assets include technical ropes teams,
swiftwater rescue crews, off-highway vehicle (OHV) SAR teams, and helicopter crews, all of
which maintain specific qualifications and training.

Figure 1. The placement of mountain SAR within the SAR typological structure

A SAR incident refers to a request for SAR assistance, but responders are uncertain as to
whether a person is in actual distress. Geographic information science also has a term for
incident data: an incident refers to a site corresponding to a single set of coordinates (Esri n.d.).
Since the mountain SAR events in this research contain varying degrees of distress, from false
5
alarm to death, and because they correspond to a single coordinate pair, all mountain SAR events
considered for analysis in this research are referred to as mountain SAR incidents.
1.1.3 Five Stages of SAR
SAR consists of five stages: awareness, initial action, planning, operations, and
conclusion (NSARC 2011). Figure 2 depicts these stages derived from the model found in the
Land SAR Addendum. Each stage provides an opportunity for after-action lessons and process
improvement. In particular, the Planning and Operations stage are a continuous feedback
process, and advanced preparation (e.g., through research on historical incidents) can increase
the efficiency of the Operations stage for a faster time to SAR Conclusion. The spatial and
temporal analysis of historical SAR incidents identifies where and when incidents traditionally
occur so SAR operations centers and rescue teams can develop appropriate training and response
plans. For example, a consistent cluster of SAR incidents may be identified around climbing
routes that straddle a jurisdictional boundary, but only one of the jurisdictions has a technical
mountain rescue team on immediate recall that can respond to injured climbers. Questions asked
during the Initial Action and Planning stages could be tailored based on an analysis of historical
SAR incidents for optimized rescue asset preparation and utilization.

Figure 2. The five stages of SAR, as adapted from the Land SAR Addendum

6
1.2 Satellite Communication Devices
Owing to technological advances in sat-comm devices, calling for help is increasingly
accessible to the general population from anywhere on the planet that can connect to the
applicable satellite infrastructure. Sat-comm devices therefore have the potential to accelerate the
SAR process from locations that previously would have had a delay in incident notification if
subjects had to rely traditional notification methods or overdue procedures (i.e., when a person
misses a check-in, often relayed to SAR agencies by friends and family). Many of these sat-
comm devices have an “SOS” feature, the activation of which sends an emergency signal with
location information via satellite to a rescue coordination center (RCC). The RCC then takes
responsibility to inform the appropriate local rescue agency. Figure 3 presents a diagram of the
sat-comm device emergency notification process and incident record keeping. It is worth noting
that sat-comm device records are often saved in duplicate or triplicate: one record of an incident
lies with the RCC, one with the local agency accepting the tasking (e.g. county), and one with
the state agency (if they require the local agency to forward their reports). Along with coordinate
data, some models of sat-comm devices can also send and receive text messages. Because
modern sat-comm devices provide reasonably precise location data, they not only expedite the
Awareness stage of a mountain SAR incident, but the Initial Action and Planning Stages as well.
7

Figure 3. The relay of information from sat-comm device activation to rescue team launch

Because sat-comm devices make it easier to initiate a SAR response from remote
locations, they conceivably increase the demand for SAR services, with implications for SAR
asset management and support requirements. SAR agencies have a responsibility to investigate
all requests for aid until an incident is resolved. Variations in mountain SAR incident spatial and
temporal patterns, perhaps due to an increase in sat-comm device SOS activations, might
therefore impact how agencies manage emergency resources. Resources include time, money,
and lives: the time and money devoted to training; the time spent verifying the authenticity of an
emergency incident; and the cost of ground and aviation SAR assets to search for, locate, and
transport the subject in question. Agencies responsible for efficiently and safely planning and
coordinating SAR efforts, and the rescue assets tasked to assist, therefore benefit from knowing
not only when and where mountain SAR incidents traditionally occur, but how sat-comm devices
might alter the patterns and trends of these incidents with implications for future caseloads.
8
1.2.1 Personal Locator Beacons
Personal locator beacons (PLBs) are portable devices that, once activated, act as both a
radio beacon and a sat-comm device. The radio beacon function allows external assets with
direction-finding capability to locate the PLB signal, while the sat-comm component increases
communications coverage. PLBs send signals over the 406 MHz internationally recognized
emergency frequency to initiate a SAR response once the signals are picked up by SAR sensors
onboard international, publicly managed satellites. In addition to the 406 MHz frequency, PLBs
also emit radio frequencies over designated emergency channels which rescue units can home in
on. The modern, portable PLB has comparable functions to emergency position-indicating radio
beacons (EPIRBs) traditionally carried by maritime craft, and to the emergency locator
transmitters (ELTs) onboard aircraft. What differentiates PLBs from EPIRBs and ELTs is
registration: instead of being registered to a transport system, PLBs are registered to an
individual. PLBs were approved for civilian use in 2003 (US Air Force n.d.) and have since
grown in global popularity as technological advances have improved their functionality and
accuracy, with most models advertising satellite positional data accurate to within 100 m (US Air
Force n.d.).
PLB activation signals are detected by the international COSPAS-SARSAT satellite
constellation, passed to a ground station, and routed to the appropriate RCC (LandSAR n.d.). In
the continental United States, the Air Force Rescue Coordination Center (AFRCC), located at
Tyndall Air Force Base, Florida, is currently responsible for notifying the appropriate local
agencies of device activation within their jurisdiction based on AFRCC and State coordination
procedures (US Air Force n.d.). PLBs no longer have market dominance in portable, satellite-
capable, emergency assistance devices, however, and consumers can currently choose from a
9
range of products linked to commercial satellite systems like Zoleo, SPOT, and Garmin’s
InReach. These are discussed further in the section below.
1.2.1 Satellite Emergency Notification Devices
The commercial sat-comm products that have emerged over the past couple of decades
are referred to as satellite emergency notification devices (SENDs). Depending on the device and
the associated satellite system, SENDs can provide SAR responders with coordinate data
accurate within 5-15 m under most operating conditions (Garmin n.d.; SPOT n.d.). Unlike PLBs,
SENDs do not emit homing frequencies and instead rely solely on signal relay through the
partnered satellite infrastructure. For example, Garmin and Zoleo products use the Iridium
satellite network (Garmin n.d.; Zoleo n.d.), while SPOT uses Globalstar satellites and ground
stations (SPOT n.d.). SEND activation results in coordination through a partnered RCC, with
most devices going through the International Emergency Response Coordination Centre
(IERCC) (IERCC n.d.).
Tracking data from SEND activations might provide a wealth of information, offering
insights into patterns in SAR incidents and implications for future trends. For instance, as of
October 2022, Garmin announced 10,000 SOS activations from its InReach products after just
over a decade on the market, with over a third of activation originating from backpacking and
hiking users and over half due to medical or injury needs (Garmin 2022). Emergency
notifications reliant on satellite infrastructure will only increase as more devices connect to
satellite networks. In 2022, Apple announced their iPhone 14 smartphone models will be capable
of emergency notifications via satellite systems, and the company has invested in improving the
Globalstar satellite infrastructure (Apple 2022). With the future looking like every person who
goes into the wilderness will be able to call for help with the press of a button, SAR coordinators
10
and responders would do well to be armed with as much advanced information as possible on the
spatial and temporal trends associated with satellite-initiated mountain SAR incidents in an age
of omnipresent connectivity.
1.3 Study Area
California’s Sierra Nevada mountain range is of particular interest to mountain SAR
operations due to high visitor numbers, diverse terrain features, and opportunities for recreation
in backcountry areas. These characteristics also make the range interesting for spatial
exploration, as spatial phenomena influence how visitors interact with the landscape. For
example, trail networks tend to invite higher numbers of visitors than off-trail locations (Doherty
et al. 2011), and some viewpoints might draw particularly large crowds of people looking to
enhance their social media profile (Lu et al. 2021). The study area for this research falls within
the portion of the range commonly referred to as the High Sierras, as this section includes world
famous – and heavily traveled – trail systems weaving amongst some of the highest peaks in the
continental United States (James and Eardley 2021).
Several websites that keep track of visitor permits provide an indication of the high traffic
volumes. The National Park System (NPS) reports Yosemite National Park hosts more than four
million visitors per year, and that permits for the John Muir Trail (JMT) – which runs from
Yosemite to Mt. Whitney – doubled from 2011 to 2015, leading to a cap of 45 permits per day
(NPS n.d.). The non-profit Pacific Crest Trail Association reports a similar jump in permit
numbers, from 1,879 issued in 2013, to 7,888 issued in 2019 (PCTS n.d.). More visitors might
equate to more opportunities for sat-comm device activation, intentionally or accidentally, and
potentially an increased demand for mountain SAR support.
11
The Sierra Nevada mountains include several wilderness areas governed by three public
agencies: the Bureau of Land Management (BLM), the Forest Service (FS), and the NPS.
Wilderness areas are lands protected by federal law to provide opportunities for solitude and to
limit access to man-made infrastructure and technology (Wilderness Connect n.d.). Due to the
lack of infrastructure, wilderness areas effectively exclude non-mountain SAR events (e.g., car
accidents), and the study area is based on the boundaries of these wilderness areas (Figure 4).

Figure 4. A map of the study area, detailing land divisions by jurisdiction

Sixteen of California’s 58 counties include some portion of the High Sierras, and the
study area crosses into thirteen of these. In California, the counties are responsible for
12
developing procedures to respond to all Land SAR incidents not covered by federal agencies,
i.e., excluding aviation emergencies, maritime emergencies in the navigable waters within the
United States, and within the NPS lands (California Public Law n.d.; NSARC 2011). If the
counties require additional assistance, they can coordinate with state and federal agencies for
external assets. The NPS is a federal organization under the Department of the Interior, and it
maintains its own SAR response for the lands it administers. While the NPS does not act in the
capacity of a RCC, it maintains incident coordination functions, and NPS SAR assets may assist
neighboring jurisdictions in the SAR process, if required (NSARC 2011). Since neither the BLM
or FS are assigned federal SAR responsibilities, the county is responsible for SAR incidents
within BLM and FS wilderness areas. Incidents originating with a sat-comm device enter the
SAR process as depicted in Figure 3 above.
Regardless of who administers the land, the High Sierras encompass challenging
environments for visitors and rescue teams alike. Mountain peaks reach upwards of fourteen
thousand feet in some areas, and the higher elevations pose a risk to unacclimatized visitors, as
well as to helicopters that have reduced performance at higher elevations. Hazardous conditions
become more pronounced at night when rescue teams lose the benefits of daylight, and teams
with infrared (IR) and/or night vision device (NVD) capabilities are often required. Exploring
historical mountain SAR incidents and the influence of sat-comm devices on spatial and
temporal distributions thus helps SAR organizations prepare for future SAR operations occurring
in this challenging geographic region.
1.4 Search and Rescue Data
The quality of spatial and temporal analysis output depends on the quality of the input
data. Incomplete datasets might suggest spatial patterns which are inaccurate, and temporal
13
trends might also be conservative or exaggerated. Datasets which are limited to one
administrative unit can fail to capture spatial relationships near their borders, as data from
neighboring units would not be considered. One way to mitigate concerns over data
completeness is to have one SAR agency set the standards for data collection and serve as a data
repository.
There is precedent for SAR data standardization in the United States: maritime SAR
operations are coordinated through the US Coast Guard (USCG), the agency which also manages
the historical maritime SAR database. This database spans all USCG coverage areas – from
inland lakes to the waters off Hawaii and Alaska – and adheres to detailed standards, making it
ideal for studies exploring the spatial and temporal relationships of emergency events and
looking for ways to improve the SAR process (Hornberger, Cox, and Lunday 2022; Malik et al.
2014). The USCG also employs a spatial analysis and predictive program, the SAR Optimal
Planning System (SAROPS), to integrate real-time environmental conditions and geographic
information for improved SAR asset response (USCG n.d.).
Unlike the maritime environment, mountain SAR – excepting NPS data – has neither an
equivalent, comprehensive, national-scale dataset, nor SAROPS-type software available to first
responders, although efforts are being made in this direction. The International SAR Incident
Database (ISID), created with a grant from the US Department of Agriculture, intends to serve as
a data repository for multiple nations – the United States included – to better understand lost
person behavior in varying overland environments. However, the ISID currently does not
represent data from all fifty states, and California is not yet a contributor (dbs Productions n.d.).
SEND activation data through either the companies that support the devices or the IERCC could
provide another source of national-scale data, but SEND data are not available from any private
14
companies due to privacy concerns. Instead, public agencies currently offer the best available
options to academic researchers. The US Air Force, as the responsible authority for PLB
activations in the United States, is one data source that can offer nation-scale insights on a slice
of mountain SAR incidents. Another source that can provide state-wide data on emergency
incidents is the California Office of Emergency Services (CALOES).
1.4.1 Air Force Rescue Coordination Center PLB Dataset
The AFRCC offers a limited option for spatial and temporal analysis of mountain SAR
incidents spanning administrative boundaries. The AFRCC is the primary inland SAR
coordinator at the federal level within the continental United States (NSARC 2011). They are
responsible for managing all distress calls originating from PLBs and ELTs. Since the latter are
associated with aircraft, and the intent of this research is to examine patterns in mountain
recreation, only PLB data are considered an appropriate representation of mountain SAR
incidents per the definition used in this paper. While the AFRCC dataset is limited to PLBs and
does not capture private sat-comm device activations, it could represent how people use sat-
comm device technology in remote mountain areas to initiate the SAR process. Access to the
AFRCC data requires a Freedom of Information Act (FOIA) request, and the request is restricted
to no more than seven years’ worth of records. While there are not enough PLB activations in the
Sierra Nevadas to support meaningful spatial statistical analysis, the PLB data from the AFRCC
complements the more numerous CALOES SAR incident dataset, serving to reinforce findings
on the role of sat-comm devices in the mountain SAR process.
1.4.2 California Office of Emergency Services Dataset
The main option to assess mountain SAR incidents and sat-comm activations at the scale
of the study area is to pull from the state-level dataset originating with the CALOES. In
15
California, all mountain SAR incidents except those within a National Park are managed at the
county level, unless a county requests additional assets, at which point they reach out to state
assets via CALOES and federal assets coordinated through AFRCC. One of the results of the
division of SAR responsibilities is the lack of a historical data repository managed by one entity
for the state. That changed in 2018 when CALOES started collecting SAR incidents from the
array of jurisdictional entities. While the robustness of the CALOES dataset relies on the
reporting quality of the counties and NPS, it offers unparalleled access to large numbers of SAR
incidents with attributes on location, date, and time. Unlike the AFRCC dataset, the CALOES
dataset includes all mountain SAR incidents regardless of means of distress notification, be it by
cell phone, sat-comm device, overdue procedures, or other method of relay. Because AFRCC
passes PLB distress notifications to the appropriate local rescue agency, it would be assumed
both the AFRCC and CALOES datasets overlap, except for accidental PLB activations if
AFRCC were able to verify the false alarm without involving additional assets. Together, these
two datasets, the one from AFRCC which contains only PLB data and the one from CALOES
which contains all land emergency incidents, are used for exploration and analysis of sat-comm
devices in mountain SAR incidents.
1.5 Motivation
Knowing when, where, and what mechanisms are influencing mountain SAR incidents
provides SAR agencies and rescue teams with actionable information to design effective training
plans and maintain the appropriate equipment for safe rescue operations. Despite the benefits
associated with understanding how sat-comm devices are impacting the mountain SAR
landscape, no academic research has either applied spatiotemporal analyses to mountain SAR
incidents or explored the impact of sat-comm devices in the SAR process. For California’s High
16
Sierra mountain region specifically, filling this gap in the scholarly literature has the potential to
improve the SAR process for multiple emergency response jurisdictions, as well as assist the
additional state and federal assets the counties and NPS might call upon for assistance.
Statistically supported results identifying locations where mountain SAR incidents are not only
occurring year after year, but also where incidents are exhibiting a positive or negative trend,
could support policy requests for new resource or funding allocations. SAR ground teams and
helicopter crews could use the results from this study to train new members and better prepare
them for conditions they can expect to encounter. One popular albeit unofficial SAR motto
states, “the first rule of SAR is don’t make more SAR.” Having a thorough understanding of
when and where mountain SAR incidents historically occur across an entire geographic area
would bolster local experience, increase safety margins through adaptations to policy and
training, and decrease the odds that the rescue team could become, in turn, a SAR case.
1.6 Thesis Overview
Chapter 1 has reviewed the background and motivation for this research, as well as
provided a description of the terms and topics used in this study. Chapter 2 delves into where
SAR features in the academic literature, supplemented by research on the spatiotemporal
analysis of non-SAR emergency incident data to support the methods used in this study. Chapter
3 covers the methods employed in this research to identify and compare the spatial and temporal
patterns of sat-comm device-initiated mountain SAR incidents against other means of SAR
notification. Chapter 4 presents the results of this study. Chapter 5 offers a discussion of the
results, a review of the study’s limitations, and recommendations for future research and SAR
policy makers.
17
1.7 Methodological Overview
The goal of this study is to explore the impact of sat-comm devices on mountain SAR in
the High Sierras. The methods developed to accomplish this goal involve spatial statistics, a
trend statistic, and visual analysis. Due to data constraints, only the CALOES dataset is
examined using spatial and trend statistics, while the AFRCC dataset supplements the statistical
results through comparison and visual analysis.
Individual mountain SAR incidents from the CALOES dataset are first explored using
point pattern analysis techniques to detect the distances at which spatial associations appear.
Global and local spatial statistics are then used to detect significant spatial patterns of aggregated
mountain SAR incidents across the study area and within local neighborhoods respectively.
Conducting trend analysis in conjunction with local spatial pattern analysis identifies emerging
patterns within the mountain SAR neighborhoods, facilitating the interpretation of the mountain
SAR incident spatial distribution over time.
Using visual analysis and distance measurements, mountain SAR incidents from the
AFRCC dataset are evaluated in the context of the CALOES spatial statistical results to assess
possible relationships. Spatial and temporal attributes from both datasets are explored and
compared using descriptive statistics. The accidental activations of sat-comm devices from both
datasets are then evaluated for their potential to increase the workload of mountain SAR
organizations.

18
Chapter 2 Related Work
The intent of this research is to explore spatial trends in mountain search and rescue (SAR)
incidents across a geographic region in order to assess how satellite communication (sat-comm)
devices might affect the SAR landscape over time. To meet the research objectives, datasets
containing mountain SAR incidents from California’s Sierra Nevada mountain range are brought
into a geographic information system (GIS) – a type of software that facilitates visual and
statistical analysis of geographic data. The datasets are cross-jurisdictional to capture the spatial
and temporal scope of incidents that, like the mountains they occur in, do not pay heed to
administrative boundaries. Mountain SAR incidents originating with a sat-comm device are
compared against incidents that do not rely on these devices using visual and statistical methods
developed in Esri’s ArcGIS Pro 2.9 software suite (Esri 2021). ArcGIS Pro offers a user-friendly
interface to explore and assess historical incident data through spatial and temporal analysis. This
study expands upon prior academic research to demonstrate how SAR professionals can
incorporate GIS tools to examine SAR incidents over space and time and explore the influence
of sat-comm device activations.
There are, however, relatively few research papers and books that consider both the
spatial and temporal components of SAR incidents. Much of the academic literature to date
examines maritime SAR incidents, and of these, a growing number leverage the benefits of a
GIS to conduct spatial analysis of geographic data as computational processing capabilities
improve (Goerlandt and Siljander 2015; Guoxiang and Maofeng 2010; Stoddard and Pelot 2020).
While several maritime SAR studies examine the temporal attributes of SAR incidents (Malik et
al. 2014; Sonninen and Goerlandt 2015; Stoddard and Pelot 2020), none assess the emerging
trends of incidents tied to a specific location, possibly because the maritime domain is fluid and
19
rarely constrained by stationary topographic features. There is a dearth of published research that
reviews the spatial components of mountain SAR incidents, particularly at scales that span
multiple jurisdictions. To supplement the thin body of work that deals explicitly with mountain
SAR incidents, one needs to explore emergency incidents from other genres that occur at similar
spatial and temporal scales. To this end, there is a burgeoning number of studies that examine the
spatiotemporal patterns of wildfires (Aftergood and Flannigan 2022; Reddy et al. 2019; Visner,
Shirowzhan, and Pettit 2021). Wildfires are similar to mountain SAR incidents in that they can
occur across expansive environments and are often seasonal, making studies on wildfire patterns
a suitable genre to reference. This chapter reviews the related literature covering these three
categories of emergency incidents – maritime SAR, mountain SAR, and wildfires – and
discusses how techniques and lessons from past research can inform the methodological design
of this paper.
2.1 Spatial and Temporal Analysis of SAR Incidents
As of 2023, there are far more academic works advancing the maritime SAR process than
land SAR missions. In respect to spatial analysis, the scope of maritime SAR differs from
mountain SAR. Maritime SAR is largely two-dimensional: maritime SAR incidents are rarely
associated with a topographic feature (exceptions would be narrow waterways and littoral
hazards) and instead are vulnerable to the drift of currents and winds. By contrast, mountain
SAR incidents occur in a three-dimensional space and tend to be stationary. Despite these
differences, the spatial and temporal findings from maritime SAR studies offer implications on
hazard identification and resource allocation that are similar to mountain SAR.
Though sparse, prior research on mountain SAR incidents covers a spectrum of topics,
including demographics, injury patterns, lost person behavior, and the digitization of historical
20
datasets. The scale of analysis conducted in prior research is, however, constrained, and often
limited to a single SAR jurisdiction (e.g., a national park). Very few mountain SAR studies
conduct analyses across a region that encompasses several administrative boundaries. Another
limitation with the mountain SAR literature concerns the breadth of analyses employed: most
research articles to date that examine historical mountain SAR incidents rely on aspatial
analytical methods, which can reveal temporal patterns but lacks the spatial considerations
available with a GIS. The few studies that do employ a GIS make use of datasets spanning
several years but elect to conduct purely spatial rather than spatiotemporal pattern analysis. A
review of past research on mountain SAR incidents highlights the gaps in analysis, but also
reveals why a thorough understanding of mountain SAR is critical to mitigating the risks posed
to rescue teams and subjects in distress.
2.1.1 Analysis of Maritime SAR Incidents
The maritime environment is the main domain for scholarly research on SAR incidents.
This bias is possibly due to a drive to protect businesses and promote safety. Fishing,
recreational boating, and commercial shipping operations all occur in potentially hazardous
environments, and if there is a low chance of a successful rescue, poor SAR support could hurt
public and private sector interests (Marven, Canessa, and Keller 2007). The bias might also be
due to the relatively high profile of maritime emergencies compared to mountain ones: not only
do ships and boats contain more lives than the average hiking party, but there are also
environmental concerns associated with oil spills and contaminants entering the water (Goerlandt
et al. 2017). The higher percentage of maritime SAR research studies might also be due to
dataset availability. Comprehensive datasets for maritime SAR incidents are maintained by a
nation’s Coast Guard, and these datasets generally suffer less from the fragmentation or varying
21
standards seen with mountain SAR datasets at scale, though there are still data quality concerns
associated with missing data (Malik et al. 2014; Stoddard and Pelot 2020). While spatial analysis
methods are not always necessary to examine maritime SAR data, they are common in maritime
SAR studies to account for the spatial nature of incident data.
2.1.1.1 Spatial analysis of maritime SAR incidents
A common theme in the academic research on maritime SAR incidents is the
identification of incident hot spots and clusters, often to determine whether current rescue asset
locations offer sufficient coverage. Azofra et al. (2007) conducted a type of point pattern analysis
– weighted density analysis – to develop an objective, apolitical method to determine the best
placement of maritime rescue assets. They designed two distribution models that categorized the
suitability of a rescue boat’s or rescue helicopter’s base station represented by changes in a
coefficient. Azofra et al. found their zonal distribution model, which involved constructing zones
based on SAR asset capabilities, preferable to their individual distribution model, which
considered a single asset to every incident. This is because a zone smooths out the effects of
outliers. Within each zone, a single set of coordinates representing a “superaccident” site was
identified and used as input in the model. The superaccident coordinates were based on the
arithmetic mean of the incidents occurring within the zone and were weighted by the total
severity of incidents. Severity was based on a four-point scale, and it encompassed medical
concerns, sea surface temperatures, and hazards in the area. While Azofra et al.’s model offers an
objective approach to guide decisions on resource allocation amongst local and regional entities,
the authors recognize their model –since it is built from historical incident data – assumes future
incidents will follow similar spatial patterns. They therefore recommended continually updating
the model’s inputs to identify the most efficient distribution of resources for SAR success.
22
While the identification of SAR incident clusters based on historical incident analysis
may inform asset placement strategies, it may also provide insights as to whether existing
administrative boundaries should be redrawn to facilitate more efficient SAR tasking. Marven,
Canessa, and Keller (2007), in their book chapter on exploratory spatial data analysis (ESDA)
and maritime SAR, reviewed how point pattern analysis and spatial statistics can support
effective decision making and evaluate jurisdictional boundaries. The authors demonstrated their
methods using the GIS tools of the day and the Canadian Coast Guard’s (CCG) incident data
from the Pacific Region, 1993-1999. After cleaning the CCG data to remove inaccurate or
incomplete incidents, the authors were left with 11,457 maritime SAR incidents spanning
approximately 157,000 km
2
. Visualizing the point patterns of incidents revealed obvious spatial
heterogeneity. Aggregating the incidents by jurisdiction would preclude a realistic assessment of
spatial patterns, since jurisdictions encompass a large amount of open water, but maritime
incidents are mostly distributed across the small area of sheltered waters. In contrast, point
pattern analysis methods, like visual analysis and kernel density estimates (KDE), do not suffer
from aggregation pitfalls like unnatural jurisdictional lines or the modifiable areal unit problem,
but point pattern analysis does lack the significance metrics provided by spatial statistics.
In order to have a statistical foundation for spatial pattern analysis, Marven, Canessa, and
Keller turned to CrimeStat version 3 (Levine 2004), a statistical software package that can
analyze geographic incident data. The authors applied two point pattern analysis methods to the
CCG dataset to find statistically significant spatial clustering: the Spatial and Temporal Analysis
of Crime (STAC) and nearest neighbor hierarchical (NNH) clustering. Both methods require an
element of subjectivity. With STAC, the analyst needs to specify a grid cell size and the
minimum number of points in a cluster for comparing densities. With NNH clustering, the
23
analyst needs to similarly define the number of points that constitute a cluster as well as the
threshold distance between points to consider them neighbors. The authors found NNH useful for
comparing incident clusters over time, though they felt KDE was the best for a visual
comparison of datasets. While Marven, Canessa, and Keller provide an expert review of how to
maximize the benefits of spatial analysis to advance maritime SAR efforts, the authors did not
discuss the efficacy of using a GIS at a regional scale, nor did they provide guidance on how
analysts should set parameters to achieve results that most closely represent the underlying
spatial associations.
Spatial analysis tools available in a GIS can produce intuitive and visually accessible
results, although the spatial conclusions are based on an imperfect representation of reality
largely due to computational processing limitations. Goerlandt, Venäläinen, and Siljander (2015)
constructed a risk-based model to review rescue boat capabilities in the Gulf of Finland from
2007 to 2012, in which they used a GIS to identify high-density accident sites. Their study area
stretched along the southern coast of Finland and covered about 11,500 km
2
. The authors used
descriptive statistics and charts to evaluate several risk indicators that were not associated with
specific coordinates (e.g., the temporal distribution of incidents and mission attributes). They
used GIS tools to evaluate the spatial distribution of incidents and run a cost-distance analysis of
rescue boats to high-density accident site. Using ArcMap software, the authors created a density
surface of the study area for a visual analysis of incident hot spots and to measure rescue boat
response times to the high-density sites under a variety of wind and wave simulations. Goerlandt,
Venäläinen, and Siljander found the ArcMap tools offered a higher level of fidelity than aspatial
methods when developing their spatial risk indicators. However, the authors noted their methods
were time consuming, owing to the 10 m resolution cost surface required to accurately represent
24
the coastal topographic features (e.g. islands and waterways). The authors discussed the
limitations of resolution further, as well as specific ArcGIS software model limitations when
modeling the maritime environment, in another paper published the same year (Siljander et al.
2015). While GIS-based analysis facilitates the exploration of geographic data, models and
methods reliant on GIS products must balance the study area size and scale of analysis with
computational demands for effective research.
2.1.1.2 Temporal analysis of maritime SAR incidents
Although maritime and mountain SAR incidents occur in different operating
environments, the emphasis by maritime SAR researchers to identify SAR incident clusters and
improve maritime SAR policy is equally applicable across domains. Similarly, maritime SAR
studies that explore temporal patterns of SAR incidents offer relevant methodological techniques
to mountain SAR research due to the emphasis on resource distribution and hazard mitigation. In
the academic research to date on maritime SAR, temporal analysis is generally aspatial, often
taking the form of a graphical representation or a trend statistic.
Whether it is a GIS-derived map, a chart, or a matrix, a visual representation of spatial
and temporal data often increases the accessibility of research products to a broad audience base
and prompts new questions about what drives the patterns under observation. To assist the
USCG Ninth District, whose area of responsibility covers the Great Lakes, Malik et al. (2014)
created an interactive visual analytics system for exploring spatial and temporal patterns of
historical incidents in the region. Their system is based on a custom GIS supported by Microsoft
Windows software, and it incorporates OpenStreetMap base layers through several programming
languages. SAR incidents may be viewed as unique values or as density-based heatmaps, and
they may be selected by attributes and color-coded. Users can interact with the data through
25
linked windows that present graphical representations of attributes over space and time. Graphics
include line and bar graphs, as well as calendar and clock graphs. This visual analytics system
successfully incorporates a large quantity of data and creates temporal visualizations of SAR
incidents that support the USCG decision making process; for example, Mondays and Tuesdays
look particularly busy, so the USCG might want to rethink making some stations operate only on
the weekends. The authors do, however, recommend future work integrates temporal prediction
algorithms rather than relying purely on graphical representations.
Exploring SAR incidents over time does not necessarily require the manufacture of a new
analytics system, and there have been commercial products to date other than a GIS that can help
policy makers explore the temporal attributes of their data. Stoddard and Pelot (2020) reviewed
how an open-source JavaScript library, Data-Driven Document(D3), can increase the
accessibility of the CCG’s SAR Program Information Management System (SISAR) dataset. The
authors focused on SISAR data from 2005 to 2013, with 2007 omitted as it was unavailable. D3
takes an underlying spatial and temporal dataset like SISAR and creates web visualizations that a
user can manipulate as a web dashboard. The authors organized the maritime SAR incidents by
hour, day, month, and year to support temporal analysis via graphical representations (i.e.,
graphs, pie charts, and incident heat maps). The authors found the dashboard-style visual
analysis techniques to be effective decision-making tools, highlighting when and where the CCG
should concentrate their resources. While Stoddard and Pelot acknowledged there were concerns
associated with under-reporting of incidents, they concluded a visual analysis of maritime SAR
incidents over time and space supports effective decision making. However, neither Stoddard
and Pelot nor Malik et al. (2014) discussed how merging temporal and spatial analysis could
26
provide their target audience with enhanced information on how incidents may change in space
over time, which could possibly reveal underlying processes of interest to the CCG and USCG.
Graphing incident attributes that are specific to when an incident occurs is another
temporal analysis technique. Environmental and meteorological conditions vary throughout the
days, months, and seasons. While such ambient conditions may be a contributing factor to a SAR
incident, they also influence the types of hazards a SAR rescue team may encounter when
responding to a distress notification. Using maritime SAR incident data from 2007 to 2012 in the
Gulf of Finland on boating accidents, Sonninen and Goerlandt (2015) were able to match
historical environmental and meteorological conditions to when an incident occurred. They could
then analyze the incidents aggregated by day, weekend, holidays, week, month, and year. Their
goal was to determine whether incidents occurred more frequently under certain conditions, and
to identify which graphical representations were optimal for revealing temporal patterns and
outliers. The authors pulled meteorological conditions from several temporal scales: for instance,
most weather stations collected data every 10 minutes, while precipitation was recorded by the
hour. Wave data came from a single buoy, so while it was representative of the study area,
results could not be considered accurate for each accident site. The authors examined the
attributes through graphical products created in the GeoViz Toolkit, a software that supports
geographic data. They found a parallel coordinate plot was the most effective presentation for
detecting attribute patterns over time, as it allows multiple variables to be displayed at once and
clearly reveals outliers. The multiform bivariate matrix was the best option for a comparison of
two variables. While the authors demonstrated methods to graph SAR incident attributes over
time, their work lacks a combined spatiotemporal element. Furthermore, Sonninen and
Goerlandt’s graphical representations were difficult to interpret if too much data were visualized
27
in a single plot, whereas a density surface on a map would hold up better, particularly for non-
scientific viewers – a useful insight for future research if attempting to visualize temporal
patterns and detect trends.
2.1.2 Analysis of Mountain SAR Incidents
As the work above demonstrates, understanding when and where maritime accidents tend
to occur can support an effective distribution of rescue resources and provide insights into the
risks associated with the highest densities of SAR incidents. A spatial and temporal analysis of
mountain SAR incidents similarly impacts resource management decisions to increase the safety
margins for rescue assets and persons in distress alike. However, there are few studies to date
that explore the spatial components of mountain SAR incidents.
2.1.2.1 Aspatial statistical analysis of mountain SAR incidents
A review of the professional literature on mountain SAR makes it clear that aspatial
statistical analysis is the most common method to evaluate historical SAR incidents. The
dominant software used in studies from the past couple of decades is SPSS, a statistical software
suite developed by IBM that offers statistical results and graphics (IBM n.d.). The benefits to
aspatial statistical analysis methods include fast processing times and easy integration with user-
friendly, graphical products. Then downside is the results cannot be connected to a specific
geographic location, making it difficult to detect changes in SAR incident accessibility over
time, or to explore spatial attributes that could contribute to an incident or alter the tasking for
rescue teams.
Traditional statistical analysis can incorporate temporal attributes. Kaufmann, Moser, and
Lederer (2006) looked at changes in the frequency and types of incidents involving helicopter-
based emergency medical services (HEMS) in Tyrol, Austria, from 1998 to 2003. Their working
28
hypothesis was that the types of emergency requests were changing in favor of less critical
injuries, which would reduce the demand for air-ambulance transport but not necessarily for
helicopter-based intervention. The authors created a severity score for incidents based on the
National Advisory Committee of Aeronautics, and binned incidents into categories of minor,
serious, severe, and critical. Separate scores for head injuries were based on the Glasgow Coma
Scale. They then broke the dataset into two temporal bins, 1998-2000 and 2001-2003, to identify
patterns and detect changes using SPSS version 11. The authors found 5% of all incidents were
false alarms, and leisure-related requests for HEMS support increased in frequency by almost
40% each year. Based on the injury patterns and changes in injury severity over time, the authors
surmised that the relative accessibility of HEMS services, increased use of mobile phones, and
popularity of technical equipment without matching experience levels might contribute to greater
risk exposure with the assumption a HEMS could always get a subject out of a bad situation. The
authors also found a seasonal influence on injury patterns, likely due to the exposure of ground
hazards. Kaufmann, Moser, and Lederer acknowledged their study is limited by a reliance on
incident figures from a single HEMS agency when several operate in the same area, potentially
creating a bias. However, they believe their results offer useful information to HEMS agencies
on how they can best support their helicopter rescue teams. For example, the very small number
of incidents requiring technical gear for recovery from canyons, crevasses, and ledges could
justify removing those capabilities, since cliff-side rescues require a lot of training for
proficiency and the cost may not be justified. While not spatial in nature, the authors demonstrate
how statistical analysis can reveal temporal patterns in incident attributes that can inform SAR
management services.
29
High resolutions of temporal and spatial data may be desirable for accurate interpretation
of spatial incidents: the weather conditions, slope, and elevation at the site of a SAR incident
would mean more to stakeholders than the average conditions for a geographic area. However,
analysis that intends to assess jurisdictional characteristics and funding requirements would not
necessarily require data at the individual level. Heggie and Amundson (2009) noted there were
no national-scale studies assessing land SAR incidents, and so they designed their research to
compare SAR incidents in national parks across the United States using SPSS version 15. The
authors pulled data from National Park Service (NPS) reports spanning 1992 to 2007, and
aggregated incidents by national park unit and by year. The authors decided to only look at the
total number of incidents, the number of subjects, rescue outcomes, and costs, since other
attributes suffered from a lack of standardization. Additionally, the authors examined 2005 data
at the same spatial scale but a finer temporal scale, considering the date, time, operating
environment, demographics, activity, and the SAR process. The authors were able to compare
the number of SAR incidents against the total costs per NPS region by year, how those costs
were broken up, and discuss what the results meant for NPS fiscal planning. For example, they
found there was an average of 11.2 SAR missions in the US national parks every day, costing
about $895 each. Using the 2005 data, the authors demonstrated how they could assess incident
attributes, finding about a quarter of SAR incidents were in mountainous terrain between 1,524-
4,572 m, followed by canyon areas, and that hiking was the most common activity driving a
SAR call, followed by boating and suicides. Heggie and Amundson demonstrated how a national
incident dataset could reveal patterns not captured by individual NPS units. However, an analysis
of incident patterns at the same scale would provide a higher fidelity analysis on where within
30
each national park incidents were historically occurring, which could possibly support mitigation
measures that could save costs over time.
To understand how to mitigate the number of SAR incidents requiring a rescue response
and better prepare rescue teams, SAR agencies require details on what factors contribute to the
severity of an incident. While an analyst can gather some attributes after-the-fact, like weather or
topography, behavioral attributes require input from the subject of a SAR case. Boore and Bock
(2013) wanted to ascertain the causal factors in backcountry SAR cases in National Parks in
order to recommend prevention measures, since they found research on SAR in the NPS tended
to focus on patterns of incident outcomes (i.e., medical injuries). The authors defined
backcountry SAR as incidents unreachable by ground ambulance. The authors pulled data from
Yosemite National Park Patient Care Reports from 2000 to 2009, since prior to 2000 SAR cases
were inconsistently reported. The authors also sent out surveys to the subjects of the most recent
cases (i.e., from 2007 to 2009, to mitigate recall bias) where there was a valid mailing address on
file. The authors asked for the subject’s experience levels, the time of day when they found
themselves in distress, the stage of activity, environmental conditions, and what the subject
thought would have helped them avoid the incident (e.g., better equipment). Incidents were
aggregated by subdistrict within Yosemite National Park. Statistical analysis was run in SPSS, to
include Pearson’s chi-squared test to check for significant correlation between demographics
against type of activity, injury against type of activity, and incident attributes against subdistrict.
Boore and Bock found most backcountry SAR incidents occurred during the day, in clear
weather, and during the second half of the subject’s activity. Of interest to this research, 6% of
the survey respondents reported having a GPS device and considered themselves experts, and no
respondent believed the GPS device would have prevented their incident. Respondents who only
31
had a cell phone self-reported as beginners. The authors acknowledged their results likely suffer
from low survey response rates, researcher bias and subjectivity, and the omission of incidents
whose outcome was self-rescue. While their study provides context to backcountry SAR cases,
they recommend future NPS efforts incorporate more details at the time a Patient Care Report is
written to offer more details for future analysis and support mitigation measures.
2.1.2.2 GIS and mountain SAR
In contrast to the mountain SAR studies that rely on aspatial software, a GIS incorporates
the spatial attributes of an incident to provide location-specific pattern and trend analysis. A
review of the SAR literature shows there are only a handful studies incorporating GIS tools to
explore the mountain SAR process, and these studies either employ a GIS for real-time SAR
tasking or for post-task analysis. No academic research on historical mountain SAR incidents to
date has taken a spatiotemporal approach.
GIS tools and products can assist real-time decision making. When GIS tools started to
become mainstream for emergency management, Ferguson (2008) presented a paper at an Esri
Federal User Conference to showcase how a GIS could support a wilderness SAR mission, using
a case study of the search for a missing child with autism in West Virginia. Ferguson had
observed a reluctance by land SAR organizations to use a GIS due to a lack of familiarity with
the tools available to conduct spatial assessments of SAR incidents in a multidimensional
environment. Ferguson demonstrates how different spatial layers, like satellite imagery or trail
networks, can increase an analyst’s situational awareness of the environment for improved
planning operations. He also discussed how some GIS tools can operate in three-dimensions,
potentially identifying locations where two-way line-of-sight communications might not be
possible, and how software extensions available at the time could support the real-time tracking
32
of rescue teams within the GIS representation of reality. While Ferguson does not consider the
after-action analysis of historical SAR incidents, his paper reveals how the adoption of GIS
software by land SAR agencies is a relatively new development.
Maximizing the benefits of a GIS, however, requires some degree of training. Durkee and
Glynn-Linaris (2012) aim to provide SAR teams with some basic training on how to incorporate
a GIS into the SAR process. The authors use the term “wildland SAR” to refer to incidents which
happen in open spaces like parks, wilderness areas, and mountainous terrain. Durkee and Glynn-
Linaris describe how, with proper training, a GIS can help planners and first responders shorten
the time from receiving a distress call to mission resolution in a repeatable and professional
manner. They detail how field operations, data management, planning, and analysis all benefit
from the integration of SAR data in a GIS and the resulting increase in an analyst’s situational
awareness. The ebook is an Esri product that focuses on how to use the MapSAR template in
ArcGIS software, starting with the basics of how to choose coordinate systems, and reviewing
the different types of data a GIS can handle. MapSAR provides integrated layers compatible with
mobile and desktop dashboards for quick visual and statistical analysis, and the template
originates from the National Alliance for Public Safety GIS (NAPSG) Foundation’s WiSAR
project (NAPSG n.d.). Even with the training offered in their ebook, Durkee and Glynn-Linaris
advocate for a GIS specialist to be activated whenever a SAR case opens, and the overall focus
of the ebook is on improving the SAR process through managing and presenting information
real-time, whether you are in an office or the field, rather than on the spatial and temporal
analysis of historical incidents.
The studies to date that have incorporated a GIS to explore historical mountain SAR
incidents are limited by data availability and data quality. In order to use historical incident data
33
as point data in a GIS, the incidents need to be tied to coordinates. This can become complicated
if the source records rely on descriptive locations of where the incident occurred, or if rescue
teams record incidents under different coordinate systems. Doherty et al. (2011) considered a
couple of techniques to georeference historic SAR incidents from 2005 to 2010 in Yosemite
National Park to input the data into a GIS to visualize and assess for spatial patterns and possible
spatial dependence. Their study explored the challenges of conducting spatial statistical analysis
on data that was collected without an anticipation of digitization and the analytical capabilities to
processes large amounts of spatial data.
Doherty et al. (2011, 776) claim they are the “first spatially-explicit study of SAR
incidents.” Using a blend of commercial GIS software and web-mapping applications, the
authors georeference data from Yosemite SAR incident reports using either a ‘point-radius’ or a
‘shape’ method. The former creates a radius of uncertainty around a location based on the
description found in the SAR incident report, which was faster to develop than the latter, and
required careful exclusion of areas that did not fit the description in the incident report. The
authors used the point-radius method for further analysis on 1,356 incidents. Since 95% of the
uncertainty radii were just over 2,000 m, the authors created a two-kilometer grid cell fishnet to
aggregate the data for spatial analysis. Smaller cells, though perhaps more appropriate to
accommodate topographic variation, may not have included the actual coordinates of the incident
within the area of uncertainty. The fishnet came to 1,560 grid cells, or over 6,000 km
2
, with cells
containing an incident count of 0-226 SAR incidents. The authors found statistically significant
clustering amongst the incidents after running the Global Moran’s I statistic, as well as 10 cells
to be statistically significant hot spots based on Getis-Ord Gi* statistic. A visual analysis of the
hot spots in relation to the terrain suggested a correlation between incidents and the Yosemite
34
Valley trails, as well as one backcountry location by a camp. Based on this study, Yosemite SAR
teams will be required to carry GPS devices for accurate location data and integration with the
park’s new digital records management system. The authors recommended future studies
consider temporal as well as spatial uncertainty of historical SAR incident datasets.
The application of a GIS to SAR incidents should match the end users’ needs and
capabilities, particularly since GIS training can be time consuming or require the hiring of a GIS
specialist. Using spatial analysis, Pfau and Blanford (2018) found the Alpine Rescue Team
(ART) were able to complete 75-95% of searches within 6-12 hours, and they recommended
ART personnel leverage the benefits of a GIS for post-mission analysis rather than real-time
application since the ART’s search time is within safe margins. ART is a non-profit group
accredited by the Mountain Rescue Association (MRA), which is a standardizing and
educational agency (MRA n.d.). ART responds to SAR incidents in three counties in Colorado,
an area of approximately 1,309 mi
2
(about 3,400 km
2
). Pfau and Blanford pulled incident data
from 2008 to 2011 from the International Search and Rescue Incident Database (ISRID) to
demonstrate how ART can use a GIS to explore lost person behavior based on historical cases
and improve their SAR process. After cleaning the data, the researchers were left with 133
missions for spatial analysis. Pfau and Blanford used descriptive statistics to explore incident
attributes on subject’s activity and time of year. Incidents that involved a missing person were
further analyzed in a GIS to determine distance, direction, time, and elevation attributes between
lost and found locations. Found locations were further run through KDE to assess the terrain
most common to high density areas. Pfau and Blanford found lost persons traveled on average
4.41 km (SE +/- 1.1) and had an elevation change between lost and found locations of 259 m (+/-
80.1). There was little variation between the type of activity and lost person behavior. The
35
authors described how locating SAR incident hot spots would be useful for training new ART
members and developing relevant training scenarios. Pfau and Blanford recommend their
methods to encourage familiarity with typical mission patterns and, assuming future incidents
follow historic patterns, increase the efficiency and safety margins of non-profit SAR groups
responding to distress calls within their jurisdiction.
The studies reviewed in this section demonstrate the interest in understanding patterns in
historical SAR incidents to increase the safety and efficiency of future SAR response. They also
highlight the growing awareness of what spatial analysis can provide to the SAR planners and
rescue assets tasked with a distress call. What is lacking in the literature to date, however, is an
exploration of mountain SAR incidents over space and time and a consideration of how changes
in spatial and temporal patterns could reveal the influence of sat-comm device activations on
mountain SAR.
2.2 Spatial and Spatiotemporal Analysis
A spatial analysis of mountain SAR incidents provides a means to measure the influence
of sat-comm technology over space. This research relied on three types of spatial analysis
methods: point pattern analysis, where each incident is examined independently; spatial
statistical analysis, where incidents are assessed in aggregate; and visual analysis, where
incidents are inspected for spatial relationships on maps. Spatial analysis methods reveal first and
second order spatial effects that are useful for organizing and interpreting the data and statistical
results. First-order effects describe the influence of topography and environmental conditions on
spatial patterns, while second-order effects represent the patterns formed through incident
interactions (O’Sullivan and Unwin 2010, 163). Mountain SAR incidents are highly unlikely to
contribute to another incident nearby – with the rare exception of when rescue personnel become
36
a SAR incident themselves during the course of a mission – and this study focuses on the
influence of first-order spatial variation in mountain SAR incidents.
Analytical techniques that rely on both spatial and temporal attributes are considered
spatiotemporal. The application of a temporal trend test in conjunction with a spatial statistic
facilitates a spatiotemporal exploration of how sat-comm device activations affect the
distribution of mountain SAR incidents over both time and space. However, spatiotemporal
analysis requires unique structuring of the incident data to run properly in a GIS.
This section of the study gives an overview of the spatial and spatiotemporal analysis
techniques used in the literature to date that contribute to the methodological design of this study.
Several papers that review historical SAR incidents demonstrate how point pattern analysis and
spatial statistical analysis methods may apply to mountain SAR incidents. Since there are no
academic studies to date that have explored the spatiotemporal patterns of mountain SAR
incidents, this paper reviews research on the spatiotemporal distribution of wildfires, as wildfires
– like mountain SAR incidents – are seasonal and often occur in remote environments.
2.2.1 Point Pattern Analysis
Point pattern analyses explore potential second order spatial variation, while also
providing insight into the first order effects on incidents’ spatial distribution. One common point
pattern analysis method to assess whether incidents demonstrate statistically significant
clustering is to compare the actual mean distance between incidents against the expected mean
for the study area. The expected mean between incidents is based on the null hypothesis of
complete spatial randomness. The nearest neighbor (mean) distance may be represented by a
ratio of the actual mean to the expected mean, and it is calculated by:
37

Nearest Neighbor Ratio =
𝑑
̅ !
𝑑
̅ "
=
∑ 𝑑
#
$
#%&
𝑛
1
25𝜌

(1)

where d is the distance between incidents, n is the number of incidents, and 𝜌 is the incident
density for the area considered a possible location for incidents (Clark and Evans 1954). If the
ratio equals one, then there is complete spatial randomness; if the ratio equal zero, there is
complete clustering. The p-values and z-scores associate with this statistic could be used to
compare incidents from different layers within the same area of analysis. However, average
nearest neighbor analysis only describes whether incidents have spatial clustering, not where
those clusters are located.
Another point pattern analysis method that can help identify sites of incident clusters is
kernel density estimation (KDE). KDE uses a kernel function to produce estimates of incident
density for locations throughout a study area. Locations may be based on a grid overlay, and the
density value for each grid depends on a kernel function to weight incidents based on their
proximity to the center of the kernel (O’Sullivan and Unwin 2010, 69). The kernel density
estimator is given by:
𝑓
8
(𝑥) =
1
𝑛𝑟
'
= 𝐾
$
#%&
?
1
𝑟
(𝑥−𝑋
#
B
(2)

where n is the number of incidents within the kernel, r is the kernel bandwidth, K is the kernel
function, x is the grid cell where the function is being estimated, and Xi are the locations of each
observation i (Silverman 1986). The distance selected for r is subjective and depends on the
incidents under examination, though the mean nearest neighbor results can provide a starting
point. Like Azofra et al.’s (2007) zonal distribution model, the kernel function can smooth the
effects of outliers to draw attention to possible cluster locations at selected bandwidths and
38
encourage effective visual analysis. The KDE output is a raster surface of density values that
lends itself to an inspection of incident hot spot locations.
Both nearest neighbor and KDE point pattern analyses feature in the research on the
spatial distribution of wildfire incidents. Wing and Long (2015), in their study on wildfires in
Oregon and Washington from 1984 to 2008, used point pattern analysis techniques to explore
wildfire hot spot locations and global spatial statistics to explore clustering amongst attribute
values. They calculated the centroid for over one thousand wildfires to use as input for average
nearest neighbor calculations, KDE, and quadrat analysis (which compares incident frequencies
across quadrat cells). The resulting nearest neighbor ratio of 0.66 at p < 0.01 suggested
statistically significant spatial clustering of wildfires across the two states, results which were
validated by the quadrat analysis. For KDE, the authors used a quartic kernel function to create a
smoothed density surface, which they used for visual analysis and were able to identify several
wildfire hot spot locations. Wing and Long also found significant clustering of temporal and
climatic wildfire attributes using the Moran’s I and Getis-Ord G global spatial statistics. The
authors conclude that point pattern and spatial statistical analyses can detect changing trends in
historical wildfire spatial patterns, but that future research should incorporate more recent fire
data for comparison.
Aftergood and Flannigan (2022) also used average nearest neighbor analysis to identify
and measure wildfire clusters in their study on 97,921 lighting-ignited wildfires in six provinces
of Western Canada from 1981 to 2018. Only fires from the beginning of April through the end of
September were included for analysis, as these are the official fire season months in the study
area. Instead of evaluating the nearest neighbor distances for all years, as Wing and Long (2015)
did, Aftergood and Flannigan ran nearest neighbor calculations for each wildfire year separately
39
using R Core Team software. The results for all years suggested positive spatial clustering. The
mean values for all years were: a mean nearest neighbor statistic of 154 km, a mean nearest
neighbor ratio for all the years of 0.474, a median z-score of -48.12, and a median p-value of
.0001. The range of average nearest neighbor values was 104 km (1994) to 272 km (2011),
which the authors hypothesize could be related to thunderstorm spread size at 170 km
2
. The
average nearest neighbor results contributed to the conceptualization of wildfire spatial
distribution.
While point pattern analyses are generally considered exploratory, they can be one of the
more viable means to assess spatial patterns if dealing with imprecise data. Koutsias,
Kalabokidis, and Allgöwer (2004) applied KDE for an exploration of wildfire incidents in the
Halkidiki peninsula in Greece, using wildfire ignition data from 1985 to 1995 that had up to 700
m and 925 m inaccuracies in the x- and y-axis respectively. Their intent was to review three
methods of point pattern analysis – quadrat analysis, moving window analysis, and KDE – to
determine which method best accommodated positional inaccuracies in historical wildfire
ignition coordinates. The authors started with a KDE bandwidth equal to the mean distance
between randomly distributed incidents. For a 3,257.63 km
2
study area with 218 wildfire
incidents, the mean random distance is 1,933 m. The authors therefore selected 2,000 m as their
intended bandwidth for exploration, although they also conducted KDE at 1,000, 4,000, and
6,000 m, all using a 250 m grid. KDE was conducted using CrimeStat version 1.1. software.
KDE using the same bandwidths was also conducted on a simulated dispersion of wildfire
ignition points with the same positional inaccuracies as the historical dataset, and the actual and
simulated results were compared. They found the 2,000 m bandwidth to be the best balance of a
large enough distance to capture positional inaccuracies but small enough to prevent over-
40
generalization that would mask the spatial influences on ignition locations. The authors also
found KDE performed the best of the three point pattern analysis methods, as KDE evaluates the
relative position of incidents to each other over space thereby accounting for positional
inaccuracies when set with the proper bandwidth parameter.
2.2.2 Global Spatial Statistics
To conduct spatial statistics that measure attributes over space, incidents need to be
aggregated to create a metric – incident frequency – whereby the spatial relationships may be
evaluated. The aggregation scheme may be based off an irregular polygon pattern, like
enumeration units, or off a repeating grid of squares or hexagons. Incident frequency may then
be compared amongst neighboring units using a neighborhood scheme defined by the analyst.
Global spatial statistics evaluate incidents within the context of the entire study area and
are a useful tool for assessing spatial trends. The Global Moran’s I statistic is used to measure
how observations correlate over space, a concept termed spatial autocorrelation. The Global
Moran’s I statistic is calculated with:

𝐼 =
𝑛
∑ ( 𝑥
#
− 𝑋
D
)
' $
#%&
×
∑ ∑ 𝑤
#(
$
(%&
$
#%&
( 𝑥
#
− 𝑋
D
)( 𝑥
(
− 𝑋
D
)
∑ ∑ 𝑤
#(
$
(%&
$
#%&

(3)

where x is the attribute value at a given location, 𝑋
D
is the mean attribute value across the study
area, wij is the weight of the distance between a location i and its neighbor j, and n is the number
of features within a specified distance band from location i (O’Sullivan and Unwin 2010).
Should the incident frequency at location i and at its neighbor j both be higher or lower than the
mean incident frequency for all features in the study area, then the index is positive, indicating
positive spatial autocorrelation. If xi and xj fall on different sides of the mean, then the index is
41
negative. The degree of spatial autocorrelation is impacted by wij, since it accounts for spatial
proximity. The range of possibilities for I is from -1 to +1, i.e., from complete dispersion to
complete clustering. Should I = 0, there is no spatial autocorrelation, and the distribution of
incidents is considered random. The variance and expected values assume complete spatial
randomness, and the output is the index value, a p-value, and a z-score.
Like average nearest neighbor calculations, the Global Moran’s I statistic indicates
whether incidents exhibit clustering or dispersion across the entire study area without defining
which areas have clusters and which are dispersed. Unlike a nearest neighbor distance, the
distances at which clustering or dispersion is most apparent can be affected by weights assigned
to incidents within a defined neighborhood. Spatial weights represent the impact of incident
interactions. Doherty et al. (2011) in their study of historical Yosemite SAR incidents used an
inverse distance squared approach based on the assumption that the incident frequency at sites
near each other should be more similar than those further away. Based on their 2 km grid of SAR
incident frequency, the Global Moran’s I results indicated statistically significant spatial
clustering, with I = 0.310, a z-score of 24.5, and a p-value < .001. Doherty et al. then ran the data
through the Getis-Ord Gi* local spatial statistic to identify where the grids comprising the
clusters were located since the global version does not describe where the clusters lie, only
whether statistically significant clusters are present in the study area.
2.2.3 Local Spatial Statistics
In contrast to a global spatial statistic, local spatial statistics account for the spatial
heterogeneity of incidents. Two dominant local spatial statistics in the field of geographic
information science are the Getis-Ord Gi* statistic, which helps define incident hot spots, and the
Anselin Local Moran’s I statistic, which can identify outliers in addition to clusters. The GIS
42
tools that use local spatial statistics create layers where each areal unit is assigned to a cluster or
outlier category depending on the z-score and p-value at each location. In this way, the locations
of hot spots, clusters, and outliers may be identified on a map and compared across layers.
The Getis-Ord Gi* statistic helps identify sites with high or low values that lie amongst
neighbors of high or low values, termed hot spots and cold spots respectively. The Gi* statistic is
given as:

and S is the variance (Getis and Ord 1992; Ord and Getis 1995). This version of the Gi* statistic
accounts for the variance and expected values and its output is the z-score. The p-values are
calculated based on a rejection of the null hypothesis that incident frequencies have a random
distribution. In the Gi* statistic, j may equal i, as i can be its own neighbor. The Gi* statistic
does not, however, identify neighborhoods containing spatial outliers since it only gives
significance to locations surrounded by similar incident values.
The Anselin Local Moran’s I statistic enables the categorization of incident sites based on
how their values compare to the values of other sites within the neighborhood. It is given by:

𝐼
#
=
𝑥
#
− 𝑋
D
𝑆
#
'
= 𝑤
#(
$
(%&,(*#
( 𝑥
(
− 𝑋
D
)
(5)

𝐺
#
∗
=
∑ 𝑤
#(
$
(%&
𝑥
(
− 𝑋
D
∑ 𝑤
#(
$
(%&
𝑆
I
𝑛∑ 𝑤
#(
' $
(%&
− ( ∑ 𝑤
#(
$
(%&
)
'
𝑛−1

(4)

where:
𝑆 =
I
∑ 𝑥
(
' $
(%&
𝑛
−𝑋
D
'

43
where:
𝑆
#
'
=
∑ ( 𝑥
(
− 𝑋
D
)
' $
(%&,(*#
𝑛−1

and the summation of distance weights and the incident-frequency attribute are for only the
neighbors of i – i.e., j – and i cannot be considered its own neighbor (Anselin 1995; Ord and
Getis 1995). Like the Global Moran’s I statistic of (3, the local version adapted by Luc Anselin
(1995) measures how the incident frequency at one site correlates with its neighbors, but it does
so at a localized scale. If xi and xj fall on the same side of the mean, either smaller or larger, then
the index is positive. If they are different, as would be seen when a location has a high incident
frequency, but its neighbors do not, then the index is negative. The sign of the index and the
magnitude of the z-score marks two types of clusters and two types of outliers: a high-high or
low-low cluster refers to a location surrounded by similarly high or low values; a location that is
identified as having a high value surrounded by low values is considered an outlier, as is a
location with a low value in a high-value neighborhood. The p-value reflects the likelihood the
spatial pattern is random.
Local spatial statistics can prove useful for detecting fine scale spatial incidents across
broad study areas. Potter et al. (2016) compared the hot spot outputs from the Getis-Ord Gi*
statistic to present their concept of the spatial association of scalable hexagons (SASH), a
technique they found useful for analysts and policy makers alike to communicate the impacts of
an ecological phenomena. The SASH method involved creating scalable hexagons and
identifying statistically significant clusters to track macro-scale patterns. Using this method, the
authors considered three ecological phenomena across broad portions of the United States
collected by three different methods. Data on invasive plants in the south eastern United States
had been collected through ground observations of US Department of Agriculture (USDA) forest
44
plots, and each hexagon cell represented two metrics: native plant species richness and the
percent cover of invasive species. Data on the spread of the mountain pine beetle in forests in the
central and western United States was captured by USDA low-altitude aerial surveys, where
hexagon cell values represented the percent of surveyed forest area that had beetle damage. Data
on wildfires in the central and western United States came from satellite sensing where each
hexagon aggregated the number of fires per 100 km
2
. Data processing and spatial statistics were
conducted in Esri’s ArcMap 10.1.
In order to restrict the study area to environments where the phenomena could occur,
Potter et al. only evaluated forested areas based on ArcMap’s forest cover layer. The size of the
aggregating hexagon unit was determined through phenomena-appropriate methods: invasive
species hexagons were near the average size of US southern counties (1,452 km
2
) to align with
USDA program goals; a mountain pine beetle generally moves within three kilometers, so the
hexagon size was 54 km
2
with about 3.8 km from the center to the edge; the authors used
semivariograms to test the spatial autocorrelation of wildfires aggregated to 54 km
2
hexagons,
settling on a hexagon size that was as wide as the spatial autocorrelation range, at 635 km
2
. The
local neighborhood considered by the Gi* statistic encompassed the 18 first and second order
neighbors around each hexagon. Potter et al. found their SASH method produced meaningful
patterns that could be applicable to policy decisions and organizational tracking of ecological
threats. The authors mention that, while Local Moran’s I statistic can reveal outliers while the
Gi* statistic cannot, they felt the Local Moran’s I statistic’s inability to capture the impact of
spatial autocorrelation at values near the mean to be a disadvantage and thus did not test for
outliers in their SASH demonstration.
45
2.2.4 Spatiotemporal Analysis
While spatial analysis can help agencies assess the distribution of incidents,
spatiotemporal analysis allows agencies to tailor policy to reflect recent trends. One method to
evaluate an incident’s spatial relationship over time is through the construct of a space-time cube
(STC). An STC incorporates time as a third dimension, allowing incidents to not only be binned
by location, but also by time, creating an array of values suitable for analysis with temporal and
spatial neighborhoods.
STCs have supported spatiotemporal research in fields as diverse as human behavior after
an earthquake (Gismondi and Huisman 2012), crime patterns (Nakaya and Yano 2010), public
health (Nielsen et al. 2019), cetacean strandings (Betty et al. 2020), and tornados (Allen et al.
2021). The field of wildfire management has benefited from the STC construct since accessible
STC-building tools became available with ArcMap 10.3 in December 2014 and ArcGIS Pro 1.0
in January 2015 (Esri n.d.). To date, there are several studies that have explored historical
wildfire incidents using spatiotemporal analysis methods facilitated by the STC design.
One spatiotemporal analysis tool commonly employed in the research on wildfire
incident patterns is Emerging Hot Spot Analysis (Esri n.d.). The Emerging Hot Spot Analysis
tool compares incident hot spots over time using the Getis-Ord Gi* local spatial statistic and a
trend test. The tool uses the Mann-Kendall trend test to evaluate hot spot bins against their
temporal neighbors to determine how past incident patterns relate to more recent incident
distributions. The Mann-Kendall trend test is given by the statistic:
𝑆 = = = 𝑎
#(
$
(%#, &
$-&
#%&

(6)

46
and where xi and xj are ranked in the time series. The values of newer and older bin pairs are
summed to represent a trend (Mann 1945; Hamed 2009). An S > 0 would indicate a positive
trend while the opposite would indicate a downward trend. If S = 0, then the trend is neither
constantly increasing nor decreasing and the null hypothesis is met. Depending on the trends
detected at an STC location over time, the Emerging Hot Spot Analysis tool assigns one of eight
trend types: new, consecutive, intensifying, persistent, diminishing, sporadic, oscillating, and
historical (Esri n.d.). These categories provide analysts with a means to ascribe the relative
importance of incident patterns within a study area.
Similar to the near-precise location information sat-comm devices provide to SAR
agencies, satellite-based sensors used to detect fires offer accurate fire location data to
researchers interested in fire patterns. Reddy et al. (2019) took fire detection data collected by
moderate resolution imaging spectrometer (MODIS) sensors on board NASA satellites to
examine emerging hot spots of forest fires in South Asia from 2003 to 2017. The authors
restricted the one square kilometer MODIS fire raster to only those areas with forest cover,
resulting in 522,348 fire incidents considered in the study. Only the months that had a greater
than 2% contribution to a nation’s annual fire count were used for descriptive statistical analysis.
Using ArcGIS software, the authors aggregated the incidents by five-kilometer grid cells and
grouped by year to create the dimensions of an STC. The authors also selected five kilometers as
the neighborhood distance to detect emerging hot spots and capture local trends. Within the
study area, they found over 30% of fires from the 15-year time span were sporadic hot spots,
making it the dominant category, and just under 8% of fires were considered new hot spots at the
where: 𝑎
#(
= 𝑠𝑖𝑔𝑛N 𝑥
(
− 𝑥
#
O = P
1
0
-1

𝑥
#
< 𝑥
(
𝑥
#
= 𝑥
(
𝑥
#
> 𝑥
(

47
low end. The authors acknowledged that a change in neighborhood size could dramatically alter
the results, but as their intent was to identify areas for forest conservation, they decided five
kilometers was optimal. However, the impact of neighborhood size on spatial and spatiotemporal
statistical analysis highlights the highly subjective nature of these methods.
Spatiotemporal analysis can advance studies seeking to assess why historical wildfires
occur in some locations and not others. In their exploration of how bushfires in New South
Wales, Australia might evolve in response to controlled burns, Visner, Shirowzhan, and Pettit
(2021) looked at data spanning 100 years of fire history using a mix of regression analysis,
correlation analysis, and spatiotemporal analysis. The authors ran a general linear regression
model and bivariate Pearson correlation statistical analysis in R statistical software, which can
handle geographic data, to identify relationships between older controlled burn areas and newer
bushfires. They then created an STC in ArcGIS Pro and used the Getis-Ord Gi* statistic and
Mann-Kendall trend test via the Emerging Hot Spot Analysis tool to identify emerging hot spots
for visual data mining. Fire polygon centroids were converted to points, and then aggregated by
municipality. The authors selected this unit of aggregation since municipalities are responsible
for fire mitigation efforts. Visner, Shirowzhan, and Pettit visually inspected the emerging hot
spots for all years, as well as for each individual year between 2010 through 2020. They
concluded most municipalities had sporadic bushfire incidents, although the authors were able to
identify four new municipality fire hot spots from the 2019 to 2020 fire season. While the
authors did not find a correlation between controlled burns and bushfires, they did find a positive
trend in the total number of bushfires occurring in New South Wales. To make their findings
accessible to the public, the authors published their findings as an Esri dashboard.
48
While fires are often characterized by burn area, fire point incident data could be useful
when pulling from a large historical dataset where location information originates from a variety
of sources. Aftergood and Flannigan (2022) used spatial and spatiotemproal pattern analysis to
explore 97,921 wildfires incidents representing the central ignition location of a lightning strike
in six provinces of Western Canada from 1981 to 2018. Only incidents occurring from the
beginning of April through the end of September were included for analysis, as these were the
official fire season months in the study area. Within ArcGIS 10.7, incidents from the fire season
and each summer month were aggregated by a grid of 30 km-wide hexagons, as this size
hexagon produced the most robust results during ESDA. The authors selected a yearly time unit
when creating the STC of the incidents, which mitigated the effects of seasonality for trend
analysis. Aftergood and Flannigan found the Mann-Kendall trend statistic revealed an overall
non-significant negative trend for the total number of incidents for all layers, though there were
areas with significant increasing and decreasing trends within the provinces. These patterns
varied in location across the summer months. The authors considered how data quality issues
related to working with a historical dataset composed of different sources of incident locations
could contribute to inaccurate results, explicitly mentioning how data from recent years might be
biased due to technological improvements that are able to capture incidents that were previously
overlooked. However, the authors felt they were able to successfully demonstrate how trend
analysis of spatial data using an STC enables an intuitive visualization of regional trends and
consolidates the variation of incidents over time into accessible map imagery.
2.3 Summary
This research aims to identify the influence of a sat-comm device usage on the spatial
distribution of mountain SAR incidents. However, when reviewing the related literature that
49
incorporates an analysis of historical SAR incident data, it was found that very few studies use
spatial statistics to explore the patterns of mountain SAR incidents. There is also a gap in the
SAR research examining location-specific temporal trends. Maritime SAR is the dominant genre
advancing the literature on the analysis of historical SAR incidents. While several studies
consider the temporal attributes of maritime SAR incidents through graphics and visual analytics
software, no maritime nor mountain SAR studies to date incorporate spatiotemporal analysis
enabled through the STC construct, possibly due to tradeoffs between the relatively large study
area sizes, fine scale spatial phenomena, and computational processing times.
As addressed by Ferguson (2008), Durkee and Glynn-Linaris (2012), and Pfau and
Blanford (2018), the low number of wilderness SAR studies incorporating a GIS, which
facilitates spatiotemporal analysis, is likely due to SAR agencies’ lack of familiarity with the
GIS toolsets and application. Since this paper aims to demonstrate how a combination of spatial
and temporal analyses can inform SAR organizations for safer and more efficient operations,
studies on the spatiotemporal patterns of wildfires were reviewed for their capabilities and
limitations. Several studies on wildfires successfully applied the STC construct and ran spatial
trend analysis at relatively large scales using a variety of spatial and temporal aggregation
schemes selected by the authors. This study takes the lessons offered in these wildfire studies and
applies them to mountain SAR incident analysis in order to explore the impact of sat-comm
devices activations on mountain SAR operations.

50

Chapter 3 Methods
The intent of this research is to explore the impact of satellite communication (sat-comm)
devices as a notification method on the spatial and temporal patterns of mountain search and
rescue (SAR) incidents in California’s Sierra Nevada mountains in order to help SAR agencies
and rescue teams prepare for future incidents. SAR organizations can use the methodology
presented in this research to assess the influence of sat-comm device activations within the
context of their sphere of influence, adjusting the allocation of resources and tailoring training
plans, as required.
The methods in this research involved three main components: data preparation, spatial
and spatiotemporal statistical analysis of incidents, and an evaluation of incident attributes.
Figure 5 presents the flow of this study and the data associated with each step. The first steps in
the methodology involved data preparation so only incidents that met the definition of mountain
SAR and occurred within the study area boundary would be considered for analysis. This study
then evaluated the CALOES mountain SAR incidents for first-order effects in ArcGIS Pro 2.9
(Esri 2021), the results of which informed the creation of neighborhood parameters. This study
used GIS tools relying on local spatial statistics and the newly defined spatial neighborhood to
reveal the presence of incident hot spots, clusters, and outliers. Developing a space-time cube
(STC) of the CALOES incidents provided the structure for spatiotemporal analysis. Lastly, the
AFRCC PLB activations were evaluated against the CALOES results through visual analysis and
descriptive statistics, as were all accidental sat-comm device activations. The comparison of
statistical results alongside an evaluation of incident attributes over time and space enabled a
51
holistic review of how sat-comm device-initiated mountain SAR incidents impact the traditional
SAR process.

Figure 5. An overview of the study's methodology

3.1 Data
This research drew from two datasets to adequately encompass the size of the study area:
one is a subset of a national-level record of personal locator beacon (PLB) activations from the
Air Force Rescue Coordination Center (AFRCC), and the other comes from the California Office
of Emergency Services (CALOES) and covers all SAR incidents originating with multiple
notification methods within the state of California. This research only ran spatial statistics on the
CALOES mountain SAR incident dataset, as it provided enough incidents to perform meaningful
analysis, whereas the AFRCC dataset was too sparse; to have statistical significance, the
distribution of incidents needs to reject the null hypothesis that incidents arise due to random
chance, which requires enough incidents associated with the physical landscape to not appear
random. Likewise, the CALOES dataset was the only input for spatiotemporal analysis. The
AFRCC’s PLB activation records supplemented the conclusions drawn from the CALOES
statistical results, and both datasets were compared through visual analysis and descriptive
statistics of incident attributes. In this way, a state- and national-level dataset complemented each
other in an exploration of sat-comm device activations across a geographic region.
52
The study area’s boundary was based off wilderness area shapefiles accessed through
Wilderness Connect (n.d.), from which the High Sierra wilderness areas were isolated and
projected into the projected coordinate system used in this study. The state and county lines that
were included in this study’s maps to orient the target audience came from a national boundary
shapefile from the USGS national map downloader (USGS n.d.). The state and county lines also
required projection.
Spatial and temporal attributes were added to the datasets to explore the environments in
which SAR incidents occur. Elevation data came from a 30 m digital elevation model (DEM)
from Esri’s World Elevation services (Esri n.d.), which sources US elevation data from the US
Geologic Survey (USGS n.d.). Elevation impacts rescue team performance capabilities: SAR
helicopter engines have degraded performance as elevation increases, and higher elevations
increase the risk of ground teams and subjects experiencing the effects of hypoxia and exposure.
This research determined incident notification time of day based on incident location and
apparent sunrise and sunset times through the National Oceanic and Atmospheric
Administration’s online solar calculator (NOAA n.d.). Four categories represented the
notification time of day: within an hour prior to sunrise, day, within an hour prior to night, and
night. The time of day when SAR agencies receive incident notification impacts rescue team
preparation. If notification is received within an hour to sunrise, there is a good chance contact or
rescue will occur during daylight hours. If notification is within an hour of sunset, then the SAR
process will most likely unfold during periods of darkness when specialized gear like night
vision goggles (NVGs) or infrared (IR) systems are required. This additional data on elevation
and time of day enabled another avenue to explore the impact of sat-comm devices on SAR
incident operating conditions through descriptive statistics.
53
3.1.1 Dataset Acquisition
The AFRCC dataset structure arrived based off the specifications made in a Freedom of
Information Act (FOIA) request to the US Air Force. Specifications included a spreadsheet
format of PLB activations in California’s Sierra Nevada mountain range from 2015 through the
date of request (i.e., September 24, 2022), as the request could not exceed seven years of records.
The FOIA request was also for the following incident attributes: coordinates, date, time,
responding assets, accidental or intentional PLB activation, and mission outcome. The FOIA
process took about two months, mainly due to the time required to search through individual
reports of PLB activations and verify incidents fell within the study area dimensions placed in
the FOIA request, since the PLB activation records are not stored in an easily searchable
database. Upon receipt, the AFRCC dataset appeared as in Figure 6, and it included 148 PLB
activations spread out across much of California. The dates of the PLB activations ran from
March 24, 2015, through September 11, 2022.
54

Figure 6. A map of the AFRCC dataset records

Acquiring the CALOES dataset involved a Public Records Act request, the processing of
which took about a month from the date of request on September 16, 2022. The request was
made for the same information as the AFRCC FOIA request except for dates: CALOES started
collecting state-wide SAR incident data from the counties in 2018, so January 1, 2018, was the
beginning date of the requested data range. CALOES provided a dataset output from ArcGIS
Survey123, a software product that integrates mobile device applications, desktop programs, and
online tools to create “surveys” that SAR teams may use to fill out an incident report (Esri n.d.).
The dropdowns for the different requirements of the report specify which information to include.
55
For the temporal information on “Incident Start Date and Time,” the guidance is, “Choose an
approximate time when the mission was initiated (Earliest action recorded, e.g., 911 phone call,
team callout, etc.).” An incident’s coordinates come from either the recording party inputting the
coordinates directly, providing a place name attached to coordinates and searching for the
location in the software, or moving a pin on a map that identifies the initial planning point for a
search, the site of injury, or the location of rescue or contact. Accidental sat-comm device
activations were annotated as such. The dataset sent from the CALOES is depicted in Figure 7. It
contained 1,679 incidents spanning the entire state, from January 1, 2018, to July 24, 2022.

Figure 7. A map of the CALOES dataset records

56
3.2 Dataset Preparation
Both the AFRCC and CALOES SAR datasets required review and refinement to make
sure the data met the definition of a mountain SAR incident and fell within the study area.
Datasets were brought into Microsoft Excel (2022) to standardize attribute categories and to
remove incidents with inaccurate or incomplete spatial and temporal attributes. The datasets
were then brought into ArcGIS Pro, projected, clipped within the study area boundaries, and
inspected for inaccuracies and redundancies. This research then split the dataset into categories
depending on if they were an actual mountain SAR incident, an actual mountain SAR incident
beginning with intentional sat-comm device activation, or an accidental activation of a sat-comm
device in order to facilitate the evaluation of analytical results.
3.2.1 Dataset Preparation in Excel
Bringing the datasets into Excel enabled an expeditious review of both datasets for
completeness as well as attribute standardization. Incidents that occurred in counties outside of
the study area or lacked coordinate or temporal information were removed. Attributes were
standardized across both datasets, with several fields added for temporal analysis. Additional
attribute fields resulted from isolating the day of the week, month, and year from the date-time-
group. For the AFRCC dataset, the dates and times needed to be converted to California’s local
time to match the CALOES dataset, as they arrived in Zulu time (i.e., the time based off the
prime meridian).
This research needed to refine the CALOES dataset to only include incidents that met the
definition of mountain SAR; the AFRCC PLB activations did not require any adjustment in this
respect. This study defines a mountain SAR incident as a land SAR event which occurs in
mountainous terrain away from the built environment, where the subject is participating in
57
outdoor recreation, and which requires the assistance of specialized SAR assets that must
maintain specific qualifications and training. Incidents that involved natural disasters, law
enforcement, body recoveries, searches, and suicides were removed. While the argument could
be made that body recoveries could reveal a pattern of hazardous areas, the comments for the
recovery cases revealed a variety of missions that did not necessarily meet the mountain SAR
criteria. For example, some recovery cases had a distress call go out, but rescue teams arrived to
find the subject dead-on-arrival, but others involved the discovery of skeletal remains – not
necessarily human. Since the comments were too incomplete to appropriately categorize body
recoveries as a mountain SAR incident or not, they were all excluded from analysis.
In the CALOES dataset, incidents that originated with a sat-comm device activation
could be labeled in the ArcGIS Survey 123 report as “PLB activated” or “SEND activated.”
However, several incidents’ comments indicated the appropriate labeling was not always used,
with incidents originating with a SEND product (e.g., the comments state the subject used a
Garmin InReach) getting labeled as PLB-activated. Thus, a new field was made to mark an
incident as either originating from a sat-comm device or not, rather than relying on potentially
inaccurate PLB and SEND labels. Incidents that did not originate with a sat-comm device
activation stemmed from an ‘other means of notification,’ a description used henceforth in this
paper to account for overdue procedures, verbal, and cellular methods of notification.
After cleaning, this research divided the datasets into separate layers based on their
method of notification and whether sat-comm device activation was intentional or accidental.
The breakdown of dataset preparation and incident categorization is depicted in Figure 8. For the
CALOES dataset, this resulted in four layers, while for the AFRCC dataset on PLB incidents,
this resulted in two layers. In this way, accidental sat-comm device activations could have their
58
locations and attributes assessed, and actual SAR missions could be compared between sat-
comm device usage and other means of communication.

Figure 8. Initial dataset preparation and organization in Microsoft Excel

3.2.2 Dataset Preparation in a GIS
To be used as input for spatial analysis, the mountain SAR incidents from both datasets
needed to represent locations on a planar surface for distance measurements. The original
coordinates in the CALOES and AFRCC datasets were in the WGS 1984 geographic coordinate
system and required projection in ArcGIS Pro. The WGS 1984 coordinate system can support
meaning visual analysis since it is a realistic representation of spatial relationships. However,
since geographic coordinate systems are based on a three-dimensional representation of the
earth, a projected coordinate system that presents a planar version of reality is required to
measure the distance between points or features during spatial statistics. This research selected
the California (Teale) Albers in NAD 1983 (meters) projected coordinate system based on
guidance set by the CDFW. The CDFW recommends using this modified Albers Conical Equal
59
Area projection for statewide datasets that require accurate distance and area measurements
(CDFW 2022). The origin of the quadrants defining the two-dimensional landscape is roughly
centered in the middle of the High Sierras, and measurements made throughout the study area
can be considered relatively accurate representations of reality since distortions increase with
distance from the origin.
Since mountain SAR incidents cannot occur in the parts of the study area devoted to
roads, housing, towns, and industry, this research needed to define a study area that would only
include land where a mountain SAR incident could occur. To this end, a shapefile representing
the wilderness areas within the High Sierras (Wilderness Connect n.d.) was brought into ArcGIS
Pro to effectively eliminate non-mountain SAR incidents and provide a landscape with the
potential for mountain SAR incident distribution. A five kilometer buffer was made around the
wilderness areas’ boundaries in order to capture incidents that bled beyond the boundaries and to
account for edge effects during subsequent spatial analysis. This buffer distance was based on
the finding by Pfau and Blanford (2018) that lost people in the mountains travel on average 4.41
km. While the present study does not take into consideration searches and lost person behavior,
Pfau and Blandford’s research offers insight into how far subjects might wander before finding
themselves in distress and initiating a call for help.
The mountain SAR incidents from the CALOES and AFRCC datasets were clipped to
within the boundaries of the five kilometer wilderness area buffer and assessed for overlap
within and between the two datasets. All incidents sharing the same date and time were visually
inspected for spatial proximity, and if enough information was available in the comments to
verify they recorded the same incident, then one was removed. Figure 9 depicts the selection
process amongst CALOES incidents. In this example, all the incidents originated with a means
60
of notification other than a sat-comm device. If incidents had the same date and time, but not
enough detail was available in the comments to determine if they were unique or separate (e.g.,
through demographic information), then both incidents were retained for analysis. Of note, one
of the comments for the records sharing dates and time in Figure 9 did have demographic
information, but it is not included in the figure to protect the subject’s privacy. Since the other
incident sharing date and time did not have demographic information that could confirm or refute
the match, both were retained per this example.

Figure 9. Depiction of incidents that were removed or retained

This research likewise identified and reviewed all incidents that shared geographic
coordinates within and between the two datasets to eliminate redundancies which would impact
analyses based on distance measurements. It is also almost impossible for incidents to occur in
the exact same location, since coordinates were accurate to at least 0.1 meter. If one of the
incidents could be determined by its attributes to be an accurate representation of a mountain
61
SAR incident, it would be retained while its duplicate or triplicate was removed. Otherwise, this
research removed all incidents in that location. For example, if two incidents appeared in the
middle of a lake under the same coordinates, but had different dates, times, and non-water sport
activities per the comments, then were both removed. Not a single AFRCC mountain SAR
incident shared the same coordinates nor same date and time as a CALOES incident. In theory,
all the AFRCC records from January 2018 through July 2022 should match a CALOES record,
since the AFRCC passes the notification of a SAR incident to the appropriate local SAR agency
(e.g., county or NPS), who in turn should record the incident and provide that information to
CALOES (reference Figure 3 in Chapter 1 above). Several CALOES sat-comm device
activations (which are a mix of PLB and SEND activations) and AFRCC PLB activations lay
near each other in time and space (see Figure 10), but due to insufficient information, all were
retained. Note in the example in Figure 10 the similarities between the CALOES local time and
AFRCC Zulu time. This was a reoccurring observation in this study, possibly pointing to a
miscommunication of date and time formats between agencies.
62

Figure 10. Example of similar incidents that were retained

This research also removed incidents from the five kilometer buffer area if ArcGIS Pro’s
satellite imagery basemap revealed they were near the built environment, such as paved roads,
parking lots (a common trail head feature), or lodging. Since wilderness areas are largely devoid
of man-made infrastructure per federal regulations, a visual inspection was not required. The
exception to the built environment restriction were incidents that occurred in winter months
when there was a high chance of snow (November through March), since several paved
mountain roads close through the winter but are available for other forms of recreation, legal or
not. This research also retained OHV incidents, as they would likely require the assistance of a
specialty-trained SAR team and would not be accessible to standard vehicles or ambulances.
This research considered the entirety of the wilderness areas a potential distress location, since
even though humans tend to follow trails (Doherty et al. 2011), they are a wily species that tend
to venture from the beaten path – or fall off it.
63
Once the feature layers only contained mountain SAR incidents within the study area,
additional spatial attributes were added using the ArcGIS Pro Summarize Elevation tool. This
tool assigns elevation, slope, and aspect to each incident based on a DEM selected by the analyst.
This research selected the 30 m DEM for reference, as it is the size that most closely resembles
the area of uncertainty associated with coordinates coming from sat-comm devices: some PLBs
set satellite position accuracy at 100 ft, which corresponds to 30.48 m (US Air Force n.d.). This
research elected to retain only the elevation output for further exploration, as slope and aspect
can vary dramatically within short distances and a single value measured at the incident-level
could be an inaccurate representation of surrounding topographic challenges. In contrast, a single
measurement can adequately represent the effects of elevation on rescue team capabilities. Table
1 presents the attributes available for analysis in each feature layer now that temporal attributes
had been added in Excel, and spatial attributes had been added in ArcGIS Pro. The only
attributes that did not overlap across datasets were the Zulu dates and times in the AFRCC data.
Table 1. The attributes associated with each dataset available for subsequent analysis
Dataset Attributes
CALOES and AFRCC
Date, Time, Time of Day, Year, Month, Day
of the Week, Mission Type, Notification
Method, Accidental Activation, Elevation,
Comments, Latitude, Longitude
AFRCC Only Zulu Date, Zulu Time

3.3 Spatial Analysis of the CALOES Dataset
The spatial analysis of mountain SAR incidents involved two main steps: point pattern
analysis and spatial statistical analysis. Due to the sparse distribution of AFRCC mountain SAR
incidents, this research only conducted statistical spatial analysis on the CALOES-derived
incidents. Figure 11 provides an overview of the analytical techniques used in this research and
64
the ArcGIS Pro tools applied to execute them. The point pattern analysis methods of nearest
neighbor measurements and kernel density estimates (KDE) provided a means to measure
incident spatial interactions. Point pattern analysis results, backed by expert opinion, aided in the
development of a neighborhood structure. Neighborhoods were structured off hexagonal grids,
and incidents were aggregated by grid cell, creating a metric for comparison: incident frequency.
Once this research had a neighborhood structure defined, local neighborhoods were compared to
the study area as a whole and tested against random simulations to determine the significance of
first-order spatial interactions that could reveal hot spots, clusters, and outliers.

Figure 11. Overview of the methods and corresponding ArcGIS tools used for analysis
65

3.3.1 Point Pattern Analysis
This research evaluated all mountain SAR incidents by a single set of coordinates without
considering the area of positional uncertainty inherent to satellite devices caused by signal
obstructions in the environment. While positional uncertainty could have been represented using
the point-radius technique described by Doherty et al. (2011), this study maintained a single
coordinate pair rather than an area to represent sat-comm device activations for three reasons:
satellite positional uncertainty is not uniform and depends on the topography; positional error
varies by device and make-and-model attributes were unknown; and if areas of uncertainty were
created, polygon centroids would still form the basis for point pattern analysis and incident
aggregation due to their computational efficiency.
Evaluating the distances between mountain SAR incidents can offer insights into how
incidents interact with each other over space. While such second-order effects were not expected
amongst mountain SAR incidents, exploring the distance relationships between incidents gave a
sense of their spatial distribution and helped define the dimensions of repeat sites for rescue
teams. This research conducted average nearest neighbor analysis for an initial assessment of
spatial clustering. Since the layer with all the actual mountain SAR incidents would have the
largest minimum bounding rectangle to run the nearest neighbor statistic, the area of this
rectangle was also applied to the layers representing only sat-comm devices or only other means
of notification. Additionally, the distances between an incident and n nearest neighbors – where
n in this case represents the number of neighbors when the minimum distance exceeds a
reasonable definition of a single topographic area – were calculated for the CALOES layer
representing actual mountain SAR incidents, since it was the layer containing the most incidents.
66
Comparing the minimum, average, and maximum distance values provided initial insights into
how incident clusters and outliers might be spaced.
This research next examined incidents for first-order effects using KDE. The kernel size
which made the most sense based on the underlying topography became a reference for spatial
neighborhood analysis. The KDE output is a raster surface, where each raster cell value
represents the density of incidents per the search distance set by the analyst. The quartic kernel
function that defines the shape of the kernel smooths the transition between cell values for a
visually appealing output that facilitates the visual analysis of point incidents across a broad area.
Cells near the edges of the study area might exhibit lower values than interior cells since fewer
incidents might lie within the kernel search distance. This research mitigated this edge effect by
the 5 km buffer around the wilderness areas that accounted for incident bleed beyond the study
area boundaries. This research explored several kernel search distances to assess which distance
resulted in clusters that corresponded to the topographic environment. Since KDE is less
computationally demanding than other forms of spatial statistical analysis, the KDE output
refined the range of distance parameters to explore in future computations.
3.3.2 Incident Aggregation and Neighborhood Structure
This research next aggregated the mountain SAR incidents so spatial statistical analysis
could be used to assess the statistical significance of the incidents’ distributions. The aggregating
unit needed to represent a site that rescue teams would associate with similar operating
conditions and consider a repeat location. For example, reoccurring SAR calls for a steep trail
section is an identifiable location to which rescue teams could develop realistic training
scenarios. Multiple incidents at that site will have different coordinates, but the site would
present similar hazards to a rescue team, like the steepness of the slope, the type of ground cover,
67
or the elevation. Hexagons were chosen as the aggregating unit because they reflect human
movement patterns more realistically than a rectangular grid and because hexagonal edges
conform better with terrain than right angles (Birch, Oom, and Beecham 2007). Expert opinion
supplemented the point pattern analysis results when settling on hexagon size by considering
how the size related to the ground and airborne rescue teams who must contend with reaching or
moving subjects in distress. The hexagon size needed to be small enough to capture topographic
variation (e.g., a climbing route, a switchback on a trail, an alpine meadow, etc.), large enough to
mitigate computational demands, and reasonable enough to represent rescue team capabilities.
Mountain SAR incidents were aggregated by grid cell, resulting in a single layer containing three
fields of incident frequencies: all actual mountain SAR incidents; actual SAR incidents that
began with a sat-comm device activation; and actual SAR incidents that originated from a
notification source other than a sat-comm device.
While the size of the hexagon reflects a single operating site for rescue teams, the size of
a SAR incident neighborhood represents a broader area of concern for SAR agencies that could
correspond to a recreational destination or pathway. Mountain SAR incidents, which have spatial
attributes related to the terrain, are not expected to occur randomly across wilderness areas. An
OHV area would presumably contain mostly OHV-related accidents, while a rock-climbing area
would have mainly technical SAR incidents. Busy areas could have a high number of cases due
to a mix of recreational experience levels. SAR incidents near each other in space will likely
have more similar attributes than those farther away. This correlation of attributes over space is
termed spatial autocorrelation. Tests for spatial autocorrelation help identify the typical
neighborhood size of mountain SAR incident locations.
68
This research used the Global Moran’s I statistic to measure the spatial autocorrelation of
mountain SAR sites based on their incident frequency across the Sierra Nevada wilderness areas.
Running the statistic for multiple distance provided a series of z-scores. A fixed distance band
was selected to conceptualize spatial relationships so as to explore multiple distances to detect
trends, since mountain SAR incidents are not known to cluster within a specific distance range.
The distance band where the index has the largest peak z-score represented the mountain SAR
incident neighborhood size where SAR incident frequency deviates the most from the mean. This
research used this distance to define subsequent neighborhood relationships.
3.3.3 Hot Spot, Cluster, and Outlier Analysis
Mountain SAR incidents play out across a diverse topographic environment, and a single,
global trend cannot capture local spatial relationships within the dataset. Two local spatial
statistics were used in this research to investigate where mountain SAR incidents occur and in
what local context: the Getis-Ord Gi* statistic and Anselin Local Moran’s I.
This research applied the Gi* statistic to identify concentrations of mountain SAR
incident hot spots that significantly differ from the rest of the study area. Hot spots are a useful
tool for visual analysis, as they reveal statistically significant areas that should concern SAR
organizations and policy makers. Hot spots are also easier to conceptualize and locate than
individual incident points on a map, and they are more compelling than a simple density analysis
since their significance can be measured. The Getis-Ord Gi* results lent themselves to a
comparison of how sat-comm device activation hot spots compare to those based on incidents
originating from other means of notification.
This research evaluated mountain SAR incident spatial clusters and outliers for
significance using the Anselin Local Moran’s I statistic. The neighborhood distance parameter
69
was the same as that set for the Gi* statistic to support evaluation between the results. The
probability a mountain SAR incident site is an outlier or belongs to a cluster is given by the p-
value after running a set number of permutations of the Monte Carlo test to reject the null
hypothesis that the attributes are randomly distributed. If the Anselin Local Moran’s I values for
the Monte Carlo permutations suggest less spatial clustering than the Anselin Local Moran’s I
values of actual mountain SAR incident distributions, then the actual distributions are registered
as significant clusters. More permutations increase the precision of the pseudo p-value from the
Monte Carlo permutations, but also require more processing time (Esri n.d.). Due to
computational constraints, this research only ran 499 permutations, limiting the pseudo p-value
threshold to p = .002. Because outliers may develop into a hot spot over time should more
incidents occur in that area, they are worth investigating using the Anselin Local Moran’s I
statistic so SAR agencies may monitor for the evolution of future patterns.
3.4 Spatiotemporal Analysis of the CALOES Dataset
This research conducted spatiotemporal analysis on the CALOES mountain SAR
incidents using the Emerging Hot Spot Analysis tool in ArcGIS Pro (Esri n.d.). This tool
combines the Getis-Ord Gi* statistic with the Mann-Kendall trend test for a spatiotemporal
consideration of mountain SAR incidents. Considering how the distribution of incidents changes
over time allows SAR agencies to distinguish between older and more recent patterns which
could reflect changes is outdoor recreation driven by sat-comm device usage.
For the Emerging Hot Spot Analysis tool to work, SAR incidents need to be structured by
space and time. A space-time cube (STC) provides the format, with two dimensions measuring
space and a third dimension measuring time. Figure 12 offers a visualization of the STC
construct, adapted from Esri (n.d.). The area of analysis forms the base of the array, and layers
70
representing a time series stack over the spatial base. Each slice of the time series represents one
month of aggregated incidents due to the data’s irregular start and end dates. While aggregating
incidents by year would have been preferable to mitigate seasonal bias (see Aftergood and
Flannigan 2022; Reddy et al. 2019; and Visner, Shirowzhan, and Pettit 2021), the Create Space
Time Cube by Aggregating Points tool requires a minimum of ten time slices to run trend tests
even though only four time slices are required to detect a trend. Time slices are defined by the
last day of the month of the most recent month with complete SAR data. Hexagonal prisms
match those used with the local statistics, although they are depicted as cubes in Figure 12 for
ease of visualization.

Figure 12. A graphical representation of a Space-Time Cube; adapted from Esri (n.d.)

The Emerging Hot Spot Analysis tool considers a location in the context of its spatial and
temporal neighborhood. The global attribute mean describing SAR incident frequency (i.e., the 𝑋
D

71
in (4) was adjusted to only consider attributes within the same temporal neighborhood, which is
set at 12 months. To keep the Emerging Hot Spot results comparable to the purely spatial Gi*
results, the spatial neighborhood fixed distance band is the same, despite potential biases
associated with fewer incidents assessed per neighborhood per temporal period when running
Emerging Hot Spot Analysis.
The Emerging Hot Spot Analysis tool’s output is a time series of hot spots and cold spots
based on the p-values and z-scores of the locations containing SAR incidents. The Emerging Hot
Spot Analysis tool applies the Mann-Kendall trend test to the hot spot and cold spot time series
to ascertain whether there is a positive or negative trend. The Mann-Kendall test is a non-
parametric rank correlation test, meaning it does not need data to be normally distributed and is
suitable for the right-skewed mountain SAR incident frequencies. Because the test looks for
consistently increasing or decreasing trends over time, seasonal data would need to be considered
by year, which is why the temporal neighborhoods are a twelve-month aggregate. The tool can
assign one of eight trends: new, consecutive, intensifying, persistent, diminishing, sporadic,
oscillating, and historical (Esri n.d.). In this manner, an analyst can identify areas that could be a
consistent or growing concern to rescue teams, particularly owing to the use of sat-comm devices
by outdoor recreationists.
3.5 Comparing the AFRCC and CALOES Datasets
The AFRCC PLB dataset augmented the CALOES SAR dataset in that it contains older
recorded incidents–and additional, contemporary data points–that support or question the spatial
patterns discovered amongst the CALOES SAR incidents. This research first examined the
spatial and temporal similarities and differences between actual SAR missions from the two
datasets through visual analysis and descriptive statistics to provide an overview of the spatial
72
and temporal attributes of each dataset. This research then inspected the PLB activations from
the AFRCC dataset in the context of the statistically significant spatial patterns of the CALOES
dataset. Additionally, the subset of mountain SAR incidents from both datasets that were
recorded as an accidental activation of a sat-comm device were reviewed and compared against
actual SAR missions. Supplementing the spatial statistics with the visual analysis and descriptive
statistics lent context to the results so rescue teams and organizations can demonstrate not just
where mountain SAR incidents occur and how those locations might differ with sat-comm
device usage, but they could begin to explore the impact of incident distributions on rescue
operations.
3.5.1 Comparing Spatial Relationships and Attributes
This research applies visual analysis and descriptive statistics to mountain SAR data in
order to evaluate how incident attributes could inform training and preparation plans for SAR
teams as they adapt to the impacts of sat-comm device activations. Both methods are considered
exploratory data analysis techniques as they do not test for significance but instead rely on
interaction between the analyst and the data for interpretation. Visual analysis involves the visual
inspection of representations like maps and GIS layers to draw conclusions (O’Sullivan and
Unwin 2010). The visual analysis techniques used in this research include an inspection of
incident locations, spatial relationships, and attributes by comparing SAR incidents against
satellite imagery in a GIS. This research visually inspected AFRCC PLB activations to identify
any that match a CALOES incident. This research used the Near tool to figure out where the
AFRCC incidents lay in relation to the CALOES incident neighborhoods and to assess whether
including the AFRCC data in future spatial statistical analysis could contribute to new – or
reinforce current – mountain SAR hot spot, cluster, and outlier locations.
73
This research used descriptive statistics to compare incident counts and attributes over
time for both the AFRCC and CALOES datasets. The results are presented in tabular or chart
formats. This research examined incident elevation and temporal attributes at the incident-level.
Considering the spatial and temporal attributes of incidents that derive from a sat-comm device
activation against those originating with other means of notification gave insight into how sat-
comm devices might be changing SAR dynamics.
3.5.2 Assessing Accidental Activations
The activation of a sat-comm device generates a SAR response until the activation can be
confirmed as accidental or the incident is resolved. Accidental activations therefore pose a
potential drain on resources if they occur in mountainous terrain and require the time and
expenses associated with the deployment of specialist rescue teams. This research examined
mountain SAR incidents determined to originate with an accidental device activation using the
visual analysis and descriptive statistics described above, and then reviewed the results for how
they might impact SAR organization planning and expectations.
3.6 Summary
This research leverages the benefits of GIS tools to explore how sat-comm device usage
impacts the SAR environment. This research ran two point pattern analysis methods on actual
mountain SAR incidents from the larger CALOES dataset, the results of which informed
subsequent methods of spatial analysis. Creating an incident neighborhood structure supported
local spatial statistical analysis to identify hot spots, clusters, and outliers. Additionally, incidents
were organized by space as well as time in an STC and assessed for emerging hot spots. This
research then evaluated the hot spots, clusters, outliers, and emerging hot spots attributed to
incidents originating with a sat-comm device for similarities and differences against the sites
74
associated with incidents originating from other means of SAR notification. The smaller AFRCC
dataset was evaluated against the CALOES dataset for redundancies. This research then
measured and visually assessed actual mountain SAR incidents unique to the AFRCC dataset in
the context of the CALOES spatial analysis results to determine possible implications. Lastly,
this research reviewed incident attributes and accidental sat-comm device activations using
descriptive statistics to determine what sat-comm usage might mean for SAR team preparation.

75
Chapter 4 Results
The intent of this study is twofold: to explore the impact of satellite communication (sat-comm)
devices on the search and rescue (SAR) landscape in California’s High Sierras; and to present a
methodology that could guide the efficient use of SAR resources and increase the safety margins
for SAR rescue teams. In order to isolate mountain SAR incidents that met the definition of
mountain SAR used in this study, the study area was restricted to within a five kilometer buffer
of the High Sierra’s wilderness areas, with a total area of 21,749.53 km
2
. After data preparation
and cleaning, 53 mountain SAR incidents represented the Air Force Rescue Coordination Center
(AFRCC) dataset on PLB activations (eight of which were attributed to accidental activations),
running from July 25, 2015, to July 17, 2022. The California Office of Emergency Services
(CALOES) dataset was distilled to a total of 416 mountain SAR incidents (27 of which were due
to the accidental activation of a sat-comm device) dated January 1, 2018, to July 27, 2022. After
removing the accidental activations, roughly one-third of actual SAR incidents in the CALOES
dataset were initiated with a sat-comm device, at 132 incidents. The remaining two-thirds started
the SAR process through a different means of notification, at 257 incidents.
It is clear from a simple visual inspection of mountain SAR incident distributions that
incidents do not have a random distribution across the Sierra Nevadas. Figure 13 presents maps
of actual mountain SAR incidents and accidental activations from the AFRCC and CALOES
datasets from 2018-2022 (AFRCC incidents pre-2018 were omitted from this figure so as to
visually compare incidents across datasets from the same period). In Figure 13a and c, incidents
appear concentrated along the eastern spine of the High Sierras, perhaps due to trail networks,
the scenery, or the challenge of extreme recreation that can attract outdoor enthusiasts. The sat-
comm device activations from both the AFRCC and CALOES incident datasets suggest positive
76
spatial clustering as well as a higher proportion of spatial outliers (see Figure 13b), while
mountain SAR incidents that begin with other means of notification have relatively fewer spatial
outliers (see Figure 13a). Furthermore, sat-comm device activations appear distributed across a
greater swath of the study area than incidents initiated by other means of notification, suggesting
sat-comm devices could impact recreational behavior and rescue team requirements. These
inferences, which were gathered from a simple assessment of points on maps, are supported by
the results of spatial statistical analysis of the CALOES dataset. This research additionally
explored how the spatial and temporal patterns could impact rescue team preparation through
visual analysis and descriptive statistics of the AFRCC and CALOES data attributes and an
evaluation of accidental sat-comm device activations.

Figure 13. Distribution of SAR incidents by means of origination: (a) other means of
notification, (b) intentional sat-comm activation, and (c) accidental sat-comm activation
4.1 Spatial Analysis of the CALOES Dataset
During data preparation, this research separated the CALOES mountain SAR incidents
into three layers for spatial pattern analysis: all actual mountain SAR incidents; actual mountain
77
SAR incidents that originated with the activation of a sat-comm device; and actual mountain
SAR incidents that are triggered by an ‘other means of notification’ (e.g., in-person notification,
cell phone, etc.). This research did not consider accidental sat-comm device activations for
statistical spatial analysis, as accidental activations are not necessarily driven by the same spatial
relationships as actual SAR incidents and because the number of accidental activations reviewed
in this study were too sparse. This research assessed mountain SAR incidents as both singular
events and as aggregations within a neighborhood structure. Point pattern analysis consisted of
average nearest neighbor calculations, a review of distances between neighboring points, and an
exploration of kernel density distances. The resulting distances were judged by expert opinion to
select a distance parameter that was small enough to reflect how SAR incidents interact over
space but large enough to support reasonable computational processing times. This research then
aggregated the mountain SAR incidents by a hexagonal grid whose cell-width matched the
selected distance parameter and evaluated the aggregations to develop a neighborhood structure.
The incidents within the neighborhood structure were assessed to determine how sat-comm
devices impact the locations of incident hot spots, clusters, and outliers. Due to the sparse
number of recorded PLB incidents from the AFRCC dataset within the study area, this study
only ran spatial statistics on the CALOES dataset.
4.1.1 Point Pattern Analysis
The point pattern analysis results provided an initial insight into the spatial patterns of
mountain SAR incidents. The nearest neighbor and kernel density estimates (KDE) point pattern
analysis methods are a simple type of spatial analysis that do not consider topographic variability
like sharp elevation changes or obstructions. While almost one-half of actual mountain SAR
incidents occurred on the same date as another incident, only about one-eighth of these occurred
78
within 250 m of another incident. These figures suggest it is unlikely a SAR incident impacted
another occurring in the same location.
This research ran average nearest neighbor calculations on all three CALOES incident
layers. The layer representing the totality of incidents regardless of notification method had a
minimum bounding rectangle with an area of 30,669.64 km
2
. To make the nearest neighbor ratio
comparable across the notification methods, this area was set for the bounding rectangle of the
other two layers. The average nearest neighbor results, presented in
Table 2, indicated significant spatial clustering for all three layers. The average nearest neighbor
observed mean distance for the mountain SAR incidents originating from a sat-comm device
activation is larger than that for the other means of notification incidents, suggesting the presence
of outliers. Because the sat-comm layer also has a larger expected mean distance since it
contained fewer incidents for the same bounding area, the nearest neighbor ratios for all three
layers are relatively similar. The ratios range from 0.42 to 0.58, falling roughly equally between
complete spatial randomness (i.e., a ratio of one) and complete spatial clustering (i.e., a ratio of
zero). The layer of sat-comm device activations had the larger ratio of 0.58, while the layer of
other means of notification had the smaller ratio of 0.42, suggesting the former has less
clustering than the latter. The sat-comm device layer also resulted in a z-score closer to zero than
the other two layers, suggesting less variation from the mean and hence less extreme clustering
distances than the other two layers. The extremely low p-values indicate more significant
clustering than could be expected due to random chance.
79
Table 2. Average nearest neighbor results
Nearest
Neighbor
Ratio
Observed
Mean
Distance (m)
Expected
Mean
Distance (m)
z-score p-value
All CALOES
Actual SAR
Incidents
0.50 2,227.63 4,439.66 -18.80 .000000
Sat-Comm
Device
Activations
0.58 4,446.48 7,621.45 -9.16 .000000
Other Means
of
Notification
0.42 2,277.72 5,462.08 -17.88 .000000

This research only calculated the minimum, average, and maximum distances between
incidents for the layer representing all actual mountain SAR incidents so as to have the
maximum number of inputs. The measurements between different numbers of neighbors are
presented in Table 3. At 10 neighbors, the minimum distance began to exceed the 500 m used in
subsequent analysis to represent a single location, so distance values beyond 10 neighbors are
not included for inspection in the table. The average distance for all actual mountain SAR
incidents to have at least one neighbor is 2,227.6 m, with the minimum distance between
incidents at less than one meter and the maximum distance at almost 26 kilometers. At least one
location within the dataset has 10 neighbors within a 600 m radius. However, the minimum
distance from eight to nine nearest neighbors jumps about 300 m from just over 100 m to just
over 400 m, suggesting most incidents that could be considered to share a location occur within
roughly 100 m of each other. Furthermore, while some sites may have multiple incidents near
each other, the average number of incidents do not, with the average number of incidents having
only three neighbors within a five kilometer radius. Table 3 suggests there are limited, if any,
interactions between incidents that impact subsequent incidents as well as the presence of several
80
spatial outliers, as the average distances between incidents are relatively large. Instead, incidents
occurring in proximity are likely location-specific rather than influenced by an incident nearby,
with some sites containing more hazards or experiencing higher traffic than others.
Table 3. Distance relationships between actual SAR incidents in the CALOES dataset
Distance
(meters)
Number of Neighbors
1 2 3 4 5 6 7 8 9 10
Minimum 0.088 38.88 47.67 60.94 67.79 94.66 95.35 106.38 403.90 573.74
Average 2227.63 3909.32 4727.88 5645.68 6354.99 7064.39 7696.88 8405.80 8951.71 9547.76
Maximum 25,926.46 37,246.16 37,556.37 45,874.96 48,106.98 67,244.65 68,099.67 79,925.45 84,268.22 89,541.62

This research next assessed actual mountain SAR incidents using KDE to determine the
kernel size where incident densities best matched the study area’s topography, which contains
mountains, alpine lakes, meadows, and glacial valleys. This research evaluated the KDE values
at 250 m, 500 m, and 1,000 m cell sizes using search distances of 1,000 m, 2,500 m, 5,000 m,
and 10,000 m. The raster surface output of incident density estimates with the 500 m cell size
provided the best resolution to represent incident density. The 2,500 m search distance offered
the best depictions of incident clusters that followed topographic characteristics based on a visual
assessment. The tool was run again using these settings for the features representing only
intentional sat-comm device activations and incidents originating with other means of
notification and were examined to check if the settings appropriately represented the smaller
datasets. Figure 14 shows how the resulting density surface raster layers compare across the
entire study area. The magnified area is just west of Whitney Portal on the Tulare-Inyo county
border and represents a roughly three kilometer zone around the tallest peak within the
continental United States. The density surface – and the smoothing effect of the kernel’s quartic
function – effectively conveys which areas have a high concentration of incidents and should be
81
the focus of SAR agencies, which areas with a low incident density should be monitored for
future trends or examined further for potentially hazardous conditions, and how the datasets
representing methods of notification compare. It is worth noting in the magnified zone in Figure
14 that the incident clusters appear elongated, possibly pointing to the impact of visitor traffic
along a trail network. The KDE results informed subsequent analyses involving incident
aggregation schemes.

Figure 14. KDE results using a 500 m grid cell and a 2,500 m search radius
82

4.1.2 Incident Aggregation and Neighborhood Structure
This research aggregated mountain SAR incidents to support spatial statistics where the
unit of analysis would represent the mountain SAR incident frequency per hexagon. After testing
sizes for 250 m and 1,000 m, a hexagon with a width of 500 m was selected for a repeating grid.
Five hundred meters effectively captured mountain SAR incidents within a single topographic
location but was large enough to support reasonable computational processing times for the area
of analysis. The grid of hexagons supported a neighborhood structure with near equal access to
all neighbors due to hexagonal geometry. From a rescue team’s perspective, 500 m is a fair
distance to measure a single SAR site: should a helicopter be able to land or hover only in the
middle of a hexagon, having rescue personnel move a person in a litter 250 m in any direction
could be considered a maximum – albeit arduous – distance for most rescue teams of two or
more people before driving a new landing or hover location. Figure 15 provides a snapshot of the
relationship between the hexagons and mountain SAR incidents. Note how incidents within
roughly 100 m of each other are generally within the same hexagonal cell, and those farther
away may be in neighboring cells. It is also worth mentioning that these distances are based off a
planar surface, and some incidents might have a greater slant range distance if separated by
extreme changes in elevation. While a grid overlay is imperfect, since incidents are arbitrarily
separated or grouped, the 500 m hexagonal grid effectively captures the local densities of
incidents per cell to facilitate a comparison with neighboring cells.
83

Figure 15. Mountain SAR incident distribution within the 500 m hexagonal grid; location is west
of Whitney Portal on the border of Tulare and Inyo counties

This research aggregated each of the three CALOES layers of actual mountain SAR
incidents by the hexagonal areal unit and tested each aggregated layer using the Global Moran’s
I statistic to explore which distances demonstrated statistically significant spatial autocorrelation.
This research used the results to develop a neighborhood distance parameter for subsequent
analysis. Ten distance bands were evaluated at 500 m, 1,000 m, and 2,000 m increments
resulting in an exploration of spatial autocorrelation out to 18.5 km. All three layers had a
significant z-score peak at 3,500 m, at which distance the mean hexagonal cell incident
frequency deviated the most from the mean for the study area and suggests positive spatial
autocorrelation. Looking at Figure 16, which depicts the results for increments of 1,000 m, other
z-score peaks are also present in all three layers. This research disregarded the two peaks at
7,500 meters, as that would constitute a neighborhood structure that would be less useful for
rescue agencies aiming to assess site-specific hazards and was not supported by the KDE results,
which suggested 2,500 m was optimal for visualizing incident clusters.
84

Figure 16. Global Moran’s I distances for maximum spatial autocorrelation

In Figure 16, a notable peak is present at 1,500 m in the layer representing sat-comm
device activations. At 1,500 m, the z-score is 12.98, compared to a z-score of 13.19 at 3,500 m.
However, when running the global spatial statistic at increments of 500 m, which produces z-
scores out to 5 km, only one z-score peak was present at 1,000 m with a z-score of 13.67. This
research disregarded these z-score peaks at 1,000 m and 1,500 m for use as neighborhood sizes
in subsequent analysis so as to facilitate comparison across all three CALOES layers. That said,
a smaller spatial neighborhood might be a more accurate representation for the sparser sat-comm
device activations layer, as larger neighborhood sizes might exaggerate the influence of a single
sat-comm device activation on a location compared to the global mean.
Based on the Global Moran’s I output and KDE results, this research set a neighborhood
distance of 3,500 m for further spatial and spatiotemporal analysis. This distance makes sense
from a topographic and SAR perspective: distances greater than 3.5 km might fail to capture the
nuances of alpine meadows, cliff faces, and lakes and suggest false spatial associations; and half
85
this distance is 1,750 m, which is roughly the distance a helicopter requires to set a safe,
established approach to a landing or hover location, and incidents occurring within the final
approach path typically pose similar challenges (e.g. vertical obstructions and vegetation) to a
helicopter rescue team. The 3,500 m neighborhood fixed distance band allowed incident
frequencies to be compared within a spatial neighborhood where they demonstrated significant
spatial autocorrelation across the study area.
4.1.3 Hot Spot Analysis
With the actual mountain SAR incidents aggregated and a distance selected for defining
the incident neighborhood, this research applied local spatial statistics to aid in the identification
of incident hot spots, clusters, and outliers. Hot spots were identified using the Getis-Ord Gi*
statistic. A hot spot is a neighborhood where the SAR incident frequency is significantly higher
than that of the study area. However, hot spots do not necessarily equate to locations containing
the highest number of incidents, but instead reveal the locations that significantly differ from the
global mean. Incident outliers may appear as a hot spot location if a dataset is so sparse that even
one incident can raise the neighborhood mean such that it is significantly greater than the global
mean, as is the case with several locations originating with a sat-comm device activation. The
global mean for all three CALOES layers has a low value, as over 99% of hexagons in the study
area in all three layers contain no incidents.
The Hot Spot Analysis tool’s results show hot spots of incidents that stem from a sat-
comm device activation are in different locations than incidents that begin with other means of
SAR notification, though there is a substantial area of overlap between the two layers. The
portion of the study area considered a hot spot with a p-value of .01 or less stands at 5.10% for
all actual mountain SAR incidents, 4.62% for incidents originating with a sat-comm device, and
86
5.21% for incidents originating with other means of notification. A comparison of the hot spots
between the different methods of SAR response activation is depicted in Figure 17. Most of the
sat-comm device hot spots and other means of notification hot spots exhibit unique distributions,
suggesting intentional sat-comm device activations increase the demand for SAR services across
a greater portion of the study area. However, as can be seen by the yellow areas in Figure 17,
13.21% of the hot spot locations with 99% confidence levels from the different methods of SAR
notification overlap. This overlap between the different types of SAR notification is largely due
to sat-comm device usage appearing in the same locations as other means of notification, rather
than occurring nearby and expanding the size of significant neighborhoods. Thus, in addition to
widening the distribution of SAR incident hot spots across the study area, these results suggest
sat-comm device usage contributes to several hot spots alongside other means of notification
where cell service or word-of-mouth relay are also possible.
87

Figure 17. Mountain SAR incident hot spot distribution and overlap

4.1.4 Cluster and Outlier Analysis
While the Hot Spot Analysis tool, and the Getis-Ord Gi* static underpinning it, is useful
for drawing attention to areas with multiple SAR incidents, the tool’s results lack the level of
detail to home-in on specific locations (i.e., the level of the 500 m hexagon) and to identify
outliers. The Cluster and Outlier Analysis tool based off the Anselin Local Moran’s I statistic
88
categorizes locations as clusters and outliers at a finer resolution than the Gi* statistic, since each
areal unit is assigned to a High-High cluster, Low-Low cluster, High-Low outlier, Low-High
outlier or no significance category. Category labels describe the areal unit and neighborhood
incident values respectively. The categorization makes Anselin Local Moran’s I results easier to
assess for site-specific spatial relationships that could set environmental expectations for rescue
teams. The four categories of cluster and outlier relationships also helps SAR agencies anticipate
the demand for SAR in a given neighborhood.
The graphical outputs of the two spatial statistics accessed through two GIS tools reveal
their advantages and disadvantages. Figure 18 compares the spatial distribution of results based
on the two types of local spatial statistics compared to point incidents. The distribution of
significant neighborhoods appears similar in Figure 18b and c, since the 3,500 m neighborhood
fixed distance band produces a similar spread of significance values. However, the Anselin Local
Moran’s I results (Figure 18c) are more nuanced and provide greater context to the
neighborhoods. The Hot Spot Analysis tool based on the Getis-Ord Gi* statistic produces layers
of hot spots that are easy to identify but appear equally relevant to SAR agencies, while the hot
spots actually contain a range of incident counts, from two to upwards of seven. The Cluster and
Outlier Analysis tool classified most of the hexagons within a neighborhood as Low-High
outliers based on their Anselin Local Moran’s index values, meaning they have lower incident
frequency values than the mean but are in neighborhoods with a higher incident frequency than
the mean. Because of the relative sparseness of the sat-comm device activations, even one
incident raises the mean value for a neighborhood. The values at the center of the neighborhoods,
which are generally marked as a High-High cluster or High-Low outlier site, allow SAR
agencies to differentiate between areas that could be of greater concern than others.
89

Figure 18. Results of local spatial statistics, including: (a) intentional sat-comm device
activations as unique incidents, (b) hot spots identified using the Getis-Ord Gi* statistic, and (c)
clusters and outliers defined using the Anselin Local Moran’s I statistic

It is worth noting in Figure 18c that several hexagons in the center of a significant
neighborhood are not considered significant, even though those are the only hexagons in the
neighborhood containing incidents (reference Figure 18a). When values are too close to the
mean, they may be considered insignificant using the Anselin Local Moran’s I statistic, an issue
Potter et al (2016) noted in their study on ecological phenomena when they elected to only run
the Getis-Ord Gi* local spatial statistic. For this reason, SAR agencies who wish to examine
cluster and outlier locations should consider the results carefully if they wish to weigh areas
more seriously depending on if clusters or outliers are present, since they might misdiagnose
SAR neighborhoods of interest to them.
90
A visual comparison of High-High cluster locations and High-Low outlier locations
between the different types of notification method reveals the contrasting spatial distributions of
the cluster and outlier categories. Figure 19a and b present the results for a visual comparison,
with the map on the left showing the distribution of High-High clusters and High-Low outliers of
sat-comm device activations, and the map on the right depicting the distribution based on other
means of notification. No Low-Low clusters were present in any of the results. Since all Low-
High outliers had an incident count of zero and served mostly to highlight neighborhoods
containing significant values, they were excluded from a comparison of results. Looking at
Figure 19a, it is apparent the layer of sat-comm device activations contains mostly High-Low
outliers, at about 70% of all cluster and High-Low outlier sites. These results likely reflect the
adoption of sat-comm devices as they increase in popularity compared to more established
methods of communication, which could influence recreational behavior and decision making in
remote areas lacking cell service or access to rescue service hubs. By contrast, Figure 19b has the
opposite dynamic, with far more High-High clusters in the layer of other means of notification
due to more incidents appearing within proximity to prior incidents and only about 20% of
clusters and High-Low outliers designated as outliers. While the prior visual inspection of
incidents and nearest neighbor distance measurements both suggested spatial outliers might play
a larger role in the layer of sat-comm device activations than other means of notification, the
Cluster and Outlier Analysis results provide measurable statistical significance of outlier versus
cluster predominance, demonstrating the challenges to anticipating where future intentional sat-
comm device activations may occur.
91

Figure 19. Anselin Local Moran’s I results for (a) sat-comm device activations and (b) other
means of notification

4.2 Spatiotemporal Analysis of the CALOES Dataset
In order to conduct statistical spatiotemporal analysis and detect emerging hot spots, this
research aggregated CALOES mountain SAR incidents by time and space in a space-time cube
92
(STC) structure. Incidents were organized spatially by the same 500 m wide hexagonal grid as
used with the local spatial statistics, but they were also binned by one-month intervals. In Arc-
GIS Pro, the Mann-Kendall trend test is run concurrently with STC creation (Esri n.d.); however,
because the data are seasonal and the Mann-Kendall test does not consider periodicity, this
research disregarded the STC trend results based off the one-month intervals produced by the
Create Space Time Cube By Aggregating Points tool. The resulting STC had a total of 593,125
hexagon grid locations and 55 time-step intervals that could be grouped into spatial and temporal
neighborhoods for analysis.
Prior to conducting spatiotemporal analysis on the STC, this research visualized the data
over time without consideration given to spatial relationships. The results are presented as a line
graph in Figure 20, where the number of sat-comm device activations are compared with other
means of notification by month. The seasonal nature of the data is immediately apparent, with
the summer months seeing far more SAR activity than winter months. The graph also indicates
sat-comm device usage is supplementing and replacing other methods of communication since
2020, though there appears to be minimal sat-comm device usage recorded for 2022 at the time
the CALOES dataset was requested. The results from the local spatial statistical analysis suggest
the shift in method of SAR notification towards sat-comm devices is spatial as well as temporal,
since sat-comm device activations occur in the same areas where other methods of notification
are available. Conducting spatiotemporal analysis on incidents bounded within an STC allows
SAR agencies to discern how these shifts occur over both time and space and further explore the
role of sat-comm device activations on mountain SAR incident distribution.
93

Figure 20. Incidents by notification method over time, CALOES dataset

Each of the three CALOES layers were the input for one STC, and this research tested all
three STCs for emerging hot spot trends based on 3,500 m spatial neighborhoods and twelve-
month temporal neighborhoods (to account for seasonal bias). The Getis-Ord Gi* statistic
categorized locations within the spatiotemporal neighborhood as hot spots if they had high
incident frequencies in a high incident-frequency neighborhood that significantly differed from
what could be expected due to random chance. The Mann-Kendall test detected trends by
comparing the Gi* z-score of a bin against previous bins to identify a constant increase or
decrease in values over time. While more time intervals can provide trends with higher fidelity, a
minimum of four intervals are required to detect a trend (which the four and a half years of
CALOES incident data provided). Running the Emerging Hot Spot Analysis tool results in a
two-dimensional layer where each areal unit is assigned a trend type.
Eight emerging hot spot trends are possible: new, consecutive, intensifying, persistent,
diminishing, sporadic, oscillating, and historical (Esri n.d.). Two trends did not appear in any of
the results: oscillating hot spots, which require a location to be considered a cold spot as well as
a hot spot over time; and historical hot spots, where a location has been a hot spot for every time
interval except the most recent one. Of the trend types, two mark emerging hot spots as relatively
recent phenomena: new hot spots, which describe a location that is considered a hot spot for the
94
first time during the most recent time interval (i.e., year), and consistent hot spots, which
represent locations that have been a hot spot only for the last two years. The intensifying trend
type is perhaps the most concerning to SAR agencies looking to put mitigation measures in
place, as an intensifying hot spot marks where a location’s neighborhood has been a hot spot for
90% of the time intervals and the Gi* z-score values have been increasing over time (i.e., the
neighborhood’s mean incident frequency is increasingly larger than the study area mean). The
other trend types provide useful descriptions of hot spot patterns over time: a persistent hot spot
is like an intensifying hot spot, but the Gi* z-scores are not increasing; a diminishing hot spot is
a persistent hot spot with a significant decrease in Gi* z-score values over time; and sporadic hot
spots represent locations that are a hot spot for the final year as well as during an earlier year.
Visualizing the Emerging Hot Spot Analysis results on a map gives context to the distribution of
hot spot trend types.
Like the purely spatial Hot Spot Analysis results, the significant neighborhoods from sat-
comm device activations identified from the Emerging Hot Spot Analysis tool show some
overlap with significant neighborhoods from other means of notification. The spatial overlap of
several emerging hot spots suggests sat-comm device activations have shared neighborhoods
with other means of notification for multiple years. However, the notification methods exhibit
different spatial patterns of emerging hot spots when viewed in isolation. The emerging hot spots
due to sat-comm device activations are smaller and sparser than the other means of notification
emerging hot spots, reflecting the more recent usage of sat-comm devices and the smaller
number of sat-comm device incidents aggregated per temporal neighborhood in an STC.
Figure 21 depicts the patterns of emerging hot spots for a section of the Sierras known at
the Ritter Range west of Mammoth, CA, along the border of Madera, Fresno, and Inyo counties.
95
In Figure 21a, which presents the emerging hot spots of all actual mountain SAR incidents, both
the emerging hot spot shapes and the trend types reflect the influences of the different methods
of SAR notification. Figure 21b depicts the emerging hot spots resulting from the other means of
notification STC. It shows how the other means of notification incidents exert the strongest
influence on the shapes of the emerging hot spots in Figure 21a, since there are more incidents
originating from other means of notification in the same neighborhoods over time than seen with
sat-comm device activations. Figure 21b also contains all the persistent, intensifying, and
diminishing trend types for the other means of notification emerging hot spots. Figure 21c shows
the emerging hot spots from sat-comm device activations. The emerging hot spots are smaller,
and the dominant trend type is consecutive, reflecting the more recent usage of sat-comm devices
as a method of SAR notification. Sat-comm device activations influence the trend types in Figure
21a, since exploring sat-comm device activations and other means of notification in conjunction
decreases the Gi* z-values for a spatial neighborhood by raising the study area mean for recent
temporal neighborhoods. Comparing maps of emerging hot spots by notification method allows
SAR agencies to consider the extent sat-comm device activations influence the spatial and
temporal patterns of mountain SAR incidents, enabling a deeper examination of the underlying
spatial and temporal relationships.
96

Figure 21. Emerging Hot Spot results on Ritter Range from (a) all actual mountain SAR
incidents, (b) other means of notification, and (c) sat-comm device activations

Intensifying hot spots are concerning to SAR agencies, as those locations not only
suggest the historical presence of SAR incident hot spots, but also that the numbers of incidents
occurring in those neighborhoods are significantly increasing over time compared to the study
area. Figure 22 presents the emerging hot spots identified around Mount Whitney, the tallest
mountain within the continental United States. This area straddles the border between Tulare and
Inyo counties, with ramifications for which county may be tasked with rescue responsibilities.
Figure 22a depicts the entirety of intensifying hot spot locations for all actual mountain SAR
incidents in the study area, representing 3.30% of hot spot trend types. Neither the other means
of notification SAR incidents nor the sat-comm device activations (Figure 22b and c
respectively) could capture this intensification trend when considered in isolation. The
dominance of consecutive hot spots amongst sat-comm device activations added to the sporadic
hot spots of the other means of notification incidents, serving to significantly increase the Gi* z-
97
score values in the Whitney Portal area in more recent years. Creating an STC and running tests
to detect emerging hot spots produces results that allow SAR agencies to identify areas of
perpetual and increasing concern. Results similar to those depicted in Figure 22 can support
decisions on how to manage resources and determine where mitigation measures should perhaps
be concentrated.

Figure 22. Emerging Hot Spot results near Mt. Whitney from (a) all actual mountain SAR
incidents, (b) other means of notification, and (c) sat-comm device activations

Consecutive and sporadic trend types are the dominant categories across all CALOES
mountain SAR incident layers. Table 4 gives a breakdown of the proportion of hot spot trend
types per layer. Consecutive hot spots are the dominant trend type for sat-comm device
activations. Sporadic hot spots are the dominant trend type for other means of notification.
Consecutive and sporadic hot spots account for roughly 90% of mountain SAR incident
locations. However, about 70% of sat-comm device activation hot spots are consecutive while
about 20% are sporadic, while the reverse is true for other means of notification hot spots. Along
98
with the prevalence of consecutive hot spots, sat-comm device locations have the highest
percentage of new hot spots, at 8.45%. These results suggest sat-comm device usage is a recent
phenomenon and that all mountain SAR incidents occur mainly in neighborhoods that have been
a hot spot of at least two years of the four and a half years considered.
Table 4. Distribution of actual mountain SAR incidents per hot spot trend type
Hot Spot Trend Type All Mountain SAR
Incidents
Sat-Comm Device
Activation
Other Means of
Notification
New 2.62% 8.45% 3.93%
Intensifying 3.30% - 0.034%
Diminishing 0.23% - 0.068%
Consecutive 46.09% 70.16% 25.67%
Persistent 0.75% - 0.68%
Sporadic 47.02% 21.39% 69.62%

4.3 Comparing the AFRCC and CALOES Datasets
This research compared the 45 intentional PLB activations from the AFRCC dataset
against the 389 actual mountain SAR incidents from the CALOES dataset to check for
redundancies. Only six of the recorded PLB activations possibly correspond to an incident in the
CALOES dataset, all occurring within one day and one kilometer of each other. Two of the six
were not categorized as either a PLB or SEND activation in the CALOES dataset, suggesting
they were mislabeled in the CALOES dataset or possible represent separate incidents that took
place in questionable spatial and temporal proximity to another one. Neither the time nor
location of the 6 potential overlapping records are exact matches, but they are similar enough to
raise questions. The time of day varies by an average of 7 hours and 50 minutes when comparing
the CALOES local time to the AFRCC local time. However, the AFRCC stores PLB SAR
incidents in Zulu time and date format, and the average difference between the AFRCC Zulu
99
time and the CALOES local time is only 40 minutes, suggesting a possible miscommunication or
misinterpretation of time format between the agencies. The average distance apart between the
possibly redundant incidents is 358.07 m, and the average difference in elevation is 65.07 m.
Since more information would be required to confirm the six PLB activations are duplicates
across the datasets, they were not removed from the AFRCC dataset for subsequent analysis.
The PLB activations from the AFRCC dataset were next measured by their proximity to
CALOES incidents in order to explore how the historical SAR incident datasets relate over
space. This research evaluated distances between incidents based on the CALOES neighborhood
structure used for spatial analysis, since the PLB activations represent contemporary and older
incidents that could impact the results of future research should the two datasets be merged. Of
the PLB activations representing actual SAR incidents, seven fell within 500 m of a CALOES-
recorded sat-comm device activation. A further 27 PLB activations lay within the same
neighborhood as a CALOES sat-comm device activation, of which 10 occurred before January
2018, 11 were in the same neighborhood as a sat-comm device activation cluster, and 16 were in
the same neighborhood as a sat-comm device activation outlier. The average distance between a
PLB activation and a CALOES sat-comm device activation is 3,735.88 m, and the median
distance is 2,653.42 m. Two more PLB activations fell within 500 m of another means of
notification incident from the CALOES dataset, and three PLB activations were within the same
neighborhood as another means of notification incident. The average distance between a PLB
activation and a CALOES other means of notification incident is 10,471.20 m, and the median
distance is 6,465.72 m.
These proximity measurements imply the PLB activations since 2015 occur largely in the
same neighborhoods as other sat-comm device activations recorded in the CALOES dataset,
100
while they have a lesser degree of spatial overlap with incidents originating from other means of
notification. If combined with the CALOES dataset, the PLB additions would likely increase the
number of significant clusters of sat-comm device activations and increase the intensity of hot
spots, while reducing the relative number of sat-comm device outliers. Additionally, because the
AFRCC dataset includes data from three years prior to the CALOES dataset, comparing the
datasets suggest sat-comm devices have potentially been used to initiate a SAR response in
locations different than other methods of SAR notification for a longer period than the CALOES
dataset can capture.
4.3.1 Attribute Comparison
The temporal and spatial attributes of the AFRCC and CALOES actual mountain SAR
incidents were explored in order to provide context to SAR agencies and rescue teams. The
temporal attributes examined were the month, the day of the week, and the time of day the SAR
process started. For both datasets and all methods of SAR incident notification, the majority of
SAR incidents occur from late spring through early autumn, starting in May and running through
September. Figure 23 visualizes the results as data clocks. The data clock format reveals that the
busiest month for SAR rescue teams varies by year, though is typically July or August. October
through the end of January are historically the quietest months, with fewer than three actual
mountain SAR incidents occurring per month during any year in either dataset. In the CALOES
dataset, the proportion of incidents that occurred during the peak season and originated with a
sat-comm device is 90.15%, while the proportion of peak-season incidents from other means of
notification is 78.99%. Similar to the sat-comm device ratio in the CALOES dataset, the
proportion of incidents that occur during the peak season from the AFRCC dataset stands at
91.11%. The higher percentage for sat-comm devices during the summer vice winter months
101
could be attributed to a change in activity during the winter months, where the user might not
expect to be either outside of cellular range or isolated from other people. Users also might not
associate winter activities as risky and are instead caught off guard while conducting an activity
they consider routine: a review of the available comments from CALOES incidents in the winter
months suggest most SAR calls are the result of having a vehicle stuck in the snow or getting
snowed in a building.

Figure 23. Data clocks of the AFRCC (left) and CALOES (right) datasets, 2018-2022

While the data clocks in Figure 23 provide a useful visual for capturing peak-season
trends over time, an aggregation of mountain SAR incident solely by month and not broken up
by year emphasizes the size of the demand during the summer season. Figure 24 presents the
CALOES mountain SAR incidents by month and notification method as a bar graph. It must be
noted in Figure 24 the ratios of peak months to off-season months do not include the second half
of 2022 and therefore cannot capture the full seasonal impact of the 2022 case load. Even with
those missing incidents, the spike in demand for SAR missions via the different notification
methods is striking.

102

Figure 24. Incidents by month and method of notification, CALOES dataset

Like the monthly data, there is not an even spread of actual mountain SAR incidents
throughout the course of the week. Most SAR incidents occur during the long-weekend days –
i.e., Friday through Monday – while fewer tend to occur Tuesday through Thursday. The
AFRCC incidents by day of the week are depicted as a graph in Figure 25. While almost 70% of
incidents occur during long weekends, Wednesdays historically have the second highest number
of incidents, at 15.56%, after Sunday, at 31.11%. Since the total number of PLB activations in
the AFRCC dataset is relatively low, a few incidents can bias the conclusions drawn from a bar
graph and more data would be required to accurately interpret incident spread by day of the
week.

Figure 25. Incidents originating with a PLB activation per day of the week, AFRCC dataset

The CALOES incidents by day of the week and method of notification are also depicted
as a bar graph in Figure 26. The percentage of incidents that occur during a long weekend for
103
sat-comm device activations and other means of notification are 71.21% and 65.76%
respectively. Saturdays and Sundays historically have the highest number of incidents that
originate with another means of notification, while Sundays and Mondays have the highest
number of incidents that begin with a sat-comm device activation. The relatively high percentage
for all datasets and methods of notification favoring the long weekend days could likely be
attributed to increased visitor traffic during those days.

Figure 26. Incidents by day of the week and method of notification, CALOES dataset

This research reviewed actual mountain SAR incidents for when the notification occurred
to consider implications for rescue team proficiency in day or night operations. Historically, the
majority of PLB activations supporting an actual mountain SAR event occurred during the day,
at 75.56%. Only one incident occurred within an hour to sunset, and the rest – at 22.22% –
occurred at night. In contrast, most of the incidents recorded in the CALOES dataset took place
at night. The predominance of night-time missions in the CALOES dataset applies to both sat-
comm device activations and other means of notification, as can be seen from the bar graph in
Figure 27. After including the mountain SAR incidents where the earliest recorded time of
notification is within an hour of sunset, a total of 62.88% of rescues stemming from a sat-comm
device activation and 71.60% of rescues originating from other means of notification likely
required a rescue team to effect the rescue during hours of darkness. A minority of SAR
incidents began during the hour prior to sunrise, possibly due to a lack of early morning
104
recreational activity, or perhaps because of the renewed hope that inevitably rises with the dawn
mitigates the desire by a subject to request a SAR response.

Figure 27. Incidents by time of day and method of notification, CALOES dataset

This research next assessed actual mountain SAR incidents by their elevation.
Proportionally, incidents originating with a sat-comm device activation occur at higher
elevations than other means of notification. The elevations of the intentional PLB activations
from the AFRCC dataset are presented in Figure 28. Incidents were binned by 500 m intervals
for ease of visualization. Mountain SAR incidents which occur above 6,000 ft (just under 2,000
m), can impact the ability of different helicopter models to perform a rescue, particularly in the
summer months when 6,000 ft can “feel” like 10,000 ft to a helicopter’s engines (see Fisher 2021
for a greater discussion on helicopter operations in California’s mountains south of the High
Sierras). In the AFRCC dataset, 88.89% of PLB activations supporting an actual mountain SAR
incident occur above 2,000 m. Over half occur above 3,000 m (just under 10,000 ft). These
results suggest the demand for SAR services increases with elevation.
105

Figure 28. Incident elevations by PLB activation; AFRCC dataset

The elevations of the actual SAR incidents from the CALOES dataset are presented in
Figure 29. The results reflect similar findings by Heggie and Amundson (2009), who determined
in their study of US national parks that about a quarter of all SAR incidents in national parks
were in mountainous terrain between 1,524-4,572 m. The relatively high elevation of all
mountain SAR incidents reflects the underlying mountainous terrain, although the substantial
number of incidents that occur at extreme elevations suggest SAR incidents are less likely in
low-lying areas like foothills and the bottoms of canyons. Of the sat-comm device incidents,
89.39% occur above 2,000 m, while 91.05% of the incidents stemming from other means of
notification occur above 2,000 m. After 2,000 m, however, incidents originating with other
means of notification start to decrease, while sat-comm device usage starts to increase. This
juxtaposition of increasing and decreasing notification methods over elevation could be due to
the increased reliance by users on sat-comm devices to improve the perception of safety, even
when experience levels do not match. This pattern could also be explained by experienced users
carrying sat-comm devices (reference Boore and Bock 2013) and tackling challenges at higher
elevation, at which point an accident occurs. Without more information on user behavior, rescue
106
teams can only conclude from this research that sat-comm devices are playing an increasing role
at higher elevations but cannot explain why.

Figure 29. Incident elevations by method of notification, CALOES dataset

4.3.2 Accidental Activations
The accidental activation of a sat-com device could deplete mountain SAR resources and
contribute to asset fatigue if the activation requires a rescue team to respond. Should the sat-
comm device not have two-way messaging capabilities, or if the device’s owner fails to catch the
accidental activation and not actively monitor their device, then rescue teams will treat the
activation as an actual incident until proven otherwise. There were a total of eight accidental
activations of a PLB between January 2015 through July 2022 recorded in the AFRCC dataset
for the study area. Five of these accidental activations occurred during the same time period as
the CALOES dataset. The CALOES dataset contained a total of 27 accidental activations of a
sat-comm device between January 2018 through July of 2022 in the study area. A comparison of
the two datasets suggests only one accidental activation from each dataset could reference the
same incident; however, as seen during the comparison of intentional activations described
107
above, neither the earliest recorded time of notification nor geographic location were the same,
leaving the dubious possibility that two separate incidents occurred within 400 m, one foot of
elevation, and 14 hours of each other.
The temporal attributes of the accidental activations suggest the accidental activation of a
sat-comm device could occur during any day of the week under day or night lighting conditions,
and they occur most often during the summer months. Just under 90% of the accidental
activations recorded in the CALOES dataset and 50% of those in the AFRCC dataset occurred
during June through August, while 37.5% of the accidental activations in the AFRCC dataset
occurred in the autumn months. Unlike actual mountain SAR incidents, which see fewer mid-
week incidents, the CALOES dataset suggests accidental activations of sat-comm devices do not
favor long weekends. Figure 30 presents a bar graph of incidents by day of the week. The results
suggest a lower number of accidental activations historically occur on Saturday and Sunday. This
is surprising, since outdoor recreation is typically higher during the weekends, and the increased
number of visitors to wilderness areas during those dates would presumably increase the
probability of an accidental activation. In contrast, all but one of the accidental PLB activations
recorded in the AFRCC dataset occurred Friday through Sunday at two to three per day, with one
incident occurring on a Tuesday. In the CALOES dataset, 59.26% of the accidental activations
placed the earliest time of notification either within an hour of sunset or after sunset, while
62.5% of the accidental activations in the AFRCC dataset began during the day. These results
suggest the accidental activation of sat-comm devices may place a burden on SAR resources, as
the occur during the busy summer season, on all days of the week, day or night.
108

Figure 30. Accidental activations of a sat-comm device by day of the week, CALOES dataset

Like actual mountain SAR incidents originating from a sat-comm device, accidental sat-
comm device activations occur in relatively inaccessible locations for rescue teams. Regarding
elevation, all of the accidental PLB activations were above 2,000 m, while 85.18% of the
accidental sat-comm device activations in the CALOES dataset occurred about 2,000 m. Using
the distance between an incident and the area of analysis boundary as an accessibility metric,
accidental activations and actual mountain SAR incidents can be compared to explore the
challenges they may pose to rescue teams due a lack of infrastructure and challenging
topography within wilderness areas. The results of the minimum, mean, and maximum distances
of incidents to the study area boundary are presented in Table 5. On average, the accidental
activations of sat-comm devices occur as far from the built environment as intentional distress
calls. This result likely corresponds to the higher proportion of outliers found during the spatial
statistical analysis of sat-comm device activations, indicating sat-comm device owners are
venturing into remote areas removed from traditional SAR hot spot locations where they then
intentionally call for rescue or accidentally activate their sat-comm device. SAR agencies should
therefore expect to budget for the extra costs associated with accidental sat-comm device
activations that may require extensive time or expense to verify a false alarm.
109
Table 5. Distances of sat-comm device activations from the area of analysis boundary
Distance
(meters)
Sat-Comm Activations
Accidental
(AFRCC)
Intentional
(AFRCC)
Accidental
(CALOES)
Intentional
(CALOES)
Minimum 39.45 1,712.65 3,542.76 252.14
Mean 13,873.53 15,179.06 11,835.38 11,177.72
Maximum 26,810.77 30,135.48 27,627.11 33,114.79

110
Chapter 5 Discussion and Conclusions
This research conducted an exploratory spatial and temporal analysis of historical mountain
search and rescue (SAR) incidents in order to examine how satellite communication (sat-comm)
devices might impact traditional mountain SAR patterns. This study’s methodology leveraged
the capabilities of a geographic information system (GIS) to identify statistically significant
spatial patterns over time. While aspatial statistical programs support most of the research on
mountain SAR incidents to date (Boore and Bock 2013; Heggie and Amundson 2009;
Kaufmann, Moser, and Lederer 2006), the tools available in a GIS can describe the distribution
of incidents through simple visualizations as well as more complex statistical metrics, the
products of which can help SAR organizations understand the operating environment and
demands for their rescue teams. The methodology presented in this research could be tailored by
overland SAR organizations to maintain an informed approach to their SAR resource and
training plans. However, the output of this methodology is limited by the quality and quantity of
historical incident records, and the results are only as good as the input used to generate them.
This chapter reviews the study’s results, considers the limitations associated with the
datasets and research methodology, and addresses possible research avenues that future works on
mountain SAR spatial analysis could incorporate. Additionally, a section of this chapter proposes
best practices for collecting and maintaining data by SAR agencies at large, as this research
identified several deficiencies in the quality of mountain SAR datasets that could have lasting
impacts on the SAR community’s ability to analyze historical incidents and prepare for future
SAR cases. The chapter concludes with a reflection on how this study advances the field of
mountain SAR.
111
5.1 Project Findings
This study found sat-comm devices impact the spatial distribution of mountain SAR
incidents in recent years, although sat-comm device activations exhibit similar temporal patterns
to incidents originating with a traditional method of communication. An exploration of
statistically significant hot spot, cluster, and outlier locations suggests sat-comm device
activations exhibit a broader spatial distribution than SAR incidents stemming from other means
of notification. Spatiotemporal analysis results emphasize the recent impact of sat-comm devices
on the mountain SAR incident spatial patterns. An analysis of incident attributes reveals similar
temporal patterns between sat-comm devices and other means of notification by week and
month, although there is a clear increase in the number of mountain SAR incidents beginning
with a sat-comm device over time. The attributes of accidental sat-comm device activations do
not share these temporal trends, although they do suggest spatial similarity to intentional sat-
comm device activations. Despite the benefits of improved positional accuracy that sat-comm
devices provide to rescue teams, the introduction of portable sat-comm devices to outdoor
recreation ultimately increase the burden placed on rescue teams by presenting a wider spatial
distribution and the possibility of accidental activations unique to modern sat-comm devices.
5.1.1 Spatial Analysis Results
The results from both the Hot Spot Analysis tool, based on the Getis-Ord Gi* statistic,
and the Cluster and Outlier Analysis tool, based on the Anselin Local Moran’s I statistic, reveal
sat-comm device activations contribute to higher incident frequencies in the same neighborhoods
as other means of notification, as well as in new, more isolated locations spread out across the
Sierra Nevada wilderness areas. Just over an eighth of mountain SAR incident hot spots
attributed to sat-comm devices overlap with hot spots from other means of notification. Since
112
most hot spots do not overlap, these results suggest different types of recreational behavior by
sat-comm device owners. The lack of overlap might also be because there are no other viable
ways to communicate in these locations other than with a sat-comm device; instead, overdue or
missing person procedures would take effect and the mission focus would be a search. SAR
missions classified as a search had been removed during the data preparation stage due to lack of
precise coordinate information, and search areas were not explored in this study to explore
whether the spatial relationship between sat-comm device activations and historical search areas
over time.
The Hot Spot Analysis tool’s results offer a visual product that facilitates rapid
identification of neighborhoods of concern across the study area. Hot spot maps clearly
communicate statistically significant neighborhoods of high mountain SAR incident frequency,
which could be useful for determining resource allocation and the regional distribution of assets.
Hot spots can also provide a means for future researchers to explore cross-jurisdictional
relationships. For example, just under a sixth of the hexagons that contribute to a hot spot
neighborhood are within 500 m of a county line in this study, a finding which highlights the
importance of maintaining cross-jurisdictional datasets and communicating rescue team
availability and capabilities. The Getis-Ord Gi* statistic, however, does not account for spatial
outliers, which can present an increased burden to rescue teams due to their inaccessibility and
the possible lack of familiarity by rescuers with outlier site hazards.
The Cluster and Outlier Analysis results highlighted the impact of spatial outliers on the
sat-comm device activation distribution. A location marked as an outlier has an incident
frequency above the global average but is surrounded by locations with low (i.e., no) incident
counts. About 70% of locations with sat-comm device activations are considered outliers based
113
on the Anselin Local Moran’s I statistic, while just over 80% of locations with incidents from
other means of notification belong to a cluster. These results indicate traditional methods of
notification are more likely to be used repeatedly in a mountain SAR incident neighborhood,
while sat-comm devices initiate the SAR process in new, remote locations, which could develop
into clusters over time. Furthermore, these spatial patterns could suggest sat-comm devices are
supplanting traditional methods of notification in addition to serving as the primary notification
method in isolated areas.
5.1.2 Spatiotemporal Analysis Results
The Emerging Hot Spot Analysis tool based on the Getis-Ord Gi* local spatial statistic
and Mann-Kendall trend test applied to an STC offer a novel approach for examining mountain
SAR incidents. The tool’s results suggest sat-comm device activations are essential to revealing
locations that are exhibiting an intensification of mountain SAR incidents over time.
Furthermore, sat-comm device activations demonstrate emerging hot spot patterns in more recent
years, marked by the relatively higher proportions of new and consecutive trend types. As more
mountain SAR incident records become available over time, SAR organizations should conduct
further spatiotemporal analysis to observe whether sat-comm device incident outliers remain as
outliers or develop into more emerging hot spots.
5.1.3 Attribute Analysis Results
The results of an exploration of mountain SAR incident attributes using descriptive
statistics indicate little variation between the methods of notification other than elevation. The
number of sat-comm device activations in both the AFRCC and CALOES datasets increased
with an increase in elevation, while the number of incidents originating with other means of
notification decreased above 2,000 m. Otherwise, while an increasing number of mountain SAR
114
incidents rely on sat-comm devices to call for help through 2022, the weekly and monthly
distribution of incidents does not show substantial variation between sat-comm device
activations and other means of communication. Most mountain SAR incidents occur Friday
through Monday during the summer months. PLB activations do not appear to favor long
weekends, but the historical records from the AFRCC dataset contain relatively few incidents,
making it difficult to capture temporal variation. Based on these findings, rescue teams should
train for high-elevation rescues and anticipate increased demand Friday through Monday during
the summer months.
5.1.4 Accidental Sat-Comm Device Activations
The results from visual analysis and descriptive statistics of accidental sat-comm device
activations suggest they present a challenge for rescue teams. Accidental activations largely
occur during the summer months, but they appear to occur at random throughout the week during
day and night conditions. Like intentional sat-comm device activations, accidental activations
mainly occur at high elevations above 2,000 m, and they occur at roughly equivalent distances
from the study area boundary. SAR organizations should expect to budget for accidental
activations that will likely occur at random in challenging, isolated locations.
5.2 Limitations
The most critical limitation in this research is data quality. The AFRCC and CALOES
datasets suffer from omissions, redundancies, and mislabeling. The results can therefore only be
considered representative of possible spatial and temporal patterns and cannot be treated as
exact. The AFRCC does not store PLB activation reports in a database that can target attributes
within the reports. This made dataset creation a laborious process of combing through seven
years of national-scale records for location information to determine whether or not a record fell
115
within the study area. Consequently, there is a chance records were missed and not included in
the final product used for evaluation. It was also interesting to discover only a fifth of the
AFRCC PLB activations possibly matched an incident in the CALOES data; ideally, 100%
would match, since tasking flows from a national-level agency like AFRCC to the state and
sheriff jurisdictions, and records would be kept at each agency executing the tasking. Of those
sat-comm device activations that were a possible match, there appeared to be a mismatch in
recorded incident start time, as the CALOES local time closely resembled the AFRCC Zulu time,
which constitutes a seven- to eight-hour difference depending on daylight savings. Such
discrepancies indicate a possible breakdown in communication as one agency hands off SAR
responsibility to another. While not an issue encountered in this research, Durkee and Glynn-
Linaris (2012) similarly emphasized the importance of relaying the basis of coordinate data when
passing location information between agencies, as a mismatch of coordinate types could
significantly alter the presumed location of a SAR incident.
In addition to missing these PLB activations, the CALOES dataset included multiple
redundant entries as well as attribute inaccuracies. Redundant records are likely due to SAR
cases handled by multiple rescue agencies when mutual aid is requested, which get combined
into the single CALOES data repository when counties input their individual data. While the data
preparation phase of this research involved extensive evaluation and cleaning in order to
appropriately categorize incidents and remove duplicates, incidents that did not share identical
times or coordinates would be overlooked for removal. Incidents that appeared near each other in
time and space and shared similar attributes were potentially the same incident, but they had to
be treated as isolated incidents due to insufficient attribute information for verification due to a
lack of standardization when entering comments. Furthermore, during preparation of the
116
CALOES dataset, several recorded incidents were mislabeled as originating from a PLB
activation when the comments stated a SEND had been employed. Additionally, several
incidents had inaccurate coordinates – for example, the comments mentioned the response was to
a hiking trail but the coordinates were in a lake – drawing into question all incidents which did
not include comments that could be used to corroborate attribute descriptions. The data quality in
both datasets would benefit from increased standardization at the organizational level and adding
a quality control component for dataset evaluation and refinement.
Several limitations also exist with the research design and spatial analysis methods. In
order to measure mountain SAR incident patterns over space, incidents required aggregation into
areal units such that the areal units could have an attribute of incident frequency for use as a
metric. Incident aggregation, however, generates several problems. The techniques an analyst
employs to group spatially heterogenous data like SAR incidents into a grid can yield differing
analytical results depending on the grid’s parameters, otherwise known as the modifiable areal
unit problem. For example, a large hexagon might contain more incidents than a small hexagon
in the same location, which in turn might result in different locations being considered as
contributing to a hot spot or being defined as an outlier.
Another problem associated with aggregation is computational processing demands.
Aggregated data, particularly across a region as large as the Sierra Nevada wilderness areas,
drive substantial processing requirements during spatial and temporal statistical analysis. One
technique to mitigate this is to decrease the resolution of the areal unit (i.e., make the areal unit
have a larger surface area), although resolution needs to be balanced with an accurate
representation of the spatial data in a digital space. The 500 m hexagon grid effectively captured
local topographic variation, preserved incident scene specificity, and supported reasonable
117
processing times. However, even the 500 m hexagon at this study’s scale of analysis exceeded
the computational processing capabilities of the ArcGIS Pro 2.9 Local Cluster and Outlier
Analysis tool, which would have lent a temporal consideration when identifying spatial clusters
and outliers using the Anselin Local Moran’s I statistic and would have complemented the
results of the Emerging Hot Spot Analysis tool that uses the Getis-Ord Gi* and Mann-Kendall
trend test combination.
Lastly, researcher bias could pose limitations on the research design and interpretation of
results. The author of this study is an expert in helicopter SAR and military SAR operations, a
perspective which creates a bias towards distances that are compatible with an airborne
perspective. For example, the 500 m hexagon represents a single site of SAR incidents. This size
of hexagon makes sense for a medium-sized helicopter crew that requires space for landing and
has multiple crew members or SAR ground team passengers who could take turns carrying
injured subjects to the helicopter. However, for a SAR hiking party or technical rescue team, an
incident half a click away from another one might not seem like they are in comparable
locations. Parameter selection is often subjective in the spatial sciences and guided by expert
opinion, but these decisions must be acknowledged and tailored to answer the research questions
at hand.
5.3 Recommendations
This study recommends future actions for two groups: spatial analysts and SAR agencies.
While this research presents methods to explore the impact of sat-comm device activations on
mountain SAR incidents, future research should continue to explore ways to advance the spatial
analysis methodology of mountain SAR incidents, as well as consider the role of SAR asset
accessibility and subject behavior. The onus is on SAR agencies to develop and retain quality
118
incident records to support meaningful research, and this study recommends SAR agencies focus
resources to standardize, collect, and share mountain SAR incident records.
5.3.1 Recommendations for Future Research
While this study demonstrates techniques to determine locations worth the attention of
SAR agencies and rescue teams, future studies could expand upon the accessibility of incident
locations or hot spots. Accessibility analysis would require a deeper knowledge of the number
and types of available rescue assets – be they different models of helicopters, off-road vehicles,
or foot traffic – and points of entry and exit for those assets. For example, helicopters would
require considerations for medical handover sites and refuel locations, as well as identifying
potential landing sites or hover-only locations in wilderness areas based on ground cover, slope,
and elevation. SAR ground vehicles, hiking teams, and climbing teams would benefit from
mapped roads and trail networks. Accessibility analysis would allow a comparison of
jurisdictions based on numbers of SAR incidents and available SAR assets in order to inform an
appropriate distribution of resources and take note of any gaps in coverage.
Another way to support SAR policy in addition to the spatial analysis of mountain SAR
incidents would be to determine the causal factors of the observed spatial patterns through
correlation analysis or regression techniques. Indices representing weather conditions,
topographic complexity, and demographics could all be used to determine what might cause
certain types of SAR incidents, and whether sat-comm devices are involved in those types of
SAR incidents. Such analysis would, however, require a breakdown of SAR incidents by activity
(e.g., hiking, technical, swimming, etc.), by severity of outcome (i.e., self-recovered through
death), or another category against which potential causal factors could be assessed. Attribute
gaps and inconsistencies in the CALOES dataset hindered such analysis herein, but correlation
119
analysis could enhance future research if the intent is to support mitigation measures so as to
reduce SAR team operating costs. Results are more meaningful if it can be determined not only
whether the number of incidents are increasing or decreasing in specific areas, but what types of
activities the subjects are engaged in that are driving those patterns since they might require
different specialty teams (e.g., technical rope training or OHV equipment). Similarly, a severity
scale would indicate whether the types of incidents that are occurring require a greater or
reduced rescue team footprint and medical support, which implications for operating costs.
Future behavioral studies of sat-comm device owners would benefit from having
additional information on demographics, as well as perhaps a comparison against comparable
technologies like cellular devices. Behaviors and demographics that correlate with sat-comm
device activations could provide insights to rescue teams responding to mountain SAR incidents
as well as other types of emergency situations, from car accidents to wildfires to flooding. While
private companies that create and support SEND products are not willing to share consumer
information, SAR agencies could consider including demographic and subject behavioral data in
their rescue reports. Non-satellite capable cell phones were once a new communications
technology altering the SAR notification landscape, and making note of whether a SAR case
began via cellular network could make an interesting comparison to the newer sat-comm
capability. Future researchers could also incorporate cellular network coverage areas into their
GIS analysis. Such layers are already available for comparison to SAR incident sites on
SARTopo, an online mapping tool that facilitates merging custom layers with environmental and
topographic layers (CalTopo 2021). Understanding user behavior would be beneficial should
rescue centers become overwhelmed with cases as satellite connectivity grows, driving SAR
agencies to evaluate mitigation measures to control operating costs.
120
Lastly, as SAR agencies collect more SAR incident data over time, future research should
revisit the methods presented in this study once there are enough inputs for statistically
significant trend analysis. Due to the main dataset used in this analysis only spanning four and a
half years, and the supporting dataset only seven years, there were not enough years of data to
compare layers in using the Mann-Kendall trend statistic, which requires a minimum of ten years
to run in ArcGIS Pro 2.9. Since mountain SAR data, like wildfire data, is highly seasonal,
incidents would need to be grouped by year for trend tests unless seasonal variation can be
accounted for. Trend analysis is a useful measure for SAR agencies aiming to anticipate the
trajectory of SAR cases and their attributes, and being able to present trend data could help SAR
agencies make decisions about resource allocation at the county and state level.
5.3.2 Recommendations for SAR Agencies
SAR agencies have a responsibility to improve their operations and mitigate the hazards
faced by rescue teams tasked to respond to a case. This responsibility requires an adequate
understanding of available technologies and the willingness to set the standard for SAR
documentation and record keeping. SAR agencies at all levels of responsibility should maintain
secure digital records in a format that can be integrated with geographic and statistical software.
Rescue teams or emergency response coordinators (e.g., watch-floor personnel and duty officers)
should be provided with clear, standardized guidance on what information to collect during, and
keep after, a SAR case that can maximize post-mission analysis. SAR agencies responsible for
maintaining SAR incident records should incorporate a quality control component, be it a
technical specialist or software service, that can corroborate incident report entries, catch
redundancies, and follow-up with subdivisions for the timely incorporation of records. In this
121
way, mountain SAR agencies can build a foundation for adapting and improving the SAR
process.
5.4 Conclusion
This study gives an overview of the implications of sat-comm device usage on mountain
SAR operations and provides methodology that SAR agencies may adopt to advance their
policies and improve SAR safety margins. Ultimately mountain SAR incidents depend on the
number of personnel partaking in outdoor recreation: without visitors, SAR in wilderness areas
would be a moot point. However, the Sierra Nevadas continue to draw high numbers of visitors
who all hold the potential to require assistance. With the advent of emergency satellite
communications capabilities on cellular phones, there is an increased likelihood people will use
satellites to call for help from any location able to connect to the satellite infrastructure.
Understanding how sat-comm technology is currently affecting SAR operations is therefore
necessary for anticipating future operations and demands on SAR resources. Not every SAR case
can be a guaranteed success. However, research like this that increases the odds to save even one
life makes the time and energy spent on preparation and analysis worthwhile.
In an effort to improve the accessibility of this research, its results, and its
recommendations, a StoryMap version of this thesis may be found at:
https://storymaps.arcgis.com/stories/7889bc805a1a4eeb87e34e5edcd7cab7.

122
References
Aftergood, O. S. R., and M. D. Flannigan. 2022. “Identifying and Analyzing Spatial and
Temporal Patterns of Lightning-Ignited Wildfires in Western Canada from 1981 to
2018.” Canadian journal of forest research 52, no. 11: 1399–1411. doi: 10.1139/cjfr-
2021-0353.
Allen, M. J., T. R. Allen, C. Davis, and G. McLeod. 2021. “Exploring Spatial Patterns of
Virginia Tornadoes Using Kernel Density and Space-Time Cube Analysis (1960-2019).”
ISPRS international journal of geo-information 10, no. 5: 310. doi:
10.3390/ijgi10050310.
Anselin, L.. 1995. “Local Indicators of Spatial Association—LISA.” Geographical analysis 27,
no. 2: 93–115. doi: 10.1111/j.1538-4632.1995.tb00338.x.
Apple. 2022. “Use Emergency SOS via satellite on your iPhone 14.” Apple Support, September
8, 2022. https://support.apple.com/en-us/HT213426.
Azofra, M., C. A. Pérez-Labajos, B. Blanco, J. J. Achútegui. 2007. "Optimum Placement of Sea
Rescue Resources." Safety Science 45, no. 9 (November): 941-951.
https://doi.org/10.1016/j.ssci.2006.09.002.
Betty, E. L., B. Bollard, S. Murphy, M. Ogle, H. Hendriks, M. B. Orams, and K. A. Stockin.
2020. “Using Emerging Hot Spot Analysis of Stranding Records to Inform Conservation
Management of a Data-Poor Cetacean Species.” Biodiversity and conservation 29, no. 2:
643–665. doi: 10.1007/s10531-019-01903-8.
Birch, C. P. D., S. P. Oom, and J. A. Beecham. 2007. “Rectangular and Hexagonal Grids Used
for Observation, Experiment and Simulation in Ecology.” Ecological modelling 206, no.
3: 347–359. doi: 10.1016/j.ecolmodel.2007.03.041.
Boore, S. M., and D. Bock. 2013. “Ten Years of Search and Rescue in Yosemite National Park:
Examining the Past for Future Prevention.” Wilderness & environmental medicine 24, no.
1: 2–7. doi:10.1016/j.wem.2012.09.001.
California Public Law. n.d. “California Government Code Sec. 26614.” Accessed November 30,
2022. https://california.public.law/codes/ca_gov't_code_section_26614.
CalTopo. 2021. “Who We Are.” https://sartopo.com/about/who-we-are.
CDFW. 2022. “CDFW Projection and Datum Guidelines.” Updated on March 3, 2022. Available
at https://nrm.dfg.ca.gov.
Clark, P. J., and F. C. Evans. 1954. “Distance to Nearest Neighbor as a Measure of Spatial
Relationships in Populations.” Ecology (Durham) 35, no. 4: 445–453. doi:
10.2307/1931034.
123
dbs Productions. n.d. “International Search and Rescue Incident Database.” Accessed September
15, 2022. https://www.dbs-sar.com/SAR_Research/ISRID.htm.
Doherty, P. J., Y. Liu, J. Doke, Q. Guo, and J. Wieczorek. 2011. “Georeferencing Incidents from
Locality Descriptions and its Applications: a Case Study from Yosemite National Park
Search and Rescue.” Transactions in GIS 15, no. 6: 775– 793.
Durkee, G. and V. Glynn-Linaris. 2012. Using GIS for Wildland Search and Rescue. Esri Press
e-Book.
Esri. n.d. “Elevation Analysis Data Coverage.” Accessed March 30, 2023.
https://developers.arcgis.com/rest/elevation/api-reference/source-data-for-elevation-
analysis-tasks.htm.
Esri. n.d. “GIS Dictionary.” Accessed November 30, 2022. https://support.Esri.com/en/other-
resources/gis-dictionary.
Esri n.d. “How Cluster and Outlier Analysis Works.” Accessed January 20, 2023.
https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/h-how-cluster-
and-outlier-analysis-anselin-local-m.htm.
Esri. n.d. “How Create Space Time Cube Works.” Accessed January 20, 2023.
https://pro.arcgis.com/en/pro-app/2.9/tool-reference/space-time-pattern-
mining/learnmorecreatecube.htm.
Esri. n.d. “How Emerging Hot Spot Analysis Works.” Accessed January 20, 2023.
https://pro.arcgis.com/en/pro-app/2.9/tool-reference/space-time-pattern-
mining/learnmoreemerging.htm.
Esri. n.d. “Space Time Pattern Mining Toolbox History.” Accessed January 20, 2023.
https://pro.arcgis.com/en/pro-app/latest/tool-reference/space-time-pattern-mining/space-
time-pattern-mining-toolbox-history-pro.htm.
Esri. n.d. “Try ArcGIS Survey123.” Accessed January 21, 2023.
https://learn.arcgis.com/en/paths/try-survey123/.
Esri. 2012. “MapSAR Website Provides Online Resources to Search and Rescue Personnel: GIS
Tools and Virtual Communities Now Available in One Location.” PR Newswire. August
8, 2012. https://www.Esri.com/news/releases/12-3qtr/mapsar-website-provides-online-
resources-to-search-and-rescue-personnel.html.
Esri. 2021. ArcGIS Pro: release 2.9. Esri, Redlands, CA.
Ferguson, D.. 2008. “GIS For Wilderness Search and Rescue.” Paper presented at the ESRI
Federal User Conference February 20–22, 2008, Washington, D.C.
https://proceedings.Esri.com/library/userconf/feduc08/papers/gis_for_wilderness_search_
and_rescue.pdf.
124
Fisher, D.. 2021. “GIS Analysis of Helicopter Rescue in San Bernardino County, California.”
Master’s thesis, University of Southern California.
Garmin. n.d. “GPS Accuracy of the inReach SE+/Explorer+ and inReach Mini.” Accessed
September 25, 2022. https://support.garmin.com/en-
US/?faq=a4BVkQAZQ78JZUUvM6puN6.
Garmin. 2022. “Data Insights from 10,000 Garmin inReach® SOS Incidents.” October 24, 2022.
https://www.garmin.com/en-US/blog/saved-by-garmin/data-insights-from-10000-garmin-
inreach-sos-incidents/.
Getis, Arthur, and J. K. Ord. 1992. “The Analysis of Spatial Association by Use of Distance
Statistics.” Geographical analysis 24, no. 3: 189–206. doi: 10.1111/j.1538-
4632.1992.tb00261.x.
Goerlandt, F., E. Venäläinen, and M. Siljander. 2015. "Utilizing Geographic Information
Systems Tools for Risk-Informed Maritime Search and Rescue Performance Evaluation."
Scientific Journals of the Maritime University of Szczecin 43 (115): 55-64.
Guoxiang, L., and L. Maofeng. 2010. “SARGIS: A GIS-Based Decision-Making Support System
for Maritime Search and Rescue.” In 2010 International Conference on E-Business and
E-Government, 1571–1574. doi: 10.1109/ICEE.2010.398.
Hamed, K. H.. 2009. “Exact Distribution of the Mann–Kendall Trend Test Statistic for Persistent
Data.” Journal of hydrology (Amsterdam) 365, no. 1: 86–94. doi:
10.1016/j.jhydrol.2008.11.024.
Heggie, T. W., and M. E. Amundson. 2009. “Dead Men Walking: Search and Rescue in US
National Parks.” Wilderness & environmental medicine 20, no. 3: 244–249. doi:
10.1580/08-WEME-OR-299R.1.
IBM. n.d. “IBM SPSS Statistics.” Accessed January 20, 2022.
https://www.ibm.com/products/spss-statistics.
IERCC. n.d. “Supported Devices and Applications.” Accessed November 30, 2022.
https://www.iercc.com/en-US/supported-devices/.
James, J. W. and A. J. Eardley. 2021. "Sierra Nevada." Encyclopedia Britannica, March 25,
2021. https://www.britannica.com/place/Sierra-Nevada-mountains.
Kaufmann, M., B. Moser, and W. Lederer. 2006. “Changes in Injury Patterns and Severity in a
Helicopter Air-Rescue System Over a 6-Year Period.” Wilderness & environmental
medicine 17, no. 1: 8–14. doi: 10.1580/1080-6032(2006)17[8:CIIPAS]2.0.CO;2.
Koutsias, N., K. D. Kalabokidis, and B. Allgöwer. 2004. “Fire Occurrence Patterns at Landscape
Level: Beyond Positional Accuracy of Ignition Points with Kernel Density Estimation
Methods.” Natural Resource Modeling 17, no. 4: 359-375.
https://doi.org/10.1111/j.1939-7445.2004.tb00141.x.
125
LandSAR. n.d. “PLB vs. SEND.” Accessed November 5, 2022.
https://www.landsar.org.nz/prevent/plb-vs-send/.
Levine, N.. 2004. CrimeStat: A Spatial Statistics Program for the Analysis of Crime Incident
Locations (v 3.0). Ned Levine & Associates, Houston, TX, and the National Institute of
Justice, Washington, DC. May.
Lu, Z. N., A. Briggs, S. Saadat, and I. M. Algaze. 2021. “The Associations Between Visitation,
Social Media Use, and Search and Rescue in United States National Parks.” Wilderness
& environmental medicine 32, no. 4 (December): 463–467. doi:
10.1016/j.wem.2021.08.005.
Malik, A., R. Maciejewski, Y. Jang, S. Oliveros, Y. Yang, B. Maule, M. White, and D. S. Ebert.
2014. “A Visual Analytics Process for Maritime Response, Resource Allocation and Risk
Assessment.” Information visualization 13, no. 2: 93–110. doi:
10.1177/1473871612460991.
Mann, Henry B. 1945. “Nonparametric Tests Against Trend.” Econometrica 13, no. 3: 245–259.
doi: 10.2307/1907187.
Marven, C. A., R. R. Canessa, and P. Keller. 2007. "Exploratory Spatial Data Analysis to
Support Maritime Search and Rescue Planning." In Geomatics Solutions for Disaster
Management, edited by Jonathan Li, Sisi Zlatanova, and Andrea Fabbri. 1st ed. Springer
:Berlin Heidelberg. doi: https://doi-org.libproxy1.usc.edu/10.1007/978-3-540-72108-
6_18.
Microsoft. 2022. “Microsoft 365.” Available at https://www.microsoft365.com/.
MRA. n.d. “The Mountain Rescue Association (MRA).” Accessed January 20, 2023.
https://mra.org/.
Nakaya, T., and K. Yano. 2010. “Visualising Crime Clusters in a Space-Time Cube: An
Exploratory Data-Analysis Approach Using Space-Time Kernel Density Estimation and
Scan Statistics.” Transactions in GIS 14, no. 3: 223–239. doi: 10.1111/j.1467-
9671.2010.01194.x.
NAPSG. n.d. “Wildland Search and Rescue.” Accessed December 15, 2022.
https://sargis.napsgfoundation.org/pages/wisar.
National Search and Rescue Committee. 2011. “Land Search and Rescue Addendum to the
National Search and Rescue Supplement to the International Aeronautical and Maritime
Search and Rescue Manual Version 1.” Accessed via https://www.dco.uscg.mil/.
National Search and Rescue Committee. 2016. “United States National Search and Rescue
Supplement to the International Aeronautical and Maritime Search and Rescue Manual
Version 2.0.” Accessed via https://www.dco.uscg.mil/.
126
National Park Service. n.d. "Traffic in Yosemite National Park." Accessed September 20, 2022.
https://www.nps.gov/yose/planyourvisit/traffic.htm.
National Park Service. n.d. "John Muir and Pacific Crest Trails." Accessed September 20, 2022.
https://www.nps.gov/yose/planyourvisit/jmt.htm.
Nielsen, C. C., C. G. Amrhein, P. S. Shah, K. Aziz, and A. R. Osornio-Vargas. 2019.
“Spatiotemporal Patterns of Small for Gestational Age and Low Birth Weight Births and
Associations with Land Use and Socioeconomic Status.” Environmental health insights
13, no. 1. doi: 10.1177/1178630219869922.
NOAA. n.d. “NOAA Solar Calculator: Find Sunrise, Sunset, Solar Noon and Solar Position for
Any Place on Earth.” Accessed January 2, 2023. https://gml.noaa.gov/grad/solcalc/.
O’Sullivan, D., and D. J. Unwin. 2010. Geographic Information Analysis, 2nd edition. New
York: John Wiley & Sons.
Ord, J. K., and A. Getis. 1995. “Local Spatial Autocorrelation Statistics: Distributional Issues
and an Application.” Geographical analysis 27, no. 4: 286–306. doi: 10.1111/j.1538-
4632.1995.tb00912.x.
Pacific Crest Trail Association. n.d. "PCT Visitor Use Statistics." Accessed September 20, 2022.
https://www.pcta.org/our-work/trail-and-land-management/pct-visitor-use-statistics/.
Pfau, L., and J. I. Blanford. 2018. “Use of Geospatial Data and Technology for Wilderness
Search and Rescue by Nonprofit Organizations.” The Professional geographer 70, no. 3:
434–42. doi:10.1080/00330124.2018.1432367.
Potter, K. M., F. H. Koch, C. M. Oswalt, and B. V. Iannone. 2016. “Data, Data Everywhere:
Detecting Spatial Patterns in Fine-Scale Ecological Information Collected Across a
Continent.” Landscape ecology 31, no. 1: 67–84. doi: 10.1007/s10980-015-0295-0.
Reddy, C. S., N. G. Bird, S. Sreelakshmi, T. M. Manikandan, M. Asra, P. H. Krishna, C. S. Jha,
P. V. N. Rao, and P. G. Diwakar. 2019. “Identification and Characterization of Spatio-
Temporal Hotspots of Forest Fires in South Asia.” Environmental monitoring and
assessment 191, no. 3: 791–791. doi: 10.1007/s10661-019-7695-6.
Siljander, M., E. Venäläinen, F. Goerlandt, and P. Pellikka. 2015. “GIS-Based Cost Distance
Modelling to Support Strategic Maritime Search and Rescue Planning: A Feasibility
Study.” Applied geography (Sevenoaks) 57: 54–70. doi: 10.1016/j.apgeog.2014.12.013.
Sonninen, M. and F. Goerlandt. 2015. "Exploring the Context of Maritime SAR Missions Using
Visual Data Mining Techniques." Scientific Journals of the Maritime University of
Szczecin 43, no. 115: 79-88.
SPOT. n.d. “Spot Satellite GPS Messenger Support.” Accessed September 25, 2022.
https://www.findmespot.com/en-ca/support/spot-satellite-gps-messenger/get-
help/messaging/how-accurate-is-the-spot-satellite-gps-messenger-w.
127
Stoddard, M. A., and R. Pelot. 2020. "Historical Maritime Search and Rescue Incident Data
Analysis." In Governance of Arctic Shipping, edited by Aldo Chircop, Floris Goerlandt,
Claudio Aporta, and Ronald Pelot. 1st ed. Cham: Springer International Publishing.
https://doi.org/10.1007/978-3-030-44975-9_3.
US Air Force. n.d. “US National Personal Locator Beacon Program - (PLBS).” Accessed
September 25, 2022. https://www.1af.acc.af.mil/Units/AFRCC/USNPLBP/.
USCG. n.d. “SAR Program Information.” Accessed September 25, 2022.
https://www.dco.uscg.mil/Our-Organization/Assistant-Commandant-for-Response-
Policy-CG-5R/Office-of-Incident-Management-Preparedness-CG-5RI/US-Coast-Guard-
Office-of-Search-and-Rescue-CG-SAR/CG-SAR-1/SAR-Program-Information/.
USGS. n.d. “3D Elevation Program.” Accessed April 12, 2022. https://www.usgs.gov/3d-
elevation-program.
USGS. n.d. “TNM Download (v2.0).” Accessed October 5, 2022.
https://apps.nationalmap.gov/downloader/.
Visner, M., S. Shirowzhan, and C. Pettit. 2021. “Spatial Analysis, Interactive Visualisation and
Gis-Based Dashboard for Monitoring Spatio-Temporal Changes of Hotspots of Bushfires
over 100 Years in New South Wales, Australia.” Buildings (Basel) 11, no. 2: 1–27.
doi:10.3390/buildings11020037.
Wilderness Connect. n.d. “GIS and GPS Downloadable Data.” Accessed September 14, 2022.
https://wilderness.net/visit-wilderness/gis-gps.php.
Wilderness Connect. n.d. “Why Regulations May Be Necessary in Wilderness.” Accessed
September 14, 2022. https://wilderness.net/learn-about-wilderness/regulations.php.
Wing, M. G., and J. Long. 2015. “A 25-Year History of Spatial and Temporal Trends in Wildfire
Activity in Oregon and Washington, U.S.A.” Modern applied science 9, no. 3: 117–117.
doi: 10.5539/mas.v9n3p117.
Woolford, M. H., L. Bugeja, C. Weller, R. J. Koester, and J. E. Ibrahim. 2022. “Search and
Rescue Incidents Among Persons 65 Years and Older in the USA: Examination of
Factors Associated with Death.” International journal of geriatric psychiatry 37, no. 11.
doi: 10.1002/gps.5821.
Wysokinski, M., R. Marcjan, and J. Dajda. 2014. "Decision Support Software for Search &
Rescue Operations." Procedia Computer Science 35: 776–85. doi:
10.1016/j.procs.2014.08.160.
zoleo. n.d. “zoleo.” Accessed November 29, 2022. https://www.zoleo.com/en-us.

Abstract (if available)

Abstract Mountain search and rescue (SAR) incidents are high risk events that consume time and money, often placing the lives of rescuers and subjects alike in precarious situations. The increasing accessibility of satellite communication (sat-comm) devices for outdoor recreation may change when and where mountain rescue teams are tasked, and California’s SAR agencies need to understand the implications of emerging sat-comm device usage on SAR requirements to mitigate future risks caused by resource and training shortfalls. To date, no academic studies have conducted a holistic assessment of SAR incidents in the Sierra Nevada mountains or considered the impacts of sat-comm device usage on the SAR caseload. Such a knowledge gap impairs the ability of federal, state, and local agencies to anticipate costs and adequately train rescue teams to respond to mountain SAR incidents. This research explores the spatial and temporal patterns of historical mountain SAR incidents in the Sierra Nevada wilderness areas to understand how sat-comm devices impact SAR services in one of the most visited mountain regions in the continental United States. The results of this study suggest sat-comm devices are replacing traditional methods of notification that alert authorities to an emergency. Incidents where the subject communicates using a sat-comm device occur at sites of historical SAR activity where traditional methods of communication are dominant, as well as at new – and more isolated – locations. A lack of confidence in data quality, however, means this study primarily serves to demonstrate spatial and spatiotemporal analysis methods that SAR agencies may adopt to explore historical mountain SAR incidents at a regional scale.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

GIS analysis of helicopter rescue in San Bernardino County, California

PDF

Evaluating the utility of a geographic information systems-based mobility model in search and rescue operations

PDF

Human suffering during wartime: a StoryMap of violations of international law during the Russo-Ukrainian War

PDF

Climate conservation priorities: using MCDA to identify future refuge for the Joshua Tree Forest in California

PDF

A spatial analysis of beef production and its environmental and health impacts in Texas

PDF

Applying least cost path analysis to search and rescue data: a case study in Yosemite National Park

PDF

Assessing the use of normalized difference chlorophyll index to estimate chlorophyll-A concentrations using Landsat 5 TM and Landsat 8 OLI imagery in the Salton Sea, California

PDF

The geographic connotations of reincarceration: a spatial analysis of recidivism in Washington State

PDF

A spatiotemporal analysis of racial disparity in the distribution of superfund sites within Santa Clara County, California

PDF

A comparison of weighted and fuzzy overlays in mapping landslide susceptibility, south-central front range, Colorado

PDF

Implementing spatial thinking with Web GIS in the non-profit sector: a case study of ArcGIS Online in the Pacific Symphony

PDF

Out-of-school suspensions by home neighborhood: a spatial analysis of student suspensions in the San Bernardino City Unified School District

PDF

Impacts of vegetation management on wildfire severity: a study of the 2021 Caldor fire

PDF

A spatiotemporal analysis of environmental risk factors of Lyme disease in the Northeastern United States

PDF

Exploring the pernicious effects of redlining and discriminatory policies on an American city: a spatio-temporal case study of New York City

PDF

An exploratory spatial analysis of fire service and EMS accessibility in northeastern Illinois communities

PDF

Temporal analysis of soil degradation in San Joaquin County, California: a close examination of soil erosion using RUSLE

PDF

Mapping future population impacts caused by sea level rise in Huntington Beach and Newport Beach: comparing the cadastral-based dasymetric system to past dasymetric mapping methods

PDF

Creating a well database and web mapping application: using geographic information systems to manage and monitor groundwater resources in Sonoma County, California

PDF

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

Asset Metadata

Creator Recca, Sophia M. (author)

Core Title A spatial and temporal exploration of how satellite communication devices impact mountain search and rescue missions in California’s Sierra Nevada mountain range

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Degree Conferral Date 2023-08

Publication Date 05/24/2023

Defense Date 05/04/2023

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

Tag cluster and outlier analysis,emerging hot spot analysis,geographic information system,hot spot analysis,OAI-PMH Harvest,SAR,search and rescue,spatial analysis,spatiotemporal analysis

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Sedano, Elisabeth (committee chair), Ter-Ghazaryan, Diana (committee member), Wu, An-Min (committee member)

Creator Email recca@usc.edu,sophrecca@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113138313

Unique identifier UC113138313

Identifier etd-ReccaSophi-11890.pdf (filename)

Legacy Identifier etd-ReccaSophi-11890

Document Type Thesis

Format theses (aat)

Rights Recca, Sophia M.

Internet Media Type application/pdf

Type texts

Source 20230530-usctheses-batch-1049 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

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

Repository Email cisadmin@lib.usc.edu